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 static struct lock reaper_lock = LOCK_INITIALIZER("reapgl", 0, 0); 90 91 int forksleep; /* Place for fork1() to sleep on. */ 92 93 /* 94 * Red-Black tree support for LWPs 95 */ 96 97 static int 98 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2) 99 { 100 if (lp1->lwp_tid < lp2->lwp_tid) 101 return(-1); 102 if (lp1->lwp_tid > lp2->lwp_tid) 103 return(1); 104 return(0); 105 } 106 107 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid); 108 109 /* 110 * When forking, memory underpinning umtx-supported mutexes may be set 111 * COW causing the physical address to change. We must wakeup any threads 112 * blocked on the physical address to allow them to re-resolve their VM. 113 * 114 * (caller is holding p->p_token) 115 */ 116 static void 117 wake_umtx_threads(struct proc *p1) 118 { 119 struct lwp *lp; 120 struct thread *td; 121 122 RB_FOREACH(lp, lwp_rb_tree, &p1->p_lwp_tree) { 123 td = lp->lwp_thread; 124 if (td && (td->td_flags & TDF_TSLEEPQ) && 125 (td->td_wdomain & PDOMAIN_MASK) == PDOMAIN_UMTX) { 126 wakeup_domain(td->td_wchan, PDOMAIN_UMTX); 127 } 128 } 129 } 130 131 /* 132 * fork() system call 133 */ 134 int 135 sys_fork(struct fork_args *uap) 136 { 137 struct lwp *lp = curthread->td_lwp; 138 struct proc *p2; 139 int error; 140 141 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2); 142 if (error == 0) { 143 PHOLD(p2); 144 start_forked_proc(lp, p2); 145 uap->sysmsg_fds[0] = p2->p_pid; 146 uap->sysmsg_fds[1] = 0; 147 PRELE(p2); 148 } 149 return error; 150 } 151 152 /* 153 * vfork() system call 154 */ 155 int 156 sys_vfork(struct vfork_args *uap) 157 { 158 struct lwp *lp = curthread->td_lwp; 159 struct proc *p2; 160 int error; 161 162 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2); 163 if (error == 0) { 164 PHOLD(p2); 165 start_forked_proc(lp, p2); 166 uap->sysmsg_fds[0] = p2->p_pid; 167 uap->sysmsg_fds[1] = 0; 168 PRELE(p2); 169 } 170 return error; 171 } 172 173 /* 174 * Handle rforks. An rfork may (1) operate on the current process without 175 * creating a new, (2) create a new process that shared the current process's 176 * vmspace, signals, and/or descriptors, or (3) create a new process that does 177 * not share these things (normal fork). 178 * 179 * Note that we only call start_forked_proc() if a new process is actually 180 * created. 181 * 182 * rfork { int flags } 183 */ 184 int 185 sys_rfork(struct rfork_args *uap) 186 { 187 struct lwp *lp = curthread->td_lwp; 188 struct proc *p2; 189 int error; 190 191 if ((uap->flags & RFKERNELONLY) != 0) 192 return (EINVAL); 193 194 error = fork1(lp, uap->flags | RFPGLOCK, &p2); 195 if (error == 0) { 196 if (p2) { 197 PHOLD(p2); 198 start_forked_proc(lp, p2); 199 uap->sysmsg_fds[0] = p2->p_pid; 200 uap->sysmsg_fds[1] = 0; 201 PRELE(p2); 202 } else { 203 uap->sysmsg_fds[0] = 0; 204 uap->sysmsg_fds[1] = 0; 205 } 206 } 207 return error; 208 } 209 210 static int 211 lwp_create1(struct lwp_params *uprm, const cpumask_t *umask) 212 { 213 struct proc *p = curproc; 214 struct lwp *lp; 215 struct lwp_params params; 216 cpumask_t *mask = NULL, mask0; 217 int error; 218 219 error = copyin(uprm, ¶ms, sizeof(params)); 220 if (error) 221 goto fail2; 222 223 if (umask != NULL) { 224 error = copyin(umask, &mask0, sizeof(mask0)); 225 if (error) 226 goto fail2; 227 CPUMASK_ANDMASK(mask0, smp_active_mask); 228 if (CPUMASK_TESTNZERO(mask0)) 229 mask = &mask0; 230 } 231 232 lwkt_gettoken(&p->p_token); 233 plimit_lwp_fork(p); /* force exclusive access */ 234 lp = lwp_fork(curthread->td_lwp, p, RFPROC | RFMEM, mask); 235 error = cpu_prepare_lwp(lp, ¶ms); 236 if (error) 237 goto fail; 238 if (params.lwp_tid1 != NULL && 239 (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid)))) 240 goto fail; 241 if (params.lwp_tid2 != NULL && 242 (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid)))) 243 goto fail; 244 245 /* 246 * Now schedule the new lwp. 247 */ 248 p->p_usched->resetpriority(lp); 249 crit_enter(); 250 lp->lwp_stat = LSRUN; 251 p->p_usched->setrunqueue(lp); 252 crit_exit(); 253 lwkt_reltoken(&p->p_token); 254 255 return (0); 256 257 fail: 258 /* 259 * Make sure no one is using this lwp, before it is removed from 260 * the tree. If we didn't wait it here, lwp tree iteration with 261 * blocking operation would be broken. 262 */ 263 while (lp->lwp_lock > 0) 264 tsleep(lp, 0, "lwpfail", 1); 265 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 266 --p->p_nthreads; 267 /* lwp_dispose expects an exited lwp, and a held proc */ 268 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 269 lp->lwp_thread->td_flags |= TDF_EXITING; 270 lwkt_remove_tdallq(lp->lwp_thread); 271 PHOLD(p); 272 biosched_done(lp->lwp_thread); 273 dsched_exit_thread(lp->lwp_thread); 274 lwp_dispose(lp); 275 lwkt_reltoken(&p->p_token); 276 fail2: 277 return (error); 278 } 279 280 /* 281 * Low level thread create used by pthreads. 282 */ 283 int 284 sys_lwp_create(struct lwp_create_args *uap) 285 { 286 287 return (lwp_create1(uap->params, NULL)); 288 } 289 290 int 291 sys_lwp_create2(struct lwp_create2_args *uap) 292 { 293 294 return (lwp_create1(uap->params, uap->mask)); 295 } 296 297 int nprocs = 1; /* process 0 */ 298 299 int 300 fork1(struct lwp *lp1, int flags, struct proc **procp) 301 { 302 struct proc *p1 = lp1->lwp_proc; 303 struct proc *p2; 304 struct proc *pptr; 305 struct pgrp *p1grp; 306 struct pgrp *plkgrp; 307 struct sysreaper *reap; 308 uid_t uid; 309 int ok, error; 310 static int curfail = 0; 311 static struct timeval lastfail; 312 struct forklist *ep; 313 struct filedesc_to_leader *fdtol; 314 315 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 316 return (EINVAL); 317 318 lwkt_gettoken(&p1->p_token); 319 plkgrp = NULL; 320 p2 = NULL; 321 322 /* 323 * Here we don't create a new process, but we divorce 324 * certain parts of a process from itself. 325 */ 326 if ((flags & RFPROC) == 0) { 327 /* 328 * This kind of stunt does not work anymore if 329 * there are native threads (lwps) running 330 */ 331 if (p1->p_nthreads != 1) { 332 error = EINVAL; 333 goto done; 334 } 335 336 vm_fork(p1, 0, flags); 337 if ((flags & RFMEM) == 0) 338 wake_umtx_threads(p1); 339 340 /* 341 * Close all file descriptors. 342 */ 343 if (flags & RFCFDG) { 344 struct filedesc *fdtmp; 345 fdtmp = fdinit(p1); 346 fdfree(p1, fdtmp); 347 } 348 349 /* 350 * Unshare file descriptors (from parent.) 351 */ 352 if (flags & RFFDG) { 353 if (p1->p_fd->fd_refcnt > 1) { 354 struct filedesc *newfd; 355 error = fdcopy(p1, &newfd); 356 if (error != 0) { 357 error = ENOMEM; 358 goto done; 359 } 360 fdfree(p1, newfd); 361 } 362 } 363 *procp = NULL; 364 error = 0; 365 goto done; 366 } 367 368 /* 369 * Interlock against process group signal delivery. If signals 370 * are pending after the interlock is obtained we have to restart 371 * the system call to process the signals. If we don't the child 372 * can miss a pgsignal (such as ^C) sent during the fork. 373 * 374 * We can't use CURSIG() here because it will process any STOPs 375 * and cause the process group lock to be held indefinitely. If 376 * a STOP occurs, the fork will be restarted after the CONT. 377 */ 378 p1grp = p1->p_pgrp; 379 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) { 380 pgref(plkgrp); 381 lockmgr(&plkgrp->pg_lock, LK_SHARED); 382 if (CURSIG_NOBLOCK(lp1)) { 383 error = ERESTART; 384 goto done; 385 } 386 } 387 388 /* 389 * Although process entries are dynamically created, we still keep 390 * a global limit on the maximum number we will create. Don't allow 391 * a nonprivileged user to use the last ten processes; don't let root 392 * exceed the limit. The variable nprocs is the current number of 393 * processes, maxproc is the limit. 394 */ 395 uid = lp1->lwp_thread->td_ucred->cr_ruid; 396 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 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 * Increment the nprocs resource before blocking can occur. There 407 * are hard-limits as to the number of processes that can run. 408 */ 409 atomic_add_int(&nprocs, 1); 410 411 /* 412 * Increment the count of procs running with this uid. This also 413 * applies to root. 414 */ 415 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1, 416 plimit_getadjvalue(RLIMIT_NPROC)); 417 if (!ok) { 418 /* 419 * Back out the process count 420 */ 421 atomic_add_int(&nprocs, -1); 422 if (ppsratecheck(&lastfail, &curfail, 1)) { 423 kprintf("maxproc limit of %jd " 424 "exceeded by \"%s\" uid %d, " 425 "please see tuning(7) and login.conf(5).\n", 426 plimit_getadjvalue(RLIMIT_NPROC), 427 p1->p_comm, 428 uid); 429 } 430 tsleep(&forksleep, 0, "fork", hz / 2); 431 error = EAGAIN; 432 goto done; 433 } 434 435 /* 436 * Allocate a new process, don't get fancy: zero the structure. 437 */ 438 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO); 439 440 /* 441 * Core initialization. SIDL is a safety state that protects the 442 * partially initialized process once it starts getting hooked 443 * into system structures and becomes addressable. 444 * 445 * We must be sure to acquire p2->p_token as well, we must hold it 446 * once the process is on the allproc list to avoid things such 447 * as competing modifications to p_flags. 448 */ 449 mycpu->gd_forkid += ncpus; 450 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid; 451 p2->p_lasttid = 0; /* first tid will be 1 */ 452 p2->p_stat = SIDL; 453 454 /* 455 * NOTE: Process 0 will not have a reaper, but process 1 (init) and 456 * all other processes always will. 457 */ 458 if ((reap = p1->p_reaper) != NULL) { 459 reaper_hold(reap); 460 p2->p_reaper = reap; 461 } else { 462 p2->p_reaper = NULL; 463 } 464 465 RB_INIT(&p2->p_lwp_tree); 466 spin_init(&p2->p_spin, "procfork1"); 467 lwkt_token_init(&p2->p_token, "proc"); 468 lwkt_gettoken(&p2->p_token); 469 470 /* 471 * Setup linkage for kernel based threading XXX lwp. Also add the 472 * process to the allproclist. 473 * 474 * The process structure is addressable after this point. 475 */ 476 if (flags & RFTHREAD) { 477 p2->p_peers = p1->p_peers; 478 p1->p_peers = p2; 479 p2->p_leader = p1->p_leader; 480 } else { 481 p2->p_leader = p2; 482 } 483 proc_add_allproc(p2); 484 485 /* 486 * Initialize the section which is copied verbatim from the parent. 487 */ 488 bcopy(&p1->p_startcopy, &p2->p_startcopy, 489 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 490 491 /* 492 * Duplicate sub-structures as needed. Increase reference counts 493 * on shared objects. 494 * 495 * NOTE: because we are now on the allproc list it is possible for 496 * other consumers to gain temporary references to p2 497 * (p2->p_lock can change). 498 */ 499 if (p1->p_flags & P_PROFIL) 500 startprofclock(p2); 501 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred); 502 503 if (jailed(p2->p_ucred)) 504 p2->p_flags |= P_JAILED; 505 506 if (p2->p_args) 507 refcount_acquire(&p2->p_args->ar_ref); 508 509 p2->p_usched = p1->p_usched; 510 /* XXX: verify copy of the secondary iosched stuff */ 511 dsched_enter_proc(p2); 512 513 if (flags & RFSIGSHARE) { 514 p2->p_sigacts = p1->p_sigacts; 515 refcount_acquire(&p2->p_sigacts->ps_refcnt); 516 } else { 517 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts), 518 M_SUBPROC, M_WAITOK); 519 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts)); 520 refcount_init(&p2->p_sigacts->ps_refcnt, 1); 521 } 522 if (flags & RFLINUXTHPN) 523 p2->p_sigparent = SIGUSR1; 524 else 525 p2->p_sigparent = SIGCHLD; 526 527 /* bump references to the text vnode (for procfs) */ 528 p2->p_textvp = p1->p_textvp; 529 if (p2->p_textvp) 530 vref(p2->p_textvp); 531 532 /* copy namecache handle to the text file */ 533 if (p1->p_textnch.mount) 534 cache_copy(&p1->p_textnch, &p2->p_textnch); 535 536 /* 537 * Handle file descriptors 538 */ 539 if (flags & RFCFDG) { 540 p2->p_fd = fdinit(p1); 541 fdtol = NULL; 542 } else if (flags & RFFDG) { 543 error = fdcopy(p1, &p2->p_fd); 544 if (error != 0) { 545 error = ENOMEM; 546 goto done; 547 } 548 fdtol = NULL; 549 } else { 550 p2->p_fd = fdshare(p1); 551 if (p1->p_fdtol == NULL) { 552 p1->p_fdtol = filedesc_to_leader_alloc(NULL, 553 p1->p_leader); 554 } 555 if ((flags & RFTHREAD) != 0) { 556 /* 557 * Shared file descriptor table and 558 * shared process leaders. 559 */ 560 fdtol = p1->p_fdtol; 561 fdtol->fdl_refcount++; 562 } else { 563 /* 564 * Shared file descriptor table, and 565 * different process leaders 566 */ 567 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 568 } 569 } 570 p2->p_fdtol = fdtol; 571 p2->p_limit = plimit_fork(p1); 572 573 /* 574 * Adjust depth for resource downscaling 575 */ 576 if ((p2->p_depth & 31) != 31) 577 ++p2->p_depth; 578 579 /* 580 * Preserve some more flags in subprocess. P_PROFIL has already 581 * been preserved. 582 */ 583 p2->p_flags |= p1->p_flags & P_SUGID; 584 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT)) 585 p2->p_flags |= P_CONTROLT; 586 if (flags & RFPPWAIT) { 587 p2->p_flags |= P_PPWAIT; 588 if (p1->p_upmap) 589 atomic_add_int(&p1->p_upmap->invfork, 1); 590 } 591 592 /* 593 * Inherit the virtual kernel structure (allows a virtual kernel 594 * to fork to simulate multiple cpus). 595 */ 596 if (p1->p_vkernel) 597 vkernel_inherit(p1, p2); 598 599 /* 600 * Once we are on a pglist we may receive signals. XXX we might 601 * race a ^C being sent to the process group by not receiving it 602 * at all prior to this line. 603 */ 604 pgref(p1grp); 605 lwkt_gettoken(&p1grp->pg_token); 606 LIST_INSERT_AFTER(p1, p2, p_pglist); 607 lwkt_reltoken(&p1grp->pg_token); 608 609 /* 610 * Attach the new process to its parent. 611 * 612 * If RFNOWAIT is set, the newly created process becomes a child 613 * of the reaper (typically init). This effectively disassociates 614 * the child from the parent. 615 * 616 * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts. 617 */ 618 if (flags & RFNOWAIT) { 619 pptr = reaper_get(reap); 620 if (pptr == NULL) { 621 pptr = initproc; 622 PHOLD(pptr); 623 } 624 } else { 625 pptr = p1; 626 } 627 p2->p_pptr = pptr; 628 p2->p_ppid = pptr->p_pid; 629 LIST_INIT(&p2->p_children); 630 631 lwkt_gettoken(&pptr->p_token); 632 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 633 lwkt_reltoken(&pptr->p_token); 634 635 if (flags & RFNOWAIT) 636 PRELE(pptr); 637 638 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 639 callout_init_mp(&p2->p_ithandle); 640 641 #ifdef KTRACE 642 /* 643 * Copy traceflag and tracefile if enabled. If not inherited, 644 * these were zeroed above but we still could have a trace race 645 * so make sure p2's p_tracenode is NULL. 646 */ 647 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) { 648 p2->p_traceflag = p1->p_traceflag; 649 p2->p_tracenode = ktrinherit(p1->p_tracenode); 650 } 651 #endif 652 653 /* 654 * This begins the section where we must prevent the parent 655 * from being swapped. 656 * 657 * Gets PRELE'd in the caller in start_forked_proc(). 658 */ 659 PHOLD(p1); 660 661 vm_fork(p1, p2, flags); 662 if ((flags & RFMEM) == 0) 663 wake_umtx_threads(p1); 664 665 /* 666 * Create the first lwp associated with the new proc. 667 * It will return via a different execution path later, directly 668 * into userland, after it was put on the runq by 669 * start_forked_proc(). 670 */ 671 lwp_fork(lp1, p2, flags, NULL); 672 673 if (flags == (RFFDG | RFPROC | RFPGLOCK)) { 674 mycpu->gd_cnt.v_forks++; 675 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + 676 p2->p_vmspace->vm_ssize; 677 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) { 678 mycpu->gd_cnt.v_vforks++; 679 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 680 p2->p_vmspace->vm_ssize; 681 } else if (p1 == &proc0) { 682 mycpu->gd_cnt.v_kthreads++; 683 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 684 p2->p_vmspace->vm_ssize; 685 } else { 686 mycpu->gd_cnt.v_rforks++; 687 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 688 p2->p_vmspace->vm_ssize; 689 } 690 691 /* 692 * Both processes are set up, now check if any loadable modules want 693 * to adjust anything. 694 * What if they have an error? XXX 695 */ 696 TAILQ_FOREACH(ep, &fork_list, next) { 697 (*ep->function)(p1, p2, flags); 698 } 699 700 /* 701 * Set the start time. Note that the process is not runnable. The 702 * caller is responsible for making it runnable. 703 */ 704 microtime(&p2->p_start); 705 p2->p_acflag = AFORK; 706 707 /* 708 * tell any interested parties about the new process 709 */ 710 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 711 712 /* 713 * Return child proc pointer to parent. 714 */ 715 *procp = p2; 716 error = 0; 717 done: 718 if (p2) 719 lwkt_reltoken(&p2->p_token); 720 lwkt_reltoken(&p1->p_token); 721 if (plkgrp) { 722 lockmgr(&plkgrp->pg_lock, LK_RELEASE); 723 pgrel(plkgrp); 724 } 725 return (error); 726 } 727 728 static struct lwp * 729 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags, 730 const cpumask_t *mask) 731 { 732 globaldata_t gd = mycpu; 733 struct lwp *lp; 734 struct thread *td; 735 736 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO); 737 738 lp->lwp_proc = destproc; 739 lp->lwp_vmspace = destproc->p_vmspace; 740 lp->lwp_stat = LSRUN; 741 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy, 742 (unsigned) ((caddr_t)&lp->lwp_endcopy - 743 (caddr_t)&lp->lwp_startcopy)); 744 if (mask != NULL) 745 lp->lwp_cpumask = *mask; 746 747 /* 748 * Reset the sigaltstack if memory is shared, otherwise inherit 749 * it. 750 */ 751 if (flags & RFMEM) { 752 lp->lwp_sigstk.ss_flags = SS_DISABLE; 753 lp->lwp_sigstk.ss_size = 0; 754 lp->lwp_sigstk.ss_sp = NULL; 755 lp->lwp_flags &= ~LWP_ALTSTACK; 756 } else { 757 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK; 758 } 759 760 /* 761 * Set cpbase to the last timeout that occured (not the upcoming 762 * timeout). 763 * 764 * A critical section is required since a timer IPI can update 765 * scheduler specific data. 766 */ 767 crit_enter(); 768 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic; 769 destproc->p_usched->heuristic_forking(origlp, lp); 770 crit_exit(); 771 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask); 772 lwkt_token_init(&lp->lwp_token, "lwp_token"); 773 spin_init(&lp->lwp_spin, "lwptoken"); 774 775 /* 776 * Assign the thread to the current cpu to begin with so we 777 * can manipulate it. 778 */ 779 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0); 780 lp->lwp_thread = td; 781 td->td_ucred = crhold(destproc->p_ucred); 782 td->td_proc = destproc; 783 td->td_lwp = lp; 784 td->td_switch = cpu_heavy_switch; 785 #ifdef NO_LWKT_SPLIT_USERPRI 786 lwkt_setpri(td, TDPRI_USER_NORM); 787 #else 788 lwkt_setpri(td, TDPRI_KERN_USER); 789 #endif 790 lwkt_set_comm(td, "%s", destproc->p_comm); 791 792 /* 793 * cpu_fork will copy and update the pcb, set up the kernel stack, 794 * and make the child ready to run. 795 */ 796 cpu_fork(origlp, lp, flags); 797 kqueue_init(&lp->lwp_kqueue, destproc->p_fd); 798 799 /* 800 * Assign a TID to the lp. Loop until the insert succeeds (returns 801 * NULL). 802 * 803 * If we are in a vfork assign the same TID as the lwp that did the 804 * vfork(). This way if the user program messes around with 805 * pthread calls inside the vfork(), it will operate like an 806 * extension of the (blocked) parent. Also note that since the 807 * address space is being shared, insofar as pthreads is concerned, 808 * the code running in the vfork() is part of the original process. 809 */ 810 if (flags & RFPPWAIT) { 811 lp->lwp_tid = origlp->lwp_tid - 1; 812 } else { 813 lp->lwp_tid = destproc->p_lasttid; 814 } 815 816 do { 817 if (++lp->lwp_tid <= 0) 818 lp->lwp_tid = 1; 819 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL); 820 821 destproc->p_lasttid = lp->lwp_tid; 822 destproc->p_nthreads++; 823 824 /* 825 * This flag is set and never cleared. It means that the process 826 * was threaded at some point. Used to improve exit performance. 827 */ 828 destproc->p_flags |= P_MAYBETHREADED; 829 830 return (lp); 831 } 832 833 /* 834 * The next two functionms are general routines to handle adding/deleting 835 * items on the fork callout list. 836 * 837 * at_fork(): 838 * Take the arguments given and put them onto the fork callout list, 839 * However first make sure that it's not already there. 840 * Returns 0 on success or a standard error number. 841 */ 842 int 843 at_fork(forklist_fn function) 844 { 845 struct forklist *ep; 846 847 #ifdef INVARIANTS 848 /* let the programmer know if he's been stupid */ 849 if (rm_at_fork(function)) { 850 kprintf("WARNING: fork callout entry (%p) already present\n", 851 function); 852 } 853 #endif 854 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 855 ep->function = function; 856 TAILQ_INSERT_TAIL(&fork_list, ep, next); 857 return (0); 858 } 859 860 /* 861 * Scan the exit callout list for the given item and remove it.. 862 * Returns the number of items removed (0 or 1) 863 */ 864 int 865 rm_at_fork(forklist_fn function) 866 { 867 struct forklist *ep; 868 869 TAILQ_FOREACH(ep, &fork_list, next) { 870 if (ep->function == function) { 871 TAILQ_REMOVE(&fork_list, ep, next); 872 kfree(ep, M_ATFORK); 873 return(1); 874 } 875 } 876 return (0); 877 } 878 879 /* 880 * Add a forked process to the run queue after any remaining setup, such 881 * as setting the fork handler, has been completed. 882 * 883 * p2 is held by the caller. 884 */ 885 void 886 start_forked_proc(struct lwp *lp1, struct proc *p2) 887 { 888 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2); 889 int pflags; 890 891 /* 892 * Move from SIDL to RUN queue, and activate the process's thread. 893 * Activation of the thread effectively makes the process "a" 894 * current process, so we do not setrunqueue(). 895 * 896 * YYY setrunqueue works here but we should clean up the trampoline 897 * code so we just schedule the LWKT thread and let the trampoline 898 * deal with the userland scheduler on return to userland. 899 */ 900 KASSERT(p2->p_stat == SIDL, 901 ("cannot start forked process, bad status: %p", p2)); 902 p2->p_usched->resetpriority(lp2); 903 crit_enter(); 904 p2->p_stat = SACTIVE; 905 lp2->lwp_stat = LSRUN; 906 p2->p_usched->setrunqueue(lp2); 907 crit_exit(); 908 909 /* 910 * Now can be swapped. 911 */ 912 PRELE(lp1->lwp_proc); 913 914 /* 915 * Preserve synchronization semantics of vfork. P_PPWAIT is set in 916 * the child until it has retired the parent's resources. The parent 917 * must wait for the flag to be cleared by the child. 918 * 919 * Interlock the flag/tsleep with atomic ops to avoid unnecessary 920 * p_token conflicts. 921 * 922 * XXX Is this use of an atomic op on a field that is not normally 923 * manipulated with atomic ops ok? 924 */ 925 while ((pflags = p2->p_flags) & P_PPWAIT) { 926 cpu_ccfence(); 927 tsleep_interlock(lp1->lwp_proc, 0); 928 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags)) 929 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0); 930 } 931 } 932 933 /* 934 * procctl (idtype_t idtype, id_t id, int cmd, void *arg) 935 */ 936 int 937 sys_procctl(struct procctl_args *uap) 938 { 939 struct proc *p = curproc; 940 struct proc *p2; 941 struct sysreaper *reap; 942 union reaper_info udata; 943 int error; 944 945 if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid) 946 return EINVAL; 947 948 switch(uap->cmd) { 949 case PROC_REAP_ACQUIRE: 950 lwkt_gettoken(&p->p_token); 951 reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO); 952 if (p->p_reaper == NULL || p->p_reaper->p != p) { 953 reaper_init(p, reap); 954 error = 0; 955 } else { 956 kfree(reap, M_REAPER); 957 error = EALREADY; 958 } 959 lwkt_reltoken(&p->p_token); 960 break; 961 case PROC_REAP_RELEASE: 962 lwkt_gettoken(&p->p_token); 963 release_again: 964 reap = p->p_reaper; 965 KKASSERT(reap != NULL); 966 if (reap->p == p) { 967 reaper_hold(reap); /* in case of thread race */ 968 lockmgr(&reap->lock, LK_EXCLUSIVE); 969 if (reap->p != p) { 970 lockmgr(&reap->lock, LK_RELEASE); 971 reaper_drop(reap); 972 goto release_again; 973 } 974 reap->p = NULL; 975 p->p_reaper = reap->parent; 976 if (p->p_reaper) 977 reaper_hold(p->p_reaper); 978 lockmgr(&reap->lock, LK_RELEASE); 979 reaper_drop(reap); /* our ref */ 980 reaper_drop(reap); /* old p_reaper ref */ 981 error = 0; 982 } else { 983 error = ENOTCONN; 984 } 985 lwkt_reltoken(&p->p_token); 986 break; 987 case PROC_REAP_STATUS: 988 bzero(&udata, sizeof(udata)); 989 lwkt_gettoken_shared(&p->p_token); 990 if ((reap = p->p_reaper) != NULL && reap->p == p) { 991 udata.status.flags = reap->flags; 992 udata.status.refs = reap->refs - 1; /* minus ours */ 993 } 994 p2 = LIST_FIRST(&p->p_children); 995 udata.status.pid_head = p2 ? p2->p_pid : -1; 996 lwkt_reltoken(&p->p_token); 997 998 if (uap->data) { 999 error = copyout(&udata, uap->data, 1000 sizeof(udata.status)); 1001 } else { 1002 error = 0; 1003 } 1004 break; 1005 default: 1006 error = EINVAL; 1007 break; 1008 } 1009 return error; 1010 } 1011 1012 /* 1013 * Bump ref on reaper, preventing destruction 1014 */ 1015 void 1016 reaper_hold(struct sysreaper *reap) 1017 { 1018 KKASSERT(reap->refs > 0); 1019 refcount_acquire(&reap->refs); 1020 } 1021 1022 /* 1023 * Drop ref on reaper, destroy the structure on the 1->0 1024 * transition and loop on the parent. 1025 */ 1026 void 1027 reaper_drop(struct sysreaper *next) 1028 { 1029 struct sysreaper *reap; 1030 1031 while ((reap = next) != NULL) { 1032 if (refcount_release(&reap->refs)) { 1033 next = reap->parent; 1034 KKASSERT(reap->p == NULL); 1035 lockmgr(&reaper_lock, LK_EXCLUSIVE); 1036 reap->parent = NULL; 1037 kfree(reap, M_REAPER); 1038 lockmgr(&reaper_lock, LK_RELEASE); 1039 } else { 1040 next = NULL; 1041 } 1042 } 1043 } 1044 1045 /* 1046 * Initialize a static or newly allocated reaper structure 1047 */ 1048 void 1049 reaper_init(struct proc *p, struct sysreaper *reap) 1050 { 1051 reap->parent = p->p_reaper; 1052 reap->p = p; 1053 if (p == initproc) { 1054 reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT; 1055 reap->refs = 2; 1056 } else { 1057 reap->flags = REAPER_STAT_OWNED; 1058 reap->refs = 1; 1059 } 1060 lockinit(&reap->lock, "subrp", 0, 0); 1061 cpu_sfence(); 1062 p->p_reaper = reap; 1063 } 1064 1065 /* 1066 * Called with p->p_token held during exit. 1067 * 1068 * This is a bit simpler than RELEASE because there are no threads remaining 1069 * to race. We only release if we own the reaper, the exit code will handle 1070 * the final p_reaper release. 1071 */ 1072 struct sysreaper * 1073 reaper_exit(struct proc *p) 1074 { 1075 struct sysreaper *reap; 1076 1077 /* 1078 * Release acquired reaper 1079 */ 1080 if ((reap = p->p_reaper) != NULL && reap->p == p) { 1081 lockmgr(&reap->lock, LK_EXCLUSIVE); 1082 p->p_reaper = reap->parent; 1083 if (p->p_reaper) 1084 reaper_hold(p->p_reaper); 1085 reap->p = NULL; 1086 lockmgr(&reap->lock, LK_RELEASE); 1087 reaper_drop(reap); 1088 } 1089 1090 /* 1091 * Return and clear reaper (caller is holding p_token for us) 1092 * (reap->p does not equal p). Caller must drop it. 1093 */ 1094 if ((reap = p->p_reaper) != NULL) { 1095 p->p_reaper = NULL; 1096 } 1097 return reap; 1098 } 1099 1100 /* 1101 * Return a held (PHOLD) process representing the reaper for process (p). 1102 * NULL should not normally be returned. Caller should PRELE() the returned 1103 * reaper process when finished. 1104 * 1105 * Remove dead internal nodes while we are at it. 1106 * 1107 * Process (p)'s token must be held on call. 1108 * The returned process's token is NOT acquired by this routine. 1109 */ 1110 struct proc * 1111 reaper_get(struct sysreaper *reap) 1112 { 1113 struct sysreaper *next; 1114 struct proc *reproc; 1115 1116 if (reap == NULL) 1117 return NULL; 1118 1119 /* 1120 * Extra hold for loop 1121 */ 1122 reaper_hold(reap); 1123 1124 while (reap) { 1125 lockmgr(&reap->lock, LK_SHARED); 1126 if (reap->p) { 1127 /* 1128 * Probable reaper 1129 */ 1130 if (reap->p) { 1131 reproc = reap->p; 1132 PHOLD(reproc); 1133 lockmgr(&reap->lock, LK_RELEASE); 1134 reaper_drop(reap); 1135 return reproc; 1136 } 1137 1138 /* 1139 * Raced, try again 1140 */ 1141 lockmgr(&reap->lock, LK_RELEASE); 1142 continue; 1143 } 1144 1145 /* 1146 * Traverse upwards in the reaper topology, destroy 1147 * dead internal nodes when possible. 1148 * 1149 * NOTE: Our ref on next means that a dead node should 1150 * have 2 (ours and reap->parent's). 1151 */ 1152 next = reap->parent; 1153 while (next) { 1154 reaper_hold(next); 1155 if (next->refs == 2 && next->p == NULL) { 1156 lockmgr(&reap->lock, LK_RELEASE); 1157 lockmgr(&reap->lock, LK_EXCLUSIVE); 1158 if (next->refs == 2 && 1159 reap->parent == next && 1160 next->p == NULL) { 1161 /* 1162 * reap->parent inherits ref from next. 1163 */ 1164 reap->parent = next->parent; 1165 next->parent = NULL; 1166 reaper_drop(next); /* ours */ 1167 reaper_drop(next); /* old parent */ 1168 next = reap->parent; 1169 continue; /* possible chain */ 1170 } 1171 } 1172 break; 1173 } 1174 lockmgr(&reap->lock, LK_RELEASE); 1175 reaper_drop(reap); 1176 reap = next; 1177 } 1178 return NULL; 1179 } 1180 1181 /* 1182 * Test that the sender is allowed to send a signal to the target. 1183 * The sender process is assumed to have a stable reaper. The 1184 * target can be e.g. from a scan callback. 1185 * 1186 * Target cannot be the reaper process itself unless reaper_ok is specified, 1187 * or sender == target. 1188 */ 1189 int 1190 reaper_sigtest(struct proc *sender, struct proc *target, int reaper_ok) 1191 { 1192 struct sysreaper *sreap; 1193 struct sysreaper *reap; 1194 int r; 1195 1196 sreap = sender->p_reaper; 1197 if (sreap == NULL) 1198 return 1; 1199 1200 if (sreap == target->p_reaper) { 1201 if (sreap->p == target && sreap->p != sender && reaper_ok == 0) 1202 return 0; 1203 return 1; 1204 } 1205 lockmgr(&reaper_lock, LK_SHARED); 1206 r = 0; 1207 for (reap = target->p_reaper; reap; reap = reap->parent) { 1208 if (sreap == reap) { 1209 if (sreap->p != target || reaper_ok) 1210 r = 1; 1211 break; 1212 } 1213 } 1214 lockmgr(&reaper_lock, LK_RELEASE); 1215 1216 return r; 1217 } 1218