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. 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 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $ 36 */ 37 38 #include "opt_ktrace.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/sysproto.h> 43 #include <sys/filedesc.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/malloc.h> 47 #include <sys/proc.h> 48 #include <sys/resourcevar.h> 49 #include <sys/vnode.h> 50 #include <sys/acct.h> 51 #include <sys/ktrace.h> 52 #include <sys/unistd.h> 53 #include <sys/jail.h> 54 55 #include <vm/vm.h> 56 #include <sys/lock.h> 57 #include <vm/pmap.h> 58 #include <vm/vm_map.h> 59 #include <vm/vm_extern.h> 60 61 #include <sys/vmmeter.h> 62 #include <sys/refcount.h> 63 #include <sys/thread2.h> 64 #include <sys/signal2.h> 65 #include <sys/spinlock2.h> 66 67 #include <sys/dsched.h> 68 69 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 70 71 /* 72 * These are the stuctures used to create a callout list for things to do 73 * when forking a process 74 */ 75 struct forklist { 76 forklist_fn function; 77 TAILQ_ENTRY(forklist) next; 78 }; 79 80 TAILQ_HEAD(forklist_head, forklist); 81 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 82 83 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags); 84 85 int forksleep; /* Place for fork1() to sleep on. */ 86 87 /* 88 * Red-Black tree support for LWPs 89 */ 90 91 static int 92 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2) 93 { 94 if (lp1->lwp_tid < lp2->lwp_tid) 95 return(-1); 96 if (lp1->lwp_tid > lp2->lwp_tid) 97 return(1); 98 return(0); 99 } 100 101 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid); 102 103 /* 104 * fork() system call 105 */ 106 int 107 sys_fork(struct fork_args *uap) 108 { 109 struct lwp *lp = curthread->td_lwp; 110 struct proc *p2; 111 int error; 112 113 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2); 114 if (error == 0) { 115 PHOLD(p2); 116 start_forked_proc(lp, p2); 117 uap->sysmsg_fds[0] = p2->p_pid; 118 uap->sysmsg_fds[1] = 0; 119 PRELE(p2); 120 } 121 return error; 122 } 123 124 /* 125 * vfork() system call 126 */ 127 int 128 sys_vfork(struct vfork_args *uap) 129 { 130 struct lwp *lp = curthread->td_lwp; 131 struct proc *p2; 132 int error; 133 134 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2); 135 if (error == 0) { 136 PHOLD(p2); 137 start_forked_proc(lp, p2); 138 uap->sysmsg_fds[0] = p2->p_pid; 139 uap->sysmsg_fds[1] = 0; 140 PRELE(p2); 141 } 142 return error; 143 } 144 145 /* 146 * Handle rforks. An rfork may (1) operate on the current process without 147 * creating a new, (2) create a new process that shared the current process's 148 * vmspace, signals, and/or descriptors, or (3) create a new process that does 149 * not share these things (normal fork). 150 * 151 * Note that we only call start_forked_proc() if a new process is actually 152 * created. 153 * 154 * rfork { int flags } 155 */ 156 int 157 sys_rfork(struct rfork_args *uap) 158 { 159 struct lwp *lp = curthread->td_lwp; 160 struct proc *p2; 161 int error; 162 163 if ((uap->flags & RFKERNELONLY) != 0) 164 return (EINVAL); 165 166 error = fork1(lp, uap->flags | RFPGLOCK, &p2); 167 if (error == 0) { 168 if (p2) { 169 PHOLD(p2); 170 start_forked_proc(lp, p2); 171 uap->sysmsg_fds[0] = p2->p_pid; 172 uap->sysmsg_fds[1] = 0; 173 PRELE(p2); 174 } else { 175 uap->sysmsg_fds[0] = 0; 176 uap->sysmsg_fds[1] = 0; 177 } 178 } 179 return error; 180 } 181 182 /* 183 * Low level thread create used by pthreads. 184 */ 185 int 186 sys_lwp_create(struct lwp_create_args *uap) 187 { 188 struct proc *p = curproc; 189 struct lwp *lp; 190 struct lwp_params params; 191 int error; 192 193 error = copyin(uap->params, ¶ms, sizeof(params)); 194 if (error) 195 goto fail2; 196 197 lwkt_gettoken(&p->p_token); 198 plimit_lwp_fork(p); /* force exclusive access */ 199 lp = lwp_fork(curthread->td_lwp, p, RFPROC); 200 error = cpu_prepare_lwp(lp, ¶ms); 201 if (error) 202 goto fail; 203 if (params.tid1 != NULL && 204 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid)))) 205 goto fail; 206 if (params.tid2 != NULL && 207 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid)))) 208 goto fail; 209 210 /* 211 * Now schedule the new lwp. 212 */ 213 p->p_usched->resetpriority(lp); 214 crit_enter(); 215 lp->lwp_stat = LSRUN; 216 p->p_usched->setrunqueue(lp); 217 crit_exit(); 218 lwkt_reltoken(&p->p_token); 219 220 return (0); 221 222 fail: 223 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 224 --p->p_nthreads; 225 /* lwp_dispose expects an exited lwp, and a held proc */ 226 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 227 lp->lwp_thread->td_flags |= TDF_EXITING; 228 lwkt_remove_tdallq(lp->lwp_thread); 229 PHOLD(p); 230 biosched_done(lp->lwp_thread); 231 dsched_exit_thread(lp->lwp_thread); 232 lwp_dispose(lp); 233 lwkt_reltoken(&p->p_token); 234 fail2: 235 return (error); 236 } 237 238 int nprocs = 1; /* process 0 */ 239 240 int 241 fork1(struct lwp *lp1, int flags, struct proc **procp) 242 { 243 struct proc *p1 = lp1->lwp_proc; 244 struct proc *p2; 245 struct proc *pptr; 246 struct pgrp *p1grp; 247 struct pgrp *plkgrp; 248 uid_t uid; 249 int ok, error; 250 static int curfail = 0; 251 static struct timeval lastfail; 252 struct forklist *ep; 253 struct filedesc_to_leader *fdtol; 254 255 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 256 return (EINVAL); 257 258 lwkt_gettoken(&p1->p_token); 259 plkgrp = NULL; 260 p2 = NULL; 261 262 /* 263 * Here we don't create a new process, but we divorce 264 * certain parts of a process from itself. 265 */ 266 if ((flags & RFPROC) == 0) { 267 /* 268 * This kind of stunt does not work anymore if 269 * there are native threads (lwps) running 270 */ 271 if (p1->p_nthreads != 1) { 272 error = EINVAL; 273 goto done; 274 } 275 276 vm_fork(p1, 0, flags); 277 278 /* 279 * Close all file descriptors. 280 */ 281 if (flags & RFCFDG) { 282 struct filedesc *fdtmp; 283 fdtmp = fdinit(p1); 284 fdfree(p1, fdtmp); 285 } 286 287 /* 288 * Unshare file descriptors (from parent.) 289 */ 290 if (flags & RFFDG) { 291 if (p1->p_fd->fd_refcnt > 1) { 292 struct filedesc *newfd; 293 error = fdcopy(p1, &newfd); 294 if (error != 0) { 295 error = ENOMEM; 296 goto done; 297 } 298 fdfree(p1, newfd); 299 } 300 } 301 *procp = NULL; 302 error = 0; 303 goto done; 304 } 305 306 /* 307 * Interlock against process group signal delivery. If signals 308 * are pending after the interlock is obtained we have to restart 309 * the system call to process the signals. If we don't the child 310 * can miss a pgsignal (such as ^C) sent during the fork. 311 * 312 * We can't use CURSIG() here because it will process any STOPs 313 * and cause the process group lock to be held indefinitely. If 314 * a STOP occurs, the fork will be restarted after the CONT. 315 */ 316 p1grp = p1->p_pgrp; 317 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) { 318 pgref(plkgrp); 319 lockmgr(&plkgrp->pg_lock, LK_SHARED); 320 if (CURSIG_NOBLOCK(lp1)) { 321 error = ERESTART; 322 goto done; 323 } 324 } 325 326 /* 327 * Although process entries are dynamically created, we still keep 328 * a global limit on the maximum number we will create. Don't allow 329 * a nonprivileged user to use the last ten processes; don't let root 330 * exceed the limit. The variable nprocs is the current number of 331 * processes, maxproc is the limit. 332 */ 333 uid = lp1->lwp_thread->td_ucred->cr_ruid; 334 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 335 if (ppsratecheck(&lastfail, &curfail, 1)) 336 kprintf("maxproc limit exceeded by uid %d, please " 337 "see tuning(7) and login.conf(5).\n", uid); 338 tsleep(&forksleep, 0, "fork", hz / 2); 339 error = EAGAIN; 340 goto done; 341 } 342 343 /* 344 * Increment the nprocs resource before blocking can occur. There 345 * are hard-limits as to the number of processes that can run. 346 */ 347 atomic_add_int(&nprocs, 1); 348 349 /* 350 * Increment the count of procs running with this uid. Don't allow 351 * a nonprivileged user to exceed their current limit. 352 */ 353 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1, 354 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 355 if (!ok) { 356 /* 357 * Back out the process count 358 */ 359 atomic_add_int(&nprocs, -1); 360 if (ppsratecheck(&lastfail, &curfail, 1)) 361 kprintf("maxproc limit exceeded by uid %d, please " 362 "see tuning(7) and login.conf(5).\n", uid); 363 tsleep(&forksleep, 0, "fork", hz / 2); 364 error = EAGAIN; 365 goto done; 366 } 367 368 /* 369 * Allocate a new process, don't get fancy: zero the structure. 370 */ 371 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO); 372 373 /* 374 * Core initialization. SIDL is a safety state that protects the 375 * partially initialized process once it starts getting hooked 376 * into system structures and becomes addressable. 377 * 378 * We must be sure to acquire p2->p_token as well, we must hold it 379 * once the process is on the allproc list to avoid things such 380 * as competing modifications to p_flags. 381 */ 382 p2->p_lasttid = -1; /* first tid will be 0 */ 383 p2->p_stat = SIDL; 384 385 RB_INIT(&p2->p_lwp_tree); 386 spin_init(&p2->p_spin); 387 lwkt_token_init(&p2->p_token, "proc"); 388 lwkt_gettoken(&p2->p_token); 389 390 /* 391 * Setup linkage for kernel based threading XXX lwp. Also add the 392 * process to the allproclist. 393 * 394 * The process structure is addressable after this point. 395 */ 396 if (flags & RFTHREAD) { 397 p2->p_peers = p1->p_peers; 398 p1->p_peers = p2; 399 p2->p_leader = p1->p_leader; 400 } else { 401 p2->p_leader = p2; 402 } 403 proc_add_allproc(p2); 404 405 /* 406 * Initialize the section which is copied verbatim from the parent. 407 */ 408 bcopy(&p1->p_startcopy, &p2->p_startcopy, 409 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 410 411 /* 412 * Duplicate sub-structures as needed. Increase reference counts 413 * on shared objects. 414 * 415 * NOTE: because we are now on the allproc list it is possible for 416 * other consumers to gain temporary references to p2 417 * (p2->p_lock can change). 418 */ 419 if (p1->p_flags & P_PROFIL) 420 startprofclock(p2); 421 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred); 422 423 if (jailed(p2->p_ucred)) 424 p2->p_flags |= P_JAILED; 425 426 if (p2->p_args) 427 refcount_acquire(&p2->p_args->ar_ref); 428 429 p2->p_usched = p1->p_usched; 430 /* XXX: verify copy of the secondary iosched stuff */ 431 dsched_new_proc(p2); 432 433 if (flags & RFSIGSHARE) { 434 p2->p_sigacts = p1->p_sigacts; 435 refcount_acquire(&p2->p_sigacts->ps_refcnt); 436 } else { 437 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts), 438 M_SUBPROC, M_WAITOK); 439 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts)); 440 refcount_init(&p2->p_sigacts->ps_refcnt, 1); 441 } 442 if (flags & RFLINUXTHPN) 443 p2->p_sigparent = SIGUSR1; 444 else 445 p2->p_sigparent = SIGCHLD; 446 447 /* bump references to the text vnode (for procfs) */ 448 p2->p_textvp = p1->p_textvp; 449 if (p2->p_textvp) 450 vref(p2->p_textvp); 451 452 /* copy namecache handle to the text file */ 453 if (p1->p_textnch.mount) 454 cache_copy(&p1->p_textnch, &p2->p_textnch); 455 456 /* 457 * Handle file descriptors 458 */ 459 if (flags & RFCFDG) { 460 p2->p_fd = fdinit(p1); 461 fdtol = NULL; 462 } else if (flags & RFFDG) { 463 error = fdcopy(p1, &p2->p_fd); 464 if (error != 0) { 465 error = ENOMEM; 466 goto done; 467 } 468 fdtol = NULL; 469 } else { 470 p2->p_fd = fdshare(p1); 471 if (p1->p_fdtol == NULL) { 472 p1->p_fdtol = filedesc_to_leader_alloc(NULL, 473 p1->p_leader); 474 } 475 if ((flags & RFTHREAD) != 0) { 476 /* 477 * Shared file descriptor table and 478 * shared process leaders. 479 */ 480 fdtol = p1->p_fdtol; 481 fdtol->fdl_refcount++; 482 } else { 483 /* 484 * Shared file descriptor table, and 485 * different process leaders 486 */ 487 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 488 } 489 } 490 p2->p_fdtol = fdtol; 491 p2->p_limit = plimit_fork(p1); 492 493 /* 494 * Preserve some more flags in subprocess. P_PROFIL has already 495 * been preserved. 496 */ 497 p2->p_flags |= p1->p_flags & P_SUGID; 498 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT)) 499 p2->p_flags |= P_CONTROLT; 500 if (flags & RFPPWAIT) 501 p2->p_flags |= P_PPWAIT; 502 503 /* 504 * Inherit the virtual kernel structure (allows a virtual kernel 505 * to fork to simulate multiple cpus). 506 */ 507 if (p1->p_vkernel) 508 vkernel_inherit(p1, p2); 509 510 /* 511 * Once we are on a pglist we may receive signals. XXX we might 512 * race a ^C being sent to the process group by not receiving it 513 * at all prior to this line. 514 */ 515 pgref(p1grp); 516 lwkt_gettoken(&p1grp->pg_token); 517 LIST_INSERT_AFTER(p1, p2, p_pglist); 518 lwkt_reltoken(&p1grp->pg_token); 519 520 /* 521 * Attach the new process to its parent. 522 * 523 * If RFNOWAIT is set, the newly created process becomes a child 524 * of init. This effectively disassociates the child from the 525 * parent. 526 */ 527 if (flags & RFNOWAIT) 528 pptr = initproc; 529 else 530 pptr = p1; 531 p2->p_pptr = pptr; 532 LIST_INIT(&p2->p_children); 533 534 lwkt_gettoken(&pptr->p_token); 535 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 536 lwkt_reltoken(&pptr->p_token); 537 538 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 539 callout_init_mp(&p2->p_ithandle); 540 541 #ifdef KTRACE 542 /* 543 * Copy traceflag and tracefile if enabled. If not inherited, 544 * these were zeroed above but we still could have a trace race 545 * so make sure p2's p_tracenode is NULL. 546 */ 547 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) { 548 p2->p_traceflag = p1->p_traceflag; 549 p2->p_tracenode = ktrinherit(p1->p_tracenode); 550 } 551 #endif 552 553 /* 554 * This begins the section where we must prevent the parent 555 * from being swapped. 556 * 557 * Gets PRELE'd in the caller in start_forked_proc(). 558 */ 559 PHOLD(p1); 560 561 vm_fork(p1, p2, flags); 562 563 /* 564 * Create the first lwp associated with the new proc. 565 * It will return via a different execution path later, directly 566 * into userland, after it was put on the runq by 567 * start_forked_proc(). 568 */ 569 lwp_fork(lp1, p2, flags); 570 571 if (flags == (RFFDG | RFPROC | RFPGLOCK)) { 572 mycpu->gd_cnt.v_forks++; 573 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + 574 p2->p_vmspace->vm_ssize; 575 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) { 576 mycpu->gd_cnt.v_vforks++; 577 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 578 p2->p_vmspace->vm_ssize; 579 } else if (p1 == &proc0) { 580 mycpu->gd_cnt.v_kthreads++; 581 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 582 p2->p_vmspace->vm_ssize; 583 } else { 584 mycpu->gd_cnt.v_rforks++; 585 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 586 p2->p_vmspace->vm_ssize; 587 } 588 589 /* 590 * Both processes are set up, now check if any loadable modules want 591 * to adjust anything. 592 * What if they have an error? XXX 593 */ 594 TAILQ_FOREACH(ep, &fork_list, next) { 595 (*ep->function)(p1, p2, flags); 596 } 597 598 /* 599 * Set the start time. Note that the process is not runnable. The 600 * caller is responsible for making it runnable. 601 */ 602 microtime(&p2->p_start); 603 p2->p_acflag = AFORK; 604 605 /* 606 * tell any interested parties about the new process 607 */ 608 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 609 610 /* 611 * Return child proc pointer to parent. 612 */ 613 *procp = p2; 614 error = 0; 615 done: 616 if (p2) 617 lwkt_reltoken(&p2->p_token); 618 lwkt_reltoken(&p1->p_token); 619 if (plkgrp) { 620 lockmgr(&plkgrp->pg_lock, LK_RELEASE); 621 pgrel(plkgrp); 622 } 623 return (error); 624 } 625 626 static struct lwp * 627 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags) 628 { 629 globaldata_t gd = mycpu; 630 struct lwp *lp; 631 struct thread *td; 632 633 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO); 634 635 lp->lwp_proc = destproc; 636 lp->lwp_vmspace = destproc->p_vmspace; 637 lp->lwp_stat = LSRUN; 638 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy, 639 (unsigned) ((caddr_t)&lp->lwp_endcopy - 640 (caddr_t)&lp->lwp_startcopy)); 641 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK; 642 /* 643 * Set cpbase to the last timeout that occured (not the upcoming 644 * timeout). 645 * 646 * A critical section is required since a timer IPI can update 647 * scheduler specific data. 648 */ 649 crit_enter(); 650 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic; 651 destproc->p_usched->heuristic_forking(origlp, lp); 652 crit_exit(); 653 lp->lwp_cpumask &= usched_mastermask; 654 lwkt_token_init(&lp->lwp_token, "lwp_token"); 655 spin_init(&lp->lwp_spin); 656 657 /* 658 * Assign the thread to the current cpu to begin with so we 659 * can manipulate it. 660 */ 661 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0); 662 lp->lwp_thread = td; 663 td->td_ucred = crhold(destproc->p_ucred); 664 td->td_proc = destproc; 665 td->td_lwp = lp; 666 td->td_switch = cpu_heavy_switch; 667 #ifdef NO_LWKT_SPLIT_USERPRI 668 lwkt_setpri(td, TDPRI_USER_NORM); 669 #else 670 lwkt_setpri(td, TDPRI_KERN_USER); 671 #endif 672 lwkt_set_comm(td, "%s", destproc->p_comm); 673 674 /* 675 * cpu_fork will copy and update the pcb, set up the kernel stack, 676 * and make the child ready to run. 677 */ 678 cpu_fork(origlp, lp, flags); 679 kqueue_init(&lp->lwp_kqueue, destproc->p_fd); 680 681 /* 682 * Assign a TID to the lp. Loop until the insert succeeds (returns 683 * NULL). 684 */ 685 lp->lwp_tid = destproc->p_lasttid; 686 do { 687 if (++lp->lwp_tid < 0) 688 lp->lwp_tid = 1; 689 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL); 690 destproc->p_lasttid = lp->lwp_tid; 691 destproc->p_nthreads++; 692 693 /* 694 * This flag is set and never cleared. It means that the process 695 * was threaded at some point. Used to improve exit performance. 696 */ 697 destproc->p_flags |= P_MAYBETHREADED; 698 699 return (lp); 700 } 701 702 /* 703 * The next two functionms are general routines to handle adding/deleting 704 * items on the fork callout list. 705 * 706 * at_fork(): 707 * Take the arguments given and put them onto the fork callout list, 708 * However first make sure that it's not already there. 709 * Returns 0 on success or a standard error number. 710 */ 711 int 712 at_fork(forklist_fn function) 713 { 714 struct forklist *ep; 715 716 #ifdef INVARIANTS 717 /* let the programmer know if he's been stupid */ 718 if (rm_at_fork(function)) { 719 kprintf("WARNING: fork callout entry (%p) already present\n", 720 function); 721 } 722 #endif 723 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 724 ep->function = function; 725 TAILQ_INSERT_TAIL(&fork_list, ep, next); 726 return (0); 727 } 728 729 /* 730 * Scan the exit callout list for the given item and remove it.. 731 * Returns the number of items removed (0 or 1) 732 */ 733 int 734 rm_at_fork(forklist_fn function) 735 { 736 struct forklist *ep; 737 738 TAILQ_FOREACH(ep, &fork_list, next) { 739 if (ep->function == function) { 740 TAILQ_REMOVE(&fork_list, ep, next); 741 kfree(ep, M_ATFORK); 742 return(1); 743 } 744 } 745 return (0); 746 } 747 748 /* 749 * Add a forked process to the run queue after any remaining setup, such 750 * as setting the fork handler, has been completed. 751 * 752 * p2 is held by the caller. 753 */ 754 void 755 start_forked_proc(struct lwp *lp1, struct proc *p2) 756 { 757 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2); 758 int pflags; 759 760 /* 761 * Move from SIDL to RUN queue, and activate the process's thread. 762 * Activation of the thread effectively makes the process "a" 763 * current process, so we do not setrunqueue(). 764 * 765 * YYY setrunqueue works here but we should clean up the trampoline 766 * code so we just schedule the LWKT thread and let the trampoline 767 * deal with the userland scheduler on return to userland. 768 */ 769 KASSERT(p2->p_stat == SIDL, 770 ("cannot start forked process, bad status: %p", p2)); 771 p2->p_usched->resetpriority(lp2); 772 crit_enter(); 773 p2->p_stat = SACTIVE; 774 lp2->lwp_stat = LSRUN; 775 p2->p_usched->setrunqueue(lp2); 776 crit_exit(); 777 778 /* 779 * Now can be swapped. 780 */ 781 PRELE(lp1->lwp_proc); 782 783 /* 784 * Preserve synchronization semantics of vfork. P_PPWAIT is set in 785 * the child until it has retired the parent's resources. The parent 786 * must wait for the flag to be cleared by the child. 787 * 788 * Interlock the flag/tsleep with atomic ops to avoid unnecessary 789 * p_token conflicts. 790 * 791 * XXX Is this use of an atomic op on a field that is not normally 792 * manipulated with atomic ops ok? 793 */ 794 while ((pflags = p2->p_flags) & P_PPWAIT) { 795 cpu_ccfence(); 796 tsleep_interlock(lp1->lwp_proc, 0); 797 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags)) 798 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0); 799 } 800 } 801