kern_fork.c revision 315949
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/10/sys/kern/kern_fork.c 315949 2017-03-25 13:33:23Z badger $"); 39 40#include "opt_kdtrace.h" 41#include "opt_ktrace.h" 42#include "opt_kstack_pages.h" 43#include "opt_procdesc.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/sysproto.h> 48#include <sys/eventhandler.h> 49#include <sys/fcntl.h> 50#include <sys/filedesc.h> 51#include <sys/jail.h> 52#include <sys/kernel.h> 53#include <sys/kthread.h> 54#include <sys/sysctl.h> 55#include <sys/lock.h> 56#include <sys/malloc.h> 57#include <sys/mutex.h> 58#include <sys/priv.h> 59#include <sys/proc.h> 60#include <sys/procdesc.h> 61#include <sys/pioctl.h> 62#include <sys/ptrace.h> 63#include <sys/racct.h> 64#include <sys/resourcevar.h> 65#include <sys/sched.h> 66#include <sys/syscall.h> 67#include <sys/vmmeter.h> 68#include <sys/vnode.h> 69#include <sys/acct.h> 70#include <sys/ktr.h> 71#include <sys/ktrace.h> 72#include <sys/unistd.h> 73#include <sys/sdt.h> 74#include <sys/sx.h> 75#include <sys/sysent.h> 76#include <sys/signalvar.h> 77 78#include <security/audit/audit.h> 79#include <security/mac/mac_framework.h> 80 81#include <vm/vm.h> 82#include <vm/pmap.h> 83#include <vm/vm_map.h> 84#include <vm/vm_extern.h> 85#include <vm/uma.h> 86 87#ifdef KDTRACE_HOOKS 88#include <sys/dtrace_bsd.h> 89dtrace_fork_func_t dtrace_fasttrap_fork; 90#endif 91 92SDT_PROVIDER_DECLARE(proc); 93SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); 94 95#ifndef _SYS_SYSPROTO_H_ 96struct fork_args { 97 int dummy; 98}; 99#endif 100 101/* ARGSUSED */ 102int 103sys_fork(struct thread *td, struct fork_args *uap) 104{ 105 int error; 106 struct proc *p2; 107 108 error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0); 109 if (error == 0) { 110 td->td_retval[0] = p2->p_pid; 111 td->td_retval[1] = 0; 112 } 113 return (error); 114} 115 116/* ARGUSED */ 117int 118sys_pdfork(td, uap) 119 struct thread *td; 120 struct pdfork_args *uap; 121{ 122#ifdef PROCDESC 123 int error, fd; 124 struct proc *p2; 125 126 /* 127 * It is necessary to return fd by reference because 0 is a valid file 128 * descriptor number, and the child needs to be able to distinguish 129 * itself from the parent using the return value. 130 */ 131 error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2, 132 &fd, uap->flags); 133 if (error == 0) { 134 td->td_retval[0] = p2->p_pid; 135 td->td_retval[1] = 0; 136 error = copyout(&fd, uap->fdp, sizeof(fd)); 137 } 138 return (error); 139#else 140 return (ENOSYS); 141#endif 142} 143 144/* ARGSUSED */ 145int 146sys_vfork(struct thread *td, struct vfork_args *uap) 147{ 148 int error, flags; 149 struct proc *p2; 150 151 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 152 error = fork1(td, flags, 0, &p2, NULL, 0); 153 if (error == 0) { 154 td->td_retval[0] = p2->p_pid; 155 td->td_retval[1] = 0; 156 } 157 return (error); 158} 159 160int 161sys_rfork(struct thread *td, struct rfork_args *uap) 162{ 163 struct proc *p2; 164 int error; 165 166 /* Don't allow kernel-only flags. */ 167 if ((uap->flags & RFKERNELONLY) != 0) 168 return (EINVAL); 169 170 AUDIT_ARG_FFLAGS(uap->flags); 171 error = fork1(td, uap->flags, 0, &p2, NULL, 0); 172 if (error == 0) { 173 td->td_retval[0] = p2 ? p2->p_pid : 0; 174 td->td_retval[1] = 0; 175 } 176 return (error); 177} 178 179int nprocs = 1; /* process 0 */ 180int lastpid = 0; 181SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 182 "Last used PID"); 183 184/* 185 * Random component to lastpid generation. We mix in a random factor to make 186 * it a little harder to predict. We sanity check the modulus value to avoid 187 * doing it in critical paths. Don't let it be too small or we pointlessly 188 * waste randomness entropy, and don't let it be impossibly large. Using a 189 * modulus that is too big causes a LOT more process table scans and slows 190 * down fork processing as the pidchecked caching is defeated. 191 */ 192static int randompid = 0; 193 194static int 195sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 196{ 197 int error, pid; 198 199 error = sysctl_wire_old_buffer(req, sizeof(int)); 200 if (error != 0) 201 return(error); 202 sx_xlock(&allproc_lock); 203 pid = randompid; 204 error = sysctl_handle_int(oidp, &pid, 0, req); 205 if (error == 0 && req->newptr != NULL) { 206 if (pid < 0 || pid > pid_max - 100) /* out of range */ 207 pid = pid_max - 100; 208 else if (pid < 2) /* NOP */ 209 pid = 0; 210 else if (pid < 100) /* Make it reasonable */ 211 pid = 100; 212 randompid = pid; 213 } 214 sx_xunlock(&allproc_lock); 215 return (error); 216} 217 218SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 219 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 220 221static int 222fork_findpid(int flags) 223{ 224 struct proc *p; 225 int trypid; 226 static int pidchecked = 0; 227 228 /* 229 * Requires allproc_lock in order to iterate over the list 230 * of processes, and proctree_lock to access p_pgrp. 231 */ 232 sx_assert(&allproc_lock, SX_LOCKED); 233 sx_assert(&proctree_lock, SX_LOCKED); 234 235 /* 236 * Find an unused process ID. We remember a range of unused IDs 237 * ready to use (from lastpid+1 through pidchecked-1). 238 * 239 * If RFHIGHPID is set (used during system boot), do not allocate 240 * low-numbered pids. 241 */ 242 trypid = lastpid + 1; 243 if (flags & RFHIGHPID) { 244 if (trypid < 10) 245 trypid = 10; 246 } else { 247 if (randompid) 248 trypid += arc4random() % randompid; 249 } 250retry: 251 /* 252 * If the process ID prototype has wrapped around, 253 * restart somewhat above 0, as the low-numbered procs 254 * tend to include daemons that don't exit. 255 */ 256 if (trypid >= pid_max) { 257 trypid = trypid % pid_max; 258 if (trypid < 100) 259 trypid += 100; 260 pidchecked = 0; 261 } 262 if (trypid >= pidchecked) { 263 int doingzomb = 0; 264 265 pidchecked = PID_MAX; 266 /* 267 * Scan the active and zombie procs to check whether this pid 268 * is in use. Remember the lowest pid that's greater 269 * than trypid, so we can avoid checking for a while. 270 * 271 * Avoid reuse of the process group id, session id or 272 * the reaper subtree id. Note that for process group 273 * and sessions, the amount of reserved pids is 274 * limited by process limit. For the subtree ids, the 275 * id is kept reserved only while there is a 276 * non-reaped process in the subtree, so amount of 277 * reserved pids is limited by process limit times 278 * two. 279 */ 280 p = LIST_FIRST(&allproc); 281again: 282 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 283 while (p->p_pid == trypid || 284 p->p_reapsubtree == trypid || 285 (p->p_pgrp != NULL && 286 (p->p_pgrp->pg_id == trypid || 287 (p->p_session != NULL && 288 p->p_session->s_sid == trypid)))) { 289 trypid++; 290 if (trypid >= pidchecked) 291 goto retry; 292 } 293 if (p->p_pid > trypid && pidchecked > p->p_pid) 294 pidchecked = p->p_pid; 295 if (p->p_pgrp != NULL) { 296 if (p->p_pgrp->pg_id > trypid && 297 pidchecked > p->p_pgrp->pg_id) 298 pidchecked = p->p_pgrp->pg_id; 299 if (p->p_session != NULL && 300 p->p_session->s_sid > trypid && 301 pidchecked > p->p_session->s_sid) 302 pidchecked = p->p_session->s_sid; 303 } 304 } 305 if (!doingzomb) { 306 doingzomb = 1; 307 p = LIST_FIRST(&zombproc); 308 goto again; 309 } 310 } 311 312 /* 313 * RFHIGHPID does not mess with the lastpid counter during boot. 314 */ 315 if (flags & RFHIGHPID) 316 pidchecked = 0; 317 else 318 lastpid = trypid; 319 320 return (trypid); 321} 322 323static int 324fork_norfproc(struct thread *td, int flags) 325{ 326 int error; 327 struct proc *p1; 328 329 KASSERT((flags & RFPROC) == 0, 330 ("fork_norfproc called with RFPROC set")); 331 p1 = td->td_proc; 332 333 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 334 (flags & (RFCFDG | RFFDG))) { 335 PROC_LOCK(p1); 336 if (thread_single(p1, SINGLE_BOUNDARY)) { 337 PROC_UNLOCK(p1); 338 return (ERESTART); 339 } 340 PROC_UNLOCK(p1); 341 } 342 343 error = vm_forkproc(td, NULL, NULL, NULL, flags); 344 if (error) 345 goto fail; 346 347 /* 348 * Close all file descriptors. 349 */ 350 if (flags & RFCFDG) { 351 struct filedesc *fdtmp; 352 fdtmp = fdinit(td->td_proc->p_fd); 353 fdescfree(td); 354 p1->p_fd = fdtmp; 355 } 356 357 /* 358 * Unshare file descriptors (from parent). 359 */ 360 if (flags & RFFDG) 361 fdunshare(td); 362 363fail: 364 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 365 (flags & (RFCFDG | RFFDG))) { 366 PROC_LOCK(p1); 367 thread_single_end(p1, SINGLE_BOUNDARY); 368 PROC_UNLOCK(p1); 369 } 370 return (error); 371} 372 373static void 374do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2, 375 struct vmspace *vm2, int pdflags) 376{ 377 struct proc *p1, *pptr; 378 int p2_held, trypid; 379 struct filedesc *fd; 380 struct filedesc_to_leader *fdtol; 381 struct sigacts *newsigacts; 382 383 sx_assert(&proctree_lock, SX_SLOCKED); 384 sx_assert(&allproc_lock, SX_XLOCKED); 385 386 p2_held = 0; 387 p1 = td->td_proc; 388 389 trypid = fork_findpid(flags); 390 391 sx_sunlock(&proctree_lock); 392 393 p2->p_state = PRS_NEW; /* protect against others */ 394 p2->p_pid = trypid; 395 AUDIT_ARG_PID(p2->p_pid); 396 LIST_INSERT_HEAD(&allproc, p2, p_list); 397 allproc_gen++; 398 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 399 tidhash_add(td2); 400 PROC_LOCK(p2); 401 PROC_LOCK(p1); 402 403 sx_xunlock(&allproc_lock); 404 405 bcopy(&p1->p_startcopy, &p2->p_startcopy, 406 __rangeof(struct proc, p_startcopy, p_endcopy)); 407 pargs_hold(p2->p_args); 408 PROC_UNLOCK(p1); 409 410 bzero(&p2->p_startzero, 411 __rangeof(struct proc, p_startzero, p_endzero)); 412 p2->p_treeflag = 0; 413 p2->p_filemon = NULL; 414 p2->p_ptevents = 0; 415 416 /* Tell the prison that we exist. */ 417 prison_proc_hold(p2->p_ucred->cr_prison); 418 419 PROC_UNLOCK(p2); 420 421 /* 422 * Malloc things while we don't hold any locks. 423 */ 424 if (flags & RFSIGSHARE) 425 newsigacts = NULL; 426 else 427 newsigacts = sigacts_alloc(); 428 429 /* 430 * Copy filedesc. 431 */ 432 if (flags & RFCFDG) { 433 fd = fdinit(p1->p_fd); 434 fdtol = NULL; 435 } else if (flags & RFFDG) { 436 fd = fdcopy(p1->p_fd); 437 fdtol = NULL; 438 } else { 439 fd = fdshare(p1->p_fd); 440 if (p1->p_fdtol == NULL) 441 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 442 p1->p_leader); 443 if ((flags & RFTHREAD) != 0) { 444 /* 445 * Shared file descriptor table, and shared 446 * process leaders. 447 */ 448 fdtol = p1->p_fdtol; 449 FILEDESC_XLOCK(p1->p_fd); 450 fdtol->fdl_refcount++; 451 FILEDESC_XUNLOCK(p1->p_fd); 452 } else { 453 /* 454 * Shared file descriptor table, and different 455 * process leaders. 456 */ 457 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 458 p1->p_fd, p2); 459 } 460 } 461 /* 462 * Make a proc table entry for the new process. 463 * Start by zeroing the section of proc that is zero-initialized, 464 * then copy the section that is copied directly from the parent. 465 */ 466 467 PROC_LOCK(p2); 468 PROC_LOCK(p1); 469 470 bzero(&td2->td_startzero, 471 __rangeof(struct thread, td_startzero, td_endzero)); 472 td2->td_su = NULL; 473 td2->td_sleeptimo = 0; 474 475 bcopy(&td->td_startcopy, &td2->td_startcopy, 476 __rangeof(struct thread, td_startcopy, td_endcopy)); 477 478 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 479 td2->td_sigstk = td->td_sigstk; 480 td2->td_flags = TDF_INMEM; 481 td2->td_lend_user_pri = PRI_MAX; 482 td2->td_dbg_sc_code = td->td_dbg_sc_code; 483 td2->td_dbg_sc_narg = td->td_dbg_sc_narg; 484 485#ifdef VIMAGE 486 td2->td_vnet = NULL; 487 td2->td_vnet_lpush = NULL; 488#endif 489 490 /* 491 * Allow the scheduler to initialize the child. 492 */ 493 thread_lock(td); 494 sched_fork(td, td2); 495 thread_unlock(td); 496 497 /* 498 * Duplicate sub-structures as needed. 499 * Increase reference counts on shared objects. 500 */ 501 p2->p_flag = P_INMEM; 502 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC); 503 p2->p_swtick = ticks; 504 if (p1->p_flag & P_PROFIL) 505 startprofclock(p2); 506 td2->td_ucred = crhold(p2->p_ucred); 507 508 if (flags & RFSIGSHARE) { 509 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 510 } else { 511 sigacts_copy(newsigacts, p1->p_sigacts); 512 p2->p_sigacts = newsigacts; 513 } 514 515 if (flags & RFTSIGZMB) 516 p2->p_sigparent = RFTSIGNUM(flags); 517 else if (flags & RFLINUXTHPN) 518 p2->p_sigparent = SIGUSR1; 519 else 520 p2->p_sigparent = SIGCHLD; 521 522 p2->p_textvp = p1->p_textvp; 523 p2->p_fd = fd; 524 p2->p_fdtol = fdtol; 525 526 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 527 p2->p_flag |= P_PROTECTED; 528 p2->p_flag2 |= P2_INHERIT_PROTECTED; 529 } 530 531 /* 532 * p_limit is copy-on-write. Bump its refcount. 533 */ 534 lim_fork(p1, p2); 535 536 pstats_fork(p1->p_stats, p2->p_stats); 537 538 PROC_UNLOCK(p1); 539 PROC_UNLOCK(p2); 540 541 /* Bump references to the text vnode (for procfs). */ 542 if (p2->p_textvp) 543 vref(p2->p_textvp); 544 545 /* 546 * Set up linkage for kernel based threading. 547 */ 548 if ((flags & RFTHREAD) != 0) { 549 mtx_lock(&ppeers_lock); 550 p2->p_peers = p1->p_peers; 551 p1->p_peers = p2; 552 p2->p_leader = p1->p_leader; 553 mtx_unlock(&ppeers_lock); 554 PROC_LOCK(p1->p_leader); 555 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 556 PROC_UNLOCK(p1->p_leader); 557 /* 558 * The task leader is exiting, so process p1 is 559 * going to be killed shortly. Since p1 obviously 560 * isn't dead yet, we know that the leader is either 561 * sending SIGKILL's to all the processes in this 562 * task or is sleeping waiting for all the peers to 563 * exit. We let p1 complete the fork, but we need 564 * to go ahead and kill the new process p2 since 565 * the task leader may not get a chance to send 566 * SIGKILL to it. We leave it on the list so that 567 * the task leader will wait for this new process 568 * to commit suicide. 569 */ 570 PROC_LOCK(p2); 571 kern_psignal(p2, SIGKILL); 572 PROC_UNLOCK(p2); 573 } else 574 PROC_UNLOCK(p1->p_leader); 575 } else { 576 p2->p_peers = NULL; 577 p2->p_leader = p2; 578 } 579 580 sx_xlock(&proctree_lock); 581 PGRP_LOCK(p1->p_pgrp); 582 PROC_LOCK(p2); 583 PROC_LOCK(p1); 584 585 /* 586 * Preserve some more flags in subprocess. P_PROFIL has already 587 * been preserved. 588 */ 589 p2->p_flag |= p1->p_flag & P_SUGID; 590 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 591 SESS_LOCK(p1->p_session); 592 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 593 p2->p_flag |= P_CONTROLT; 594 SESS_UNLOCK(p1->p_session); 595 if (flags & RFPPWAIT) 596 p2->p_flag |= P_PPWAIT; 597 598 p2->p_pgrp = p1->p_pgrp; 599 LIST_INSERT_AFTER(p1, p2, p_pglist); 600 PGRP_UNLOCK(p1->p_pgrp); 601 LIST_INIT(&p2->p_children); 602 LIST_INIT(&p2->p_orphans); 603 604 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 605 606 /* 607 * If PF_FORK is set, the child process inherits the 608 * procfs ioctl flags from its parent. 609 */ 610 if (p1->p_pfsflags & PF_FORK) { 611 p2->p_stops = p1->p_stops; 612 p2->p_pfsflags = p1->p_pfsflags; 613 } 614 615 /* 616 * This begins the section where we must prevent the parent 617 * from being swapped. 618 */ 619 _PHOLD(p1); 620 PROC_UNLOCK(p1); 621 622 /* 623 * Attach the new process to its parent. 624 * 625 * If RFNOWAIT is set, the newly created process becomes a child 626 * of init. This effectively disassociates the child from the 627 * parent. 628 */ 629 if ((flags & RFNOWAIT) != 0) { 630 pptr = p1->p_reaper; 631 p2->p_reaper = pptr; 632 } else { 633 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 634 p1 : p1->p_reaper; 635 pptr = p1; 636 } 637 p2->p_pptr = pptr; 638 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 639 LIST_INIT(&p2->p_reaplist); 640 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 641 if (p2->p_reaper == p1) 642 p2->p_reapsubtree = p2->p_pid; 643 else 644 p2->p_reapsubtree = p1->p_reapsubtree; 645 sx_xunlock(&proctree_lock); 646 647 /* Inform accounting that we have forked. */ 648 p2->p_acflag = AFORK; 649 PROC_UNLOCK(p2); 650 651#ifdef KTRACE 652 ktrprocfork(p1, p2); 653#endif 654 655 /* 656 * Finish creating the child process. It will return via a different 657 * execution path later. (ie: directly into user mode) 658 */ 659 vm_forkproc(td, p2, td2, vm2, flags); 660 661 if (flags == (RFFDG | RFPROC)) { 662 PCPU_INC(cnt.v_forks); 663 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 664 p2->p_vmspace->vm_ssize); 665 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 666 PCPU_INC(cnt.v_vforks); 667 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 668 p2->p_vmspace->vm_ssize); 669 } else if (p1 == &proc0) { 670 PCPU_INC(cnt.v_kthreads); 671 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 672 p2->p_vmspace->vm_ssize); 673 } else { 674 PCPU_INC(cnt.v_rforks); 675 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 676 p2->p_vmspace->vm_ssize); 677 } 678 679#ifdef PROCDESC 680 /* 681 * Associate the process descriptor with the process before anything 682 * can happen that might cause that process to need the descriptor. 683 * However, don't do this until after fork(2) can no longer fail. 684 */ 685 if (flags & RFPROCDESC) 686 procdesc_new(p2, pdflags); 687#endif 688 689 /* 690 * Both processes are set up, now check if any loadable modules want 691 * to adjust anything. 692 */ 693 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 694 695 /* 696 * Set the child start time and mark the process as being complete. 697 */ 698 PROC_LOCK(p2); 699 PROC_LOCK(p1); 700 microuptime(&p2->p_stats->p_start); 701 PROC_SLOCK(p2); 702 p2->p_state = PRS_NORMAL; 703 PROC_SUNLOCK(p2); 704 705#ifdef KDTRACE_HOOKS 706 /* 707 * Tell the DTrace fasttrap provider about the new process so that any 708 * tracepoints inherited from the parent can be removed. We have to do 709 * this only after p_state is PRS_NORMAL since the fasttrap module will 710 * use pfind() later on. 711 */ 712 if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork) 713 dtrace_fasttrap_fork(p1, p2); 714#endif 715 if (p1->p_ptevents & PTRACE_FORK) { 716 /* 717 * Arrange for debugger to receive the fork event. 718 * 719 * We can report PL_FLAG_FORKED regardless of 720 * P_FOLLOWFORK settings, but it does not make a sense 721 * for runaway child. 722 */ 723 td->td_dbgflags |= TDB_FORK; 724 td->td_dbg_forked = p2->p_pid; 725 td2->td_dbgflags |= TDB_STOPATFORK; 726 _PHOLD(p2); 727 p2_held = 1; 728 } 729 if (flags & RFPPWAIT) { 730 td->td_pflags |= TDP_RFPPWAIT; 731 td->td_rfppwait_p = p2; 732 td->td_dbgflags |= TDB_VFORK; 733 } 734 PROC_UNLOCK(p2); 735 if ((flags & RFSTOPPED) == 0) { 736 /* 737 * If RFSTOPPED not requested, make child runnable and 738 * add to run queue. 739 */ 740 thread_lock(td2); 741 TD_SET_CAN_RUN(td2); 742 sched_add(td2, SRQ_BORING); 743 thread_unlock(td2); 744 } 745 746 /* 747 * Now can be swapped. 748 */ 749 _PRELE(p1); 750 PROC_UNLOCK(p1); 751 752 /* 753 * Tell any interested parties about the new process. 754 */ 755 knote_fork(&p1->p_klist, p2->p_pid); 756 SDT_PROBE3(proc, , , create, p2, p1, flags); 757 758 /* 759 * Wait until debugger is attached to child. 760 */ 761 PROC_LOCK(p2); 762 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0) 763 cv_wait(&p2->p_dbgwait, &p2->p_mtx); 764 if (p2_held) 765 _PRELE(p2); 766 PROC_UNLOCK(p2); 767} 768 769int 770fork1(struct thread *td, int flags, int pages, struct proc **procp, 771 int *procdescp, int pdflags) 772{ 773 struct proc *p1, *newproc; 774 struct thread *td2; 775 struct vmspace *vm2; 776#ifdef PROCDESC 777 struct file *fp_procdesc; 778#endif 779 vm_ooffset_t mem_charged; 780 int error, nprocs_new, ok; 781 static int curfail; 782 static struct timeval lastfail; 783 784 /* Check for the undefined or unimplemented flags. */ 785 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 786 return (EINVAL); 787 788 /* Signal value requires RFTSIGZMB. */ 789 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 790 return (EINVAL); 791 792 /* Can't copy and clear. */ 793 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 794 return (EINVAL); 795 796 /* Check the validity of the signal number. */ 797 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 798 return (EINVAL); 799 800#ifdef PROCDESC 801 if ((flags & RFPROCDESC) != 0) { 802 /* Can't not create a process yet get a process descriptor. */ 803 if ((flags & RFPROC) == 0) 804 return (EINVAL); 805 806 /* Must provide a place to put a procdesc if creating one. */ 807 if (procdescp == NULL) 808 return (EINVAL); 809 } 810#endif 811 812 p1 = td->td_proc; 813 814 /* 815 * Here we don't create a new process, but we divorce 816 * certain parts of a process from itself. 817 */ 818 if ((flags & RFPROC) == 0) { 819 *procp = NULL; 820 return (fork_norfproc(td, flags)); 821 } 822 823#ifdef PROCDESC 824 fp_procdesc = NULL; 825#endif 826 newproc = NULL; 827 vm2 = NULL; 828 829 /* 830 * Increment the nprocs resource before allocations occur. 831 * Although process entries are dynamically created, we still 832 * keep a global limit on the maximum number we will 833 * create. There are hard-limits as to the number of processes 834 * that can run, established by the KVA and memory usage for 835 * the process data. 836 * 837 * Don't allow a nonprivileged user to use the last ten 838 * processes; don't let root exceed the limit. 839 */ 840 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 841 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, 842 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { 843 sx_xlock(&allproc_lock); 844 if (ppsratecheck(&lastfail, &curfail, 1)) { 845 printf("maxproc limit exceeded by uid %u (pid %d); " 846 "see tuning(7) and login.conf(5)\n", 847 td->td_ucred->cr_ruid, p1->p_pid); 848 } 849 sx_xunlock(&allproc_lock); 850 error = EAGAIN; 851 goto fail1; 852 } 853 854#ifdef PROCDESC 855 /* 856 * If required, create a process descriptor in the parent first; we 857 * will abandon it if something goes wrong. We don't finit() until 858 * later. 859 */ 860 if (flags & RFPROCDESC) { 861 error = falloc(td, &fp_procdesc, procdescp, 0); 862 if (error != 0) 863 goto fail1; 864 } 865#endif 866 867 mem_charged = 0; 868 if (pages == 0) 869 pages = KSTACK_PAGES; 870 /* Allocate new proc. */ 871 newproc = uma_zalloc(proc_zone, M_WAITOK); 872 td2 = FIRST_THREAD_IN_PROC(newproc); 873 if (td2 == NULL) { 874 td2 = thread_alloc(pages); 875 if (td2 == NULL) { 876 error = ENOMEM; 877 goto fail2; 878 } 879 proc_linkup(newproc, td2); 880 } else { 881 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 882 if (td2->td_kstack != 0) 883 vm_thread_dispose(td2); 884 if (!thread_alloc_stack(td2, pages)) { 885 error = ENOMEM; 886 goto fail2; 887 } 888 } 889 } 890 891 if ((flags & RFMEM) == 0) { 892 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 893 if (vm2 == NULL) { 894 error = ENOMEM; 895 goto fail2; 896 } 897 if (!swap_reserve(mem_charged)) { 898 /* 899 * The swap reservation failed. The accounting 900 * from the entries of the copied vm2 will be 901 * subtracted in vmspace_free(), so force the 902 * reservation there. 903 */ 904 swap_reserve_force(mem_charged); 905 error = ENOMEM; 906 goto fail2; 907 } 908 } else 909 vm2 = NULL; 910 911 /* 912 * XXX: This is ugly; when we copy resource usage, we need to bump 913 * per-cred resource counters. 914 */ 915 proc_set_cred_init(newproc, crhold(td->td_ucred)); 916 917 /* 918 * Initialize resource accounting for the child process. 919 */ 920 error = racct_proc_fork(p1, newproc); 921 if (error != 0) { 922 error = EAGAIN; 923 goto fail1; 924 } 925 926#ifdef MAC 927 mac_proc_init(newproc); 928#endif 929 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); 930 STAILQ_INIT(&newproc->p_ktr); 931 932 /* We have to lock the process tree while we look for a pid. */ 933 sx_slock(&proctree_lock); 934 sx_xlock(&allproc_lock); 935 936 /* 937 * Increment the count of procs running with this uid. Don't allow 938 * a nonprivileged user to exceed their current limit. 939 * 940 * XXXRW: Can we avoid privilege here if it's not needed? 941 */ 942 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 943 if (error == 0) 944 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 945 else { 946 PROC_LOCK(p1); 947 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 948 lim_cur(p1, RLIMIT_NPROC)); 949 PROC_UNLOCK(p1); 950 } 951 if (ok) { 952 do_fork(td, flags, newproc, td2, vm2, pdflags); 953 954 /* 955 * Return child proc pointer to parent. 956 */ 957 *procp = newproc; 958#ifdef PROCDESC 959 if (flags & RFPROCDESC) { 960 procdesc_finit(newproc->p_procdesc, fp_procdesc); 961 fdrop(fp_procdesc, td); 962 } 963#endif 964 racct_proc_fork_done(newproc); 965 return (0); 966 } 967 968 error = EAGAIN; 969 sx_sunlock(&proctree_lock); 970 sx_xunlock(&allproc_lock); 971#ifdef MAC 972 mac_proc_destroy(newproc); 973#endif 974 racct_proc_exit(newproc); 975fail1: 976 crfree(newproc->p_ucred); 977 newproc->p_ucred = NULL; 978fail2: 979 if (vm2 != NULL) 980 vmspace_free(vm2); 981 uma_zfree(proc_zone, newproc); 982#ifdef PROCDESC 983 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 984 fdclose(td->td_proc->p_fd, fp_procdesc, *procdescp, td); 985 fdrop(fp_procdesc, td); 986 } 987#endif 988 atomic_add_int(&nprocs, -1); 989 pause("fork", hz / 2); 990 return (error); 991} 992 993/* 994 * Handle the return of a child process from fork1(). This function 995 * is called from the MD fork_trampoline() entry point. 996 */ 997void 998fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 999 struct trapframe *frame) 1000{ 1001 struct proc *p; 1002 struct thread *td; 1003 struct thread *dtd; 1004 1005 td = curthread; 1006 p = td->td_proc; 1007 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1008 1009 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1010 td, td->td_sched, p->p_pid, td->td_name); 1011 1012 sched_fork_exit(td); 1013 /* 1014 * Processes normally resume in mi_switch() after being 1015 * cpu_switch()'ed to, but when children start up they arrive here 1016 * instead, so we must do much the same things as mi_switch() would. 1017 */ 1018 if ((dtd = PCPU_GET(deadthread))) { 1019 PCPU_SET(deadthread, NULL); 1020 thread_stash(dtd); 1021 } 1022 thread_unlock(td); 1023 1024 /* 1025 * cpu_set_fork_handler intercepts this function call to 1026 * have this call a non-return function to stay in kernel mode. 1027 * initproc has its own fork handler, but it does return. 1028 */ 1029 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1030 callout(arg, frame); 1031 1032 /* 1033 * Check if a kernel thread misbehaved and returned from its main 1034 * function. 1035 */ 1036 if (p->p_flag & P_KTHREAD) { 1037 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1038 td->td_name, p->p_pid); 1039 kthread_exit(); 1040 } 1041 mtx_assert(&Giant, MA_NOTOWNED); 1042 1043 if (p->p_sysent->sv_schedtail != NULL) 1044 (p->p_sysent->sv_schedtail)(td); 1045} 1046 1047/* 1048 * Simplified back end of syscall(), used when returning from fork() 1049 * directly into user mode. This function is passed in to fork_exit() 1050 * as the first parameter and is called when returning to a new 1051 * userland process. 1052 */ 1053void 1054fork_return(struct thread *td, struct trapframe *frame) 1055{ 1056 struct proc *p, *dbg; 1057 1058 p = td->td_proc; 1059 if (td->td_dbgflags & TDB_STOPATFORK) { 1060 sx_xlock(&proctree_lock); 1061 PROC_LOCK(p); 1062 if (p->p_pptr->p_ptevents & PTRACE_FORK) { 1063 /* 1064 * If debugger still wants auto-attach for the 1065 * parent's children, do it now. 1066 */ 1067 dbg = p->p_pptr->p_pptr; 1068 proc_set_traced(p, true); 1069 CTR2(KTR_PTRACE, 1070 "fork_return: attaching to new child pid %d: oppid %d", 1071 p->p_pid, p->p_oppid); 1072 proc_reparent(p, dbg); 1073 sx_xunlock(&proctree_lock); 1074 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1075 ptracestop(td, SIGSTOP, NULL); 1076 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1077 } else { 1078 /* 1079 * ... otherwise clear the request. 1080 */ 1081 sx_xunlock(&proctree_lock); 1082 td->td_dbgflags &= ~TDB_STOPATFORK; 1083 cv_broadcast(&p->p_dbgwait); 1084 } 1085 PROC_UNLOCK(p); 1086 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1087 /* 1088 * This is the start of a new thread in a traced 1089 * process. Report a system call exit event. 1090 */ 1091 PROC_LOCK(p); 1092 td->td_dbgflags |= TDB_SCX; 1093 _STOPEVENT(p, S_SCX, td->td_dbg_sc_code); 1094 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1095 (td->td_dbgflags & TDB_BORN) != 0) 1096 ptracestop(td, SIGTRAP, NULL); 1097 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1098 PROC_UNLOCK(p); 1099 } 1100 1101 userret(td, frame); 1102 1103#ifdef KTRACE 1104 if (KTRPOINT(td, KTR_SYSRET)) 1105 ktrsysret(SYS_fork, 0, 0); 1106#endif 1107} 1108