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 #include <sys/lwp.h> 55 56 #include <vm/vm.h> 57 #include <sys/lock.h> 58 #include <vm/pmap.h> 59 #include <vm/vm_map.h> 60 #include <vm/vm_extern.h> 61 62 #include <sys/vmmeter.h> 63 #include <sys/refcount.h> 64 #include <sys/thread2.h> 65 #include <sys/signal2.h> 66 #include <sys/spinlock2.h> 67 68 #include <sys/dsched.h> 69 70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 71 static MALLOC_DEFINE(M_REAPER, "reaper", "process reapers"); 72 73 /* 74 * These are the stuctures used to create a callout list for things to do 75 * when forking a process 76 */ 77 struct forklist { 78 forklist_fn function; 79 TAILQ_ENTRY(forklist) next; 80 }; 81 82 TAILQ_HEAD(forklist_head, forklist); 83 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 84 85 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags, 86 const cpumask_t *mask); 87 static int lwp_create1(struct lwp_params *params, 88 const cpumask_t *mask); 89 90 int forksleep; /* Place for fork1() to sleep on. */ 91 92 /* 93 * Red-Black tree support for LWPs 94 */ 95 96 static int 97 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2) 98 { 99 if (lp1->lwp_tid < lp2->lwp_tid) 100 return(-1); 101 if (lp1->lwp_tid > lp2->lwp_tid) 102 return(1); 103 return(0); 104 } 105 106 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid); 107 108 /* 109 * fork() system call 110 */ 111 int 112 sys_fork(struct fork_args *uap) 113 { 114 struct lwp *lp = curthread->td_lwp; 115 struct proc *p2; 116 int error; 117 118 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2); 119 if (error == 0) { 120 PHOLD(p2); 121 start_forked_proc(lp, p2); 122 uap->sysmsg_fds[0] = p2->p_pid; 123 uap->sysmsg_fds[1] = 0; 124 PRELE(p2); 125 } 126 return error; 127 } 128 129 /* 130 * vfork() system call 131 */ 132 int 133 sys_vfork(struct vfork_args *uap) 134 { 135 struct lwp *lp = curthread->td_lwp; 136 struct proc *p2; 137 int error; 138 139 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2); 140 if (error == 0) { 141 PHOLD(p2); 142 start_forked_proc(lp, p2); 143 uap->sysmsg_fds[0] = p2->p_pid; 144 uap->sysmsg_fds[1] = 0; 145 PRELE(p2); 146 } 147 return error; 148 } 149 150 /* 151 * Handle rforks. An rfork may (1) operate on the current process without 152 * creating a new, (2) create a new process that shared the current process's 153 * vmspace, signals, and/or descriptors, or (3) create a new process that does 154 * not share these things (normal fork). 155 * 156 * Note that we only call start_forked_proc() if a new process is actually 157 * created. 158 * 159 * rfork { int flags } 160 */ 161 int 162 sys_rfork(struct rfork_args *uap) 163 { 164 struct lwp *lp = curthread->td_lwp; 165 struct proc *p2; 166 int error; 167 168 if ((uap->flags & RFKERNELONLY) != 0) 169 return (EINVAL); 170 171 error = fork1(lp, uap->flags | RFPGLOCK, &p2); 172 if (error == 0) { 173 if (p2) { 174 PHOLD(p2); 175 start_forked_proc(lp, p2); 176 uap->sysmsg_fds[0] = p2->p_pid; 177 uap->sysmsg_fds[1] = 0; 178 PRELE(p2); 179 } else { 180 uap->sysmsg_fds[0] = 0; 181 uap->sysmsg_fds[1] = 0; 182 } 183 } 184 return error; 185 } 186 187 static int 188 lwp_create1(struct lwp_params *uprm, const cpumask_t *umask) 189 { 190 struct proc *p = curproc; 191 struct lwp *lp; 192 struct lwp_params params; 193 cpumask_t *mask = NULL, mask0; 194 int error; 195 196 error = copyin(uprm, ¶ms, sizeof(params)); 197 if (error) 198 goto fail2; 199 200 if (umask != NULL) { 201 error = copyin(umask, &mask0, sizeof(mask0)); 202 if (error) 203 goto fail2; 204 CPUMASK_ANDMASK(mask0, smp_active_mask); 205 if (CPUMASK_TESTNZERO(mask0)) 206 mask = &mask0; 207 } 208 209 lwkt_gettoken(&p->p_token); 210 plimit_lwp_fork(p); /* force exclusive access */ 211 lp = lwp_fork(curthread->td_lwp, p, RFPROC | RFMEM, mask); 212 error = cpu_prepare_lwp(lp, ¶ms); 213 if (error) 214 goto fail; 215 if (params.lwp_tid1 != NULL && 216 (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid)))) 217 goto fail; 218 if (params.lwp_tid2 != NULL && 219 (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid)))) 220 goto fail; 221 222 /* 223 * Now schedule the new lwp. 224 */ 225 p->p_usched->resetpriority(lp); 226 crit_enter(); 227 lp->lwp_stat = LSRUN; 228 p->p_usched->setrunqueue(lp); 229 crit_exit(); 230 lwkt_reltoken(&p->p_token); 231 232 return (0); 233 234 fail: 235 /* 236 * Make sure no one is using this lwp, before it is removed from 237 * the tree. If we didn't wait it here, lwp tree iteration with 238 * blocking operation would be broken. 239 */ 240 while (lp->lwp_lock > 0) 241 tsleep(lp, 0, "lwpfail", 1); 242 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 243 --p->p_nthreads; 244 /* lwp_dispose expects an exited lwp, and a held proc */ 245 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 246 lp->lwp_thread->td_flags |= TDF_EXITING; 247 lwkt_remove_tdallq(lp->lwp_thread); 248 PHOLD(p); 249 biosched_done(lp->lwp_thread); 250 dsched_exit_thread(lp->lwp_thread); 251 lwp_dispose(lp); 252 lwkt_reltoken(&p->p_token); 253 fail2: 254 return (error); 255 } 256 257 /* 258 * Low level thread create used by pthreads. 259 */ 260 int 261 sys_lwp_create(struct lwp_create_args *uap) 262 { 263 264 return (lwp_create1(uap->params, NULL)); 265 } 266 267 int 268 sys_lwp_create2(struct lwp_create2_args *uap) 269 { 270 271 return (lwp_create1(uap->params, uap->mask)); 272 } 273 274 int nprocs = 1; /* process 0 */ 275 276 int 277 fork1(struct lwp *lp1, int flags, struct proc **procp) 278 { 279 struct proc *p1 = lp1->lwp_proc; 280 struct proc *p2; 281 struct proc *pptr; 282 struct pgrp *p1grp; 283 struct pgrp *plkgrp; 284 struct sysreaper *reap; 285 uid_t uid; 286 int ok, error; 287 static int curfail = 0; 288 static struct timeval lastfail; 289 struct forklist *ep; 290 struct filedesc_to_leader *fdtol; 291 292 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 293 return (EINVAL); 294 295 lwkt_gettoken(&p1->p_token); 296 plkgrp = NULL; 297 p2 = NULL; 298 299 /* 300 * Here we don't create a new process, but we divorce 301 * certain parts of a process from itself. 302 */ 303 if ((flags & RFPROC) == 0) { 304 /* 305 * This kind of stunt does not work anymore if 306 * there are native threads (lwps) running 307 */ 308 if (p1->p_nthreads != 1) { 309 error = EINVAL; 310 goto done; 311 } 312 313 vm_fork(p1, 0, flags); 314 315 /* 316 * Close all file descriptors. 317 */ 318 if (flags & RFCFDG) { 319 struct filedesc *fdtmp; 320 fdtmp = fdinit(p1); 321 fdfree(p1, fdtmp); 322 } 323 324 /* 325 * Unshare file descriptors (from parent.) 326 */ 327 if (flags & RFFDG) { 328 if (p1->p_fd->fd_refcnt > 1) { 329 struct filedesc *newfd; 330 error = fdcopy(p1, &newfd); 331 if (error != 0) { 332 error = ENOMEM; 333 goto done; 334 } 335 fdfree(p1, newfd); 336 } 337 } 338 *procp = NULL; 339 error = 0; 340 goto done; 341 } 342 343 /* 344 * Interlock against process group signal delivery. If signals 345 * are pending after the interlock is obtained we have to restart 346 * the system call to process the signals. If we don't the child 347 * can miss a pgsignal (such as ^C) sent during the fork. 348 * 349 * We can't use CURSIG() here because it will process any STOPs 350 * and cause the process group lock to be held indefinitely. If 351 * a STOP occurs, the fork will be restarted after the CONT. 352 */ 353 p1grp = p1->p_pgrp; 354 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) { 355 pgref(plkgrp); 356 lockmgr(&plkgrp->pg_lock, LK_SHARED); 357 if (CURSIG_NOBLOCK(lp1)) { 358 error = ERESTART; 359 goto done; 360 } 361 } 362 363 /* 364 * Although process entries are dynamically created, we still keep 365 * a global limit on the maximum number we will create. Don't allow 366 * a nonprivileged user to use the last ten processes; don't let root 367 * exceed the limit. The variable nprocs is the current number of 368 * processes, maxproc is the limit. 369 */ 370 uid = lp1->lwp_thread->td_ucred->cr_ruid; 371 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 372 if (ppsratecheck(&lastfail, &curfail, 1)) 373 kprintf("maxproc limit exceeded by uid %d, please " 374 "see tuning(7) and login.conf(5).\n", uid); 375 tsleep(&forksleep, 0, "fork", hz / 2); 376 error = EAGAIN; 377 goto done; 378 } 379 380 /* 381 * Increment the nprocs resource before blocking can occur. There 382 * are hard-limits as to the number of processes that can run. 383 */ 384 atomic_add_int(&nprocs, 1); 385 386 /* 387 * Increment the count of procs running with this uid. Don't allow 388 * a nonprivileged user to exceed their current limit. 389 */ 390 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1, 391 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 392 if (!ok) { 393 /* 394 * Back out the process count 395 */ 396 atomic_add_int(&nprocs, -1); 397 if (ppsratecheck(&lastfail, &curfail, 1)) 398 kprintf("maxproc limit exceeded by uid %d, please " 399 "see tuning(7) and login.conf(5).\n", uid); 400 tsleep(&forksleep, 0, "fork", hz / 2); 401 error = EAGAIN; 402 goto done; 403 } 404 405 /* 406 * Allocate a new process, don't get fancy: zero the structure. 407 */ 408 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO); 409 410 /* 411 * Core initialization. SIDL is a safety state that protects the 412 * partially initialized process once it starts getting hooked 413 * into system structures and becomes addressable. 414 * 415 * We must be sure to acquire p2->p_token as well, we must hold it 416 * once the process is on the allproc list to avoid things such 417 * as competing modifications to p_flags. 418 */ 419 mycpu->gd_forkid += ncpus; 420 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid; 421 p2->p_lasttid = -1; /* first tid will be 0 */ 422 p2->p_stat = SIDL; 423 424 /* 425 * NOTE: Process 0 will not have a reaper, but process 1 (init) and 426 * all other processes always will. 427 */ 428 if ((reap = p1->p_reaper) != NULL) { 429 reaper_hold(reap); 430 p2->p_reaper = reap; 431 } else { 432 p2->p_reaper = NULL; 433 } 434 435 RB_INIT(&p2->p_lwp_tree); 436 spin_init(&p2->p_spin, "procfork1"); 437 lwkt_token_init(&p2->p_token, "proc"); 438 lwkt_gettoken(&p2->p_token); 439 440 /* 441 * Setup linkage for kernel based threading XXX lwp. Also add the 442 * process to the allproclist. 443 * 444 * The process structure is addressable after this point. 445 */ 446 if (flags & RFTHREAD) { 447 p2->p_peers = p1->p_peers; 448 p1->p_peers = p2; 449 p2->p_leader = p1->p_leader; 450 } else { 451 p2->p_leader = p2; 452 } 453 proc_add_allproc(p2); 454 455 /* 456 * Initialize the section which is copied verbatim from the parent. 457 */ 458 bcopy(&p1->p_startcopy, &p2->p_startcopy, 459 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 460 461 /* 462 * Duplicate sub-structures as needed. Increase reference counts 463 * on shared objects. 464 * 465 * NOTE: because we are now on the allproc list it is possible for 466 * other consumers to gain temporary references to p2 467 * (p2->p_lock can change). 468 */ 469 if (p1->p_flags & P_PROFIL) 470 startprofclock(p2); 471 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred); 472 473 if (jailed(p2->p_ucred)) 474 p2->p_flags |= P_JAILED; 475 476 if (p2->p_args) 477 refcount_acquire(&p2->p_args->ar_ref); 478 479 p2->p_usched = p1->p_usched; 480 /* XXX: verify copy of the secondary iosched stuff */ 481 dsched_enter_proc(p2); 482 483 if (flags & RFSIGSHARE) { 484 p2->p_sigacts = p1->p_sigacts; 485 refcount_acquire(&p2->p_sigacts->ps_refcnt); 486 } else { 487 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts), 488 M_SUBPROC, M_WAITOK); 489 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts)); 490 refcount_init(&p2->p_sigacts->ps_refcnt, 1); 491 } 492 if (flags & RFLINUXTHPN) 493 p2->p_sigparent = SIGUSR1; 494 else 495 p2->p_sigparent = SIGCHLD; 496 497 /* bump references to the text vnode (for procfs) */ 498 p2->p_textvp = p1->p_textvp; 499 if (p2->p_textvp) 500 vref(p2->p_textvp); 501 502 /* copy namecache handle to the text file */ 503 if (p1->p_textnch.mount) 504 cache_copy(&p1->p_textnch, &p2->p_textnch); 505 506 /* 507 * Handle file descriptors 508 */ 509 if (flags & RFCFDG) { 510 p2->p_fd = fdinit(p1); 511 fdtol = NULL; 512 } else if (flags & RFFDG) { 513 error = fdcopy(p1, &p2->p_fd); 514 if (error != 0) { 515 error = ENOMEM; 516 goto done; 517 } 518 fdtol = NULL; 519 } else { 520 p2->p_fd = fdshare(p1); 521 if (p1->p_fdtol == NULL) { 522 p1->p_fdtol = filedesc_to_leader_alloc(NULL, 523 p1->p_leader); 524 } 525 if ((flags & RFTHREAD) != 0) { 526 /* 527 * Shared file descriptor table and 528 * shared process leaders. 529 */ 530 fdtol = p1->p_fdtol; 531 fdtol->fdl_refcount++; 532 } else { 533 /* 534 * Shared file descriptor table, and 535 * different process leaders 536 */ 537 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 538 } 539 } 540 p2->p_fdtol = fdtol; 541 p2->p_limit = plimit_fork(p1); 542 543 /* 544 * Preserve some more flags in subprocess. P_PROFIL has already 545 * been preserved. 546 */ 547 p2->p_flags |= p1->p_flags & P_SUGID; 548 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT)) 549 p2->p_flags |= P_CONTROLT; 550 if (flags & RFPPWAIT) { 551 p2->p_flags |= P_PPWAIT; 552 if (p1->p_upmap) 553 atomic_add_int(&p1->p_upmap->invfork, 1); 554 } 555 556 /* 557 * Inherit the virtual kernel structure (allows a virtual kernel 558 * to fork to simulate multiple cpus). 559 */ 560 if (p1->p_vkernel) 561 vkernel_inherit(p1, p2); 562 563 /* 564 * Once we are on a pglist we may receive signals. XXX we might 565 * race a ^C being sent to the process group by not receiving it 566 * at all prior to this line. 567 */ 568 pgref(p1grp); 569 lwkt_gettoken(&p1grp->pg_token); 570 LIST_INSERT_AFTER(p1, p2, p_pglist); 571 lwkt_reltoken(&p1grp->pg_token); 572 573 /* 574 * Attach the new process to its parent. 575 * 576 * If RFNOWAIT is set, the newly created process becomes a child 577 * of the reaper (typically init). This effectively disassociates 578 * the child from the parent. 579 * 580 * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts. 581 */ 582 if (flags & RFNOWAIT) { 583 pptr = reaper_get(reap); 584 if (pptr == NULL) { 585 pptr = initproc; 586 PHOLD(pptr); 587 } 588 } else { 589 pptr = p1; 590 } 591 p2->p_pptr = pptr; 592 LIST_INIT(&p2->p_children); 593 594 lwkt_gettoken(&pptr->p_token); 595 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 596 lwkt_reltoken(&pptr->p_token); 597 598 if (flags & RFNOWAIT) 599 PRELE(pptr); 600 601 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 602 callout_init_mp(&p2->p_ithandle); 603 604 #ifdef KTRACE 605 /* 606 * Copy traceflag and tracefile if enabled. If not inherited, 607 * these were zeroed above but we still could have a trace race 608 * so make sure p2's p_tracenode is NULL. 609 */ 610 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) { 611 p2->p_traceflag = p1->p_traceflag; 612 p2->p_tracenode = ktrinherit(p1->p_tracenode); 613 } 614 #endif 615 616 /* 617 * This begins the section where we must prevent the parent 618 * from being swapped. 619 * 620 * Gets PRELE'd in the caller in start_forked_proc(). 621 */ 622 PHOLD(p1); 623 624 vm_fork(p1, p2, flags); 625 626 /* 627 * Create the first lwp associated with the new proc. 628 * It will return via a different execution path later, directly 629 * into userland, after it was put on the runq by 630 * start_forked_proc(). 631 */ 632 lwp_fork(lp1, p2, flags, NULL); 633 634 if (flags == (RFFDG | RFPROC | RFPGLOCK)) { 635 mycpu->gd_cnt.v_forks++; 636 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + 637 p2->p_vmspace->vm_ssize; 638 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) { 639 mycpu->gd_cnt.v_vforks++; 640 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 641 p2->p_vmspace->vm_ssize; 642 } else if (p1 == &proc0) { 643 mycpu->gd_cnt.v_kthreads++; 644 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 645 p2->p_vmspace->vm_ssize; 646 } else { 647 mycpu->gd_cnt.v_rforks++; 648 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 649 p2->p_vmspace->vm_ssize; 650 } 651 652 /* 653 * Both processes are set up, now check if any loadable modules want 654 * to adjust anything. 655 * What if they have an error? XXX 656 */ 657 TAILQ_FOREACH(ep, &fork_list, next) { 658 (*ep->function)(p1, p2, flags); 659 } 660 661 /* 662 * Set the start time. Note that the process is not runnable. The 663 * caller is responsible for making it runnable. 664 */ 665 microtime(&p2->p_start); 666 p2->p_acflag = AFORK; 667 668 /* 669 * tell any interested parties about the new process 670 */ 671 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 672 673 /* 674 * Return child proc pointer to parent. 675 */ 676 *procp = p2; 677 error = 0; 678 done: 679 if (p2) 680 lwkt_reltoken(&p2->p_token); 681 lwkt_reltoken(&p1->p_token); 682 if (plkgrp) { 683 lockmgr(&plkgrp->pg_lock, LK_RELEASE); 684 pgrel(plkgrp); 685 } 686 return (error); 687 } 688 689 static struct lwp * 690 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags, 691 const cpumask_t *mask) 692 { 693 globaldata_t gd = mycpu; 694 struct lwp *lp; 695 struct thread *td; 696 697 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO); 698 699 lp->lwp_proc = destproc; 700 lp->lwp_vmspace = destproc->p_vmspace; 701 lp->lwp_stat = LSRUN; 702 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy, 703 (unsigned) ((caddr_t)&lp->lwp_endcopy - 704 (caddr_t)&lp->lwp_startcopy)); 705 if (mask != NULL) 706 lp->lwp_cpumask = *mask; 707 708 /* 709 * Reset the sigaltstack if memory is shared, otherwise inherit 710 * it. 711 */ 712 if (flags & RFMEM) { 713 lp->lwp_sigstk.ss_flags = SS_DISABLE; 714 lp->lwp_sigstk.ss_size = 0; 715 lp->lwp_sigstk.ss_sp = NULL; 716 lp->lwp_flags &= ~LWP_ALTSTACK; 717 } else { 718 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK; 719 } 720 721 /* 722 * Set cpbase to the last timeout that occured (not the upcoming 723 * timeout). 724 * 725 * A critical section is required since a timer IPI can update 726 * scheduler specific data. 727 */ 728 crit_enter(); 729 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic; 730 destproc->p_usched->heuristic_forking(origlp, lp); 731 crit_exit(); 732 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask); 733 lwkt_token_init(&lp->lwp_token, "lwp_token"); 734 spin_init(&lp->lwp_spin, "lwptoken"); 735 736 /* 737 * Assign the thread to the current cpu to begin with so we 738 * can manipulate it. 739 */ 740 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0); 741 lp->lwp_thread = td; 742 td->td_ucred = crhold(destproc->p_ucred); 743 td->td_proc = destproc; 744 td->td_lwp = lp; 745 td->td_switch = cpu_heavy_switch; 746 #ifdef NO_LWKT_SPLIT_USERPRI 747 lwkt_setpri(td, TDPRI_USER_NORM); 748 #else 749 lwkt_setpri(td, TDPRI_KERN_USER); 750 #endif 751 lwkt_set_comm(td, "%s", destproc->p_comm); 752 753 /* 754 * cpu_fork will copy and update the pcb, set up the kernel stack, 755 * and make the child ready to run. 756 */ 757 cpu_fork(origlp, lp, flags); 758 kqueue_init(&lp->lwp_kqueue, destproc->p_fd); 759 760 /* 761 * Assign a TID to the lp. Loop until the insert succeeds (returns 762 * NULL). 763 * 764 * If we are in a vfork assign the same TID as the lwp that did the 765 * vfork(). This way if the user program messes around with 766 * pthread calls inside the vfork(), it will operate like an 767 * extension of the (blocked) parent. Also note that since the 768 * address space is being shared, insofar as pthreads is concerned, 769 * the code running in the vfork() is part of the original process. 770 */ 771 if (flags & RFPPWAIT) { 772 lp->lwp_tid = origlp->lwp_tid - 1; 773 } else { 774 lp->lwp_tid = destproc->p_lasttid; 775 } 776 777 do { 778 if (++lp->lwp_tid < 0) 779 lp->lwp_tid = 1; 780 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL); 781 782 destproc->p_lasttid = lp->lwp_tid; 783 destproc->p_nthreads++; 784 785 /* 786 * This flag is set and never cleared. It means that the process 787 * was threaded at some point. Used to improve exit performance. 788 */ 789 destproc->p_flags |= P_MAYBETHREADED; 790 791 return (lp); 792 } 793 794 /* 795 * The next two functionms are general routines to handle adding/deleting 796 * items on the fork callout list. 797 * 798 * at_fork(): 799 * Take the arguments given and put them onto the fork callout list, 800 * However first make sure that it's not already there. 801 * Returns 0 on success or a standard error number. 802 */ 803 int 804 at_fork(forklist_fn function) 805 { 806 struct forklist *ep; 807 808 #ifdef INVARIANTS 809 /* let the programmer know if he's been stupid */ 810 if (rm_at_fork(function)) { 811 kprintf("WARNING: fork callout entry (%p) already present\n", 812 function); 813 } 814 #endif 815 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 816 ep->function = function; 817 TAILQ_INSERT_TAIL(&fork_list, ep, next); 818 return (0); 819 } 820 821 /* 822 * Scan the exit callout list for the given item and remove it.. 823 * Returns the number of items removed (0 or 1) 824 */ 825 int 826 rm_at_fork(forklist_fn function) 827 { 828 struct forklist *ep; 829 830 TAILQ_FOREACH(ep, &fork_list, next) { 831 if (ep->function == function) { 832 TAILQ_REMOVE(&fork_list, ep, next); 833 kfree(ep, M_ATFORK); 834 return(1); 835 } 836 } 837 return (0); 838 } 839 840 /* 841 * Add a forked process to the run queue after any remaining setup, such 842 * as setting the fork handler, has been completed. 843 * 844 * p2 is held by the caller. 845 */ 846 void 847 start_forked_proc(struct lwp *lp1, struct proc *p2) 848 { 849 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2); 850 int pflags; 851 852 /* 853 * Move from SIDL to RUN queue, and activate the process's thread. 854 * Activation of the thread effectively makes the process "a" 855 * current process, so we do not setrunqueue(). 856 * 857 * YYY setrunqueue works here but we should clean up the trampoline 858 * code so we just schedule the LWKT thread and let the trampoline 859 * deal with the userland scheduler on return to userland. 860 */ 861 KASSERT(p2->p_stat == SIDL, 862 ("cannot start forked process, bad status: %p", p2)); 863 p2->p_usched->resetpriority(lp2); 864 crit_enter(); 865 p2->p_stat = SACTIVE; 866 lp2->lwp_stat = LSRUN; 867 p2->p_usched->setrunqueue(lp2); 868 crit_exit(); 869 870 /* 871 * Now can be swapped. 872 */ 873 PRELE(lp1->lwp_proc); 874 875 /* 876 * Preserve synchronization semantics of vfork. P_PPWAIT is set in 877 * the child until it has retired the parent's resources. The parent 878 * must wait for the flag to be cleared by the child. 879 * 880 * Interlock the flag/tsleep with atomic ops to avoid unnecessary 881 * p_token conflicts. 882 * 883 * XXX Is this use of an atomic op on a field that is not normally 884 * manipulated with atomic ops ok? 885 */ 886 while ((pflags = p2->p_flags) & P_PPWAIT) { 887 cpu_ccfence(); 888 tsleep_interlock(lp1->lwp_proc, 0); 889 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags)) 890 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0); 891 } 892 } 893 894 /* 895 * procctl (idtype_t idtype, id_t id, int cmd, void *arg) 896 */ 897 int 898 sys_procctl(struct procctl_args *uap) 899 { 900 struct proc *p = curproc; 901 struct proc *p2; 902 struct sysreaper *reap; 903 union reaper_info udata; 904 int error; 905 906 if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid) 907 return EINVAL; 908 909 switch(uap->cmd) { 910 case PROC_REAP_ACQUIRE: 911 lwkt_gettoken(&p->p_token); 912 reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO); 913 if (p->p_reaper == NULL || p->p_reaper->p != p) { 914 reaper_init(p, reap); 915 error = 0; 916 } else { 917 kfree(reap, M_REAPER); 918 error = EALREADY; 919 } 920 lwkt_reltoken(&p->p_token); 921 break; 922 case PROC_REAP_RELEASE: 923 lwkt_gettoken(&p->p_token); 924 release_again: 925 reap = p->p_reaper; 926 KKASSERT(reap != NULL); 927 if (reap->p == p) { 928 reaper_hold(reap); /* in case of thread race */ 929 lockmgr(&reap->lock, LK_EXCLUSIVE); 930 if (reap->p != p) { 931 lockmgr(&reap->lock, LK_RELEASE); 932 reaper_drop(reap); 933 goto release_again; 934 } 935 reap->p = NULL; 936 p->p_reaper = reap->parent; 937 if (p->p_reaper) 938 reaper_hold(p->p_reaper); 939 lockmgr(&reap->lock, LK_RELEASE); 940 reaper_drop(reap); /* our ref */ 941 reaper_drop(reap); /* old p_reaper ref */ 942 error = 0; 943 } else { 944 error = ENOTCONN; 945 } 946 lwkt_reltoken(&p->p_token); 947 break; 948 case PROC_REAP_STATUS: 949 bzero(&udata, sizeof(udata)); 950 lwkt_gettoken_shared(&p->p_token); 951 if ((reap = p->p_reaper) != NULL && reap->p == p) { 952 udata.status.flags = reap->flags; 953 udata.status.refs = reap->refs - 1; /* minus ours */ 954 } 955 p2 = LIST_FIRST(&p->p_children); 956 udata.status.pid_head = p2 ? p2->p_pid : -1; 957 lwkt_reltoken(&p->p_token); 958 959 if (uap->data) { 960 error = copyout(&udata, uap->data, 961 sizeof(udata.status)); 962 } else { 963 error = 0; 964 } 965 break; 966 default: 967 error = EINVAL; 968 break; 969 } 970 return error; 971 } 972 973 /* 974 * Bump ref on reaper, preventing destruction 975 */ 976 void 977 reaper_hold(struct sysreaper *reap) 978 { 979 KKASSERT(reap->refs > 0); 980 refcount_acquire(&reap->refs); 981 } 982 983 /* 984 * Drop ref on reaper, destroy the structure on the 1->0 985 * transition and loop on the parent. 986 */ 987 void 988 reaper_drop(struct sysreaper *next) 989 { 990 struct sysreaper *reap; 991 992 while ((reap = next) != NULL) { 993 if (refcount_release(&reap->refs)) { 994 next = reap->parent; 995 KKASSERT(reap->p == NULL); 996 reap->parent = NULL; 997 kfree(reap, M_REAPER); 998 } else { 999 next = NULL; 1000 } 1001 } 1002 } 1003 1004 /* 1005 * Initialize a static or newly allocated reaper structure 1006 */ 1007 void 1008 reaper_init(struct proc *p, struct sysreaper *reap) 1009 { 1010 reap->parent = p->p_reaper; 1011 reap->p = p; 1012 if (p == initproc) { 1013 reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT; 1014 reap->refs = 2; 1015 } else { 1016 reap->flags = REAPER_STAT_OWNED; 1017 reap->refs = 1; 1018 } 1019 lockinit(&reap->lock, "subrp", 0, 0); 1020 cpu_sfence(); 1021 p->p_reaper = reap; 1022 } 1023 1024 /* 1025 * Called with p->p_token held during exit. 1026 * 1027 * This is a bit simpler than RELEASE because there are no threads remaining 1028 * to race. We only release if we own the reaper, the exit code will handle 1029 * the final p_reaper release. 1030 */ 1031 struct sysreaper * 1032 reaper_exit(struct proc *p) 1033 { 1034 struct sysreaper *reap; 1035 1036 /* 1037 * Release acquired reaper 1038 */ 1039 if ((reap = p->p_reaper) != NULL && reap->p == p) { 1040 lockmgr(&reap->lock, LK_EXCLUSIVE); 1041 p->p_reaper = reap->parent; 1042 if (p->p_reaper) 1043 reaper_hold(p->p_reaper); 1044 reap->p = NULL; 1045 lockmgr(&reap->lock, LK_RELEASE); 1046 reaper_drop(reap); 1047 } 1048 1049 /* 1050 * Return and clear reaper (caller is holding p_token for us) 1051 * (reap->p does not equal p). Caller must drop it. 1052 */ 1053 if ((reap = p->p_reaper) != NULL) { 1054 p->p_reaper = NULL; 1055 } 1056 return reap; 1057 } 1058 1059 /* 1060 * Return a held (PHOLD) process representing the reaper for process (p). 1061 * NULL should not normally be returned. Caller should PRELE() the returned 1062 * reaper process when finished. 1063 * 1064 * Remove dead internal nodes while we are at it. 1065 * 1066 * Process (p)'s token must be held on call. 1067 * The returned process's token is NOT acquired by this routine. 1068 */ 1069 struct proc * 1070 reaper_get(struct sysreaper *reap) 1071 { 1072 struct sysreaper *next; 1073 struct proc *reproc; 1074 1075 if (reap == NULL) 1076 return NULL; 1077 1078 /* 1079 * Extra hold for loop 1080 */ 1081 reaper_hold(reap); 1082 1083 while (reap) { 1084 lockmgr(&reap->lock, LK_SHARED); 1085 if (reap->p) { 1086 /* 1087 * Probable reaper 1088 */ 1089 if (reap->p) { 1090 reproc = reap->p; 1091 PHOLD(reproc); 1092 lockmgr(&reap->lock, LK_RELEASE); 1093 reaper_drop(reap); 1094 return reproc; 1095 } 1096 1097 /* 1098 * Raced, try again 1099 */ 1100 lockmgr(&reap->lock, LK_RELEASE); 1101 continue; 1102 } 1103 1104 /* 1105 * Traverse upwards in the reaper topology, destroy 1106 * dead internal nodes when possible. 1107 * 1108 * NOTE: Our ref on next means that a dead node should 1109 * have 2 (ours and reap->parent's). 1110 */ 1111 next = reap->parent; 1112 while (next) { 1113 reaper_hold(next); 1114 if (next->refs == 2 && next->p == NULL) { 1115 lockmgr(&reap->lock, LK_RELEASE); 1116 lockmgr(&reap->lock, LK_EXCLUSIVE); 1117 if (next->refs == 2 && 1118 reap->parent == next && 1119 next->p == NULL) { 1120 /* 1121 * reap->parent inherits ref from next. 1122 */ 1123 reap->parent = next->parent; 1124 next->parent = NULL; 1125 reaper_drop(next); /* ours */ 1126 reaper_drop(next); /* old parent */ 1127 next = reap->parent; 1128 continue; /* possible chain */ 1129 } 1130 } 1131 break; 1132 } 1133 lockmgr(&reap->lock, LK_RELEASE); 1134 reaper_drop(reap); 1135 reap = next; 1136 } 1137 return NULL; 1138 } 1139