1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 37 */ 38 39#include <sys/cdefs.h> 40__FBSDID("$FreeBSD$"); 41 42#include "opt_ktrace.h" 43#include "opt_kstack_pages.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 101EVENTHANDLER_LIST_DECLARE(process_fork); 102 103/* ARGSUSED */ 104int 105sys_fork(struct thread *td, struct fork_args *uap) 106{ 107 struct fork_req fr; 108 int error, pid; 109 110 bzero(&fr, sizeof(fr)); 111 fr.fr_flags = RFFDG | RFPROC; 112 fr.fr_pidp = &pid; 113 error = fork1(td, &fr); 114 if (error == 0) { 115 td->td_retval[0] = pid; 116 td->td_retval[1] = 0; 117 } 118 return (error); 119} 120 121/* ARGUSED */ 122int 123sys_pdfork(struct thread *td, struct pdfork_args *uap) 124{ 125 struct fork_req fr; 126 int error, fd, pid; 127 128 bzero(&fr, sizeof(fr)); 129 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC; 130 fr.fr_pidp = &pid; 131 fr.fr_pd_fd = &fd; 132 fr.fr_pd_flags = uap->flags; 133 /* 134 * It is necessary to return fd by reference because 0 is a valid file 135 * descriptor number, and the child needs to be able to distinguish 136 * itself from the parent using the return value. 137 */ 138 error = fork1(td, &fr); 139 if (error == 0) { 140 td->td_retval[0] = pid; 141 td->td_retval[1] = 0; 142 error = copyout(&fd, uap->fdp, sizeof(fd)); 143 } 144 return (error); 145} 146 147/* ARGSUSED */ 148int 149sys_vfork(struct thread *td, struct vfork_args *uap) 150{ 151 struct fork_req fr; 152 int error, pid; 153 154 bzero(&fr, sizeof(fr)); 155 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 156 fr.fr_pidp = &pid; 157 error = fork1(td, &fr); 158 if (error == 0) { 159 td->td_retval[0] = pid; 160 td->td_retval[1] = 0; 161 } 162 return (error); 163} 164 165int 166sys_rfork(struct thread *td, struct rfork_args *uap) 167{ 168 struct fork_req fr; 169 int error, pid; 170 171 /* Don't allow kernel-only flags. */ 172 if ((uap->flags & RFKERNELONLY) != 0) 173 return (EINVAL); 174 /* RFSPAWN must not appear with others */ 175 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN) 176 return (EINVAL); 177 178 AUDIT_ARG_FFLAGS(uap->flags); 179 bzero(&fr, sizeof(fr)); 180 if ((uap->flags & RFSPAWN) != 0) { 181 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 182 fr.fr_flags2 = FR2_DROPSIG_CAUGHT; 183 } else { 184 fr.fr_flags = uap->flags; 185 } 186 fr.fr_pidp = &pid; 187 error = fork1(td, &fr); 188 if (error == 0) { 189 td->td_retval[0] = pid; 190 td->td_retval[1] = 0; 191 } 192 return (error); 193} 194 195int nprocs = 1; /* process 0 */ 196int lastpid = 0; 197SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 198 "Last used PID"); 199 200/* 201 * Random component to lastpid generation. We mix in a random factor to make 202 * it a little harder to predict. We sanity check the modulus value to avoid 203 * doing it in critical paths. Don't let it be too small or we pointlessly 204 * waste randomness entropy, and don't let it be impossibly large. Using a 205 * modulus that is too big causes a LOT more process table scans and slows 206 * down fork processing as the pidchecked caching is defeated. 207 */ 208static int randompid = 0; 209 210static int 211sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 212{ 213 int error, pid; 214 215 error = sysctl_wire_old_buffer(req, sizeof(int)); 216 if (error != 0) 217 return(error); 218 sx_xlock(&allproc_lock); 219 pid = randompid; 220 error = sysctl_handle_int(oidp, &pid, 0, req); 221 if (error == 0 && req->newptr != NULL) { 222 if (pid == 0) 223 randompid = 0; 224 else if (pid == 1) 225 /* generate a random PID modulus between 100 and 1123 */ 226 randompid = 100 + arc4random() % 1024; 227 else if (pid < 0 || pid > pid_max - 100) 228 /* out of range */ 229 randompid = pid_max - 100; 230 else if (pid < 100) 231 /* Make it reasonable */ 232 randompid = 100; 233 else 234 randompid = pid; 235 } 236 sx_xunlock(&allproc_lock); 237 return (error); 238} 239 240SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 241 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value"); 242 243static int 244fork_findpid(int flags) 245{ 246 struct proc *p; 247 int trypid; 248 static int pidchecked = 0; 249 250 /* 251 * Requires allproc_lock in order to iterate over the list 252 * of processes, and proctree_lock to access p_pgrp. 253 */ 254 sx_assert(&allproc_lock, SX_LOCKED); 255 sx_assert(&proctree_lock, SX_LOCKED); 256 257 /* 258 * Find an unused process ID. We remember a range of unused IDs 259 * ready to use (from lastpid+1 through pidchecked-1). 260 * 261 * If RFHIGHPID is set (used during system boot), do not allocate 262 * low-numbered pids. 263 */ 264 trypid = lastpid + 1; 265 if (flags & RFHIGHPID) { 266 if (trypid < 10) 267 trypid = 10; 268 } else { 269 if (randompid) 270 trypid += arc4random() % randompid; 271 } 272retry: 273 /* 274 * If the process ID prototype has wrapped around, 275 * restart somewhat above 0, as the low-numbered procs 276 * tend to include daemons that don't exit. 277 */ 278 if (trypid >= pid_max) { 279 trypid = trypid % pid_max; 280 if (trypid < 100) 281 trypid += 100; 282 pidchecked = 0; 283 } 284 if (trypid >= pidchecked) { 285 int doingzomb = 0; 286 287 pidchecked = PID_MAX; 288 /* 289 * Scan the active and zombie procs to check whether this pid 290 * is in use. Remember the lowest pid that's greater 291 * than trypid, so we can avoid checking for a while. 292 * 293 * Avoid reuse of the process group id, session id or 294 * the reaper subtree id. Note that for process group 295 * and sessions, the amount of reserved pids is 296 * limited by process limit. For the subtree ids, the 297 * id is kept reserved only while there is a 298 * non-reaped process in the subtree, so amount of 299 * reserved pids is limited by process limit times 300 * two. 301 */ 302 p = LIST_FIRST(&allproc); 303again: 304 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 305 while (p->p_pid == trypid || 306 p->p_reapsubtree == trypid || 307 (p->p_pgrp != NULL && 308 (p->p_pgrp->pg_id == trypid || 309 (p->p_session != NULL && 310 p->p_session->s_sid == trypid)))) { 311 trypid++; 312 if (trypid >= pidchecked) 313 goto retry; 314 } 315 if (p->p_pid > trypid && pidchecked > p->p_pid) 316 pidchecked = p->p_pid; 317 if (p->p_pgrp != NULL) { 318 if (p->p_pgrp->pg_id > trypid && 319 pidchecked > p->p_pgrp->pg_id) 320 pidchecked = p->p_pgrp->pg_id; 321 if (p->p_session != NULL && 322 p->p_session->s_sid > trypid && 323 pidchecked > p->p_session->s_sid) 324 pidchecked = p->p_session->s_sid; 325 } 326 } 327 if (!doingzomb) { 328 doingzomb = 1; 329 p = LIST_FIRST(&zombproc); 330 goto again; 331 } 332 } 333 334 /* 335 * RFHIGHPID does not mess with the lastpid counter during boot. 336 */ 337 if (flags & RFHIGHPID) 338 pidchecked = 0; 339 else 340 lastpid = trypid; 341 342 return (trypid); 343} 344 345static int 346fork_norfproc(struct thread *td, int flags) 347{ 348 int error; 349 struct proc *p1; 350 351 KASSERT((flags & RFPROC) == 0, 352 ("fork_norfproc called with RFPROC set")); 353 p1 = td->td_proc; 354 355 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 356 (flags & (RFCFDG | RFFDG))) { 357 PROC_LOCK(p1); 358 if (thread_single(p1, SINGLE_BOUNDARY)) { 359 PROC_UNLOCK(p1); 360 return (ERESTART); 361 } 362 PROC_UNLOCK(p1); 363 } 364 365 error = vm_forkproc(td, NULL, NULL, NULL, flags); 366 if (error) 367 goto fail; 368 369 /* 370 * Close all file descriptors. 371 */ 372 if (flags & RFCFDG) { 373 struct filedesc *fdtmp; 374 fdtmp = fdinit(td->td_proc->p_fd, false); 375 fdescfree(td); 376 p1->p_fd = fdtmp; 377 } 378 379 /* 380 * Unshare file descriptors (from parent). 381 */ 382 if (flags & RFFDG) 383 fdunshare(td); 384 385fail: 386 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 387 (flags & (RFCFDG | RFFDG))) { 388 PROC_LOCK(p1); 389 thread_single_end(p1, SINGLE_BOUNDARY); 390 PROC_UNLOCK(p1); 391 } 392 return (error); 393} 394 395static void 396do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 397 struct vmspace *vm2, struct file *fp_procdesc) 398{ 399 struct proc *p1, *pptr; 400 int trypid; 401 struct filedesc *fd; 402 struct filedesc_to_leader *fdtol; 403 struct sigacts *newsigacts; 404 405 sx_assert(&proctree_lock, SX_LOCKED); 406 sx_assert(&allproc_lock, SX_XLOCKED); 407 408 p1 = td->td_proc; 409 410 trypid = fork_findpid(fr->fr_flags); 411 412 p2->p_state = PRS_NEW; /* protect against others */ 413 p2->p_pid = trypid; 414 AUDIT_ARG_PID(p2->p_pid); 415 LIST_INSERT_HEAD(&allproc, p2, p_list); 416 allproc_gen++; 417 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 418 PROC_LOCK(p2); 419 PROC_LOCK(p1); 420 421 sx_xunlock(&allproc_lock); 422 sx_xunlock(&proctree_lock); 423 424 bcopy(&p1->p_startcopy, &p2->p_startcopy, 425 __rangeof(struct proc, p_startcopy, p_endcopy)); 426 p2->p_fctl0 = p1->p_fctl0; 427 pargs_hold(p2->p_args); 428 429 PROC_UNLOCK(p1); 430 431 bzero(&p2->p_startzero, 432 __rangeof(struct proc, p_startzero, p_endzero)); 433 434 /* Tell the prison that we exist. */ 435 prison_proc_hold(p2->p_ucred->cr_prison); 436 437 PROC_UNLOCK(p2); 438 439 tidhash_add(td2); 440 441 /* 442 * Malloc things while we don't hold any locks. 443 */ 444 if (fr->fr_flags & RFSIGSHARE) 445 newsigacts = NULL; 446 else 447 newsigacts = sigacts_alloc(); 448 449 /* 450 * Copy filedesc. 451 */ 452 if (fr->fr_flags & RFCFDG) { 453 fd = fdinit(p1->p_fd, false); 454 fdtol = NULL; 455 } else if (fr->fr_flags & RFFDG) { 456 fd = fdcopy(p1->p_fd); 457 fdtol = NULL; 458 } else { 459 fd = fdshare(p1->p_fd); 460 if (p1->p_fdtol == NULL) 461 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 462 p1->p_leader); 463 if ((fr->fr_flags & RFTHREAD) != 0) { 464 /* 465 * Shared file descriptor table, and shared 466 * process leaders. 467 */ 468 fdtol = p1->p_fdtol; 469 FILEDESC_XLOCK(p1->p_fd); 470 fdtol->fdl_refcount++; 471 FILEDESC_XUNLOCK(p1->p_fd); 472 } else { 473 /* 474 * Shared file descriptor table, and different 475 * process leaders. 476 */ 477 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 478 p1->p_fd, p2); 479 } 480 } 481 /* 482 * Make a proc table entry for the new process. 483 * Start by zeroing the section of proc that is zero-initialized, 484 * then copy the section that is copied directly from the parent. 485 */ 486 487 PROC_LOCK(p2); 488 PROC_LOCK(p1); 489 490 bzero(&td2->td_startzero, 491 __rangeof(struct thread, td_startzero, td_endzero)); 492 td2->td_pflags2 = 0; 493 td2->td_errno = 0; 494 495 bcopy(&td->td_startcopy, &td2->td_startcopy, 496 __rangeof(struct thread, td_startcopy, td_endcopy)); 497 498 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 499 td2->td_sigstk = td->td_sigstk; 500 td2->td_flags = TDF_INMEM; 501 td2->td_lend_user_pri = PRI_MAX; 502 503#ifdef VIMAGE 504 td2->td_vnet = NULL; 505 td2->td_vnet_lpush = NULL; 506#endif 507 508 /* 509 * Allow the scheduler to initialize the child. 510 */ 511 thread_lock(td); 512 sched_fork(td, td2); 513 thread_unlock(td); 514 515 /* 516 * Duplicate sub-structures as needed. 517 * Increase reference counts on shared objects. 518 */ 519 p2->p_flag = P_INMEM; 520 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 521 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 522 P2_TRAPCAP | 523 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC); 524 p2->p_swtick = ticks; 525 if (p1->p_flag & P_PROFIL) 526 startprofclock(p2); 527 528 if (fr->fr_flags & RFSIGSHARE) { 529 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 530 } else { 531 sigacts_copy(newsigacts, p1->p_sigacts); 532 p2->p_sigacts = newsigacts; 533 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) { 534 mtx_lock(&p2->p_sigacts->ps_mtx); 535 sig_drop_caught(p2); 536 mtx_unlock(&p2->p_sigacts->ps_mtx); 537 } 538 } 539 540 if (fr->fr_flags & RFTSIGZMB) 541 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 542 else if (fr->fr_flags & RFLINUXTHPN) 543 p2->p_sigparent = SIGUSR1; 544 else 545 p2->p_sigparent = SIGCHLD; 546 547 p2->p_textvp = p1->p_textvp; 548 p2->p_fd = fd; 549 p2->p_fdtol = fdtol; 550 551 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 552 p2->p_flag |= P_PROTECTED; 553 p2->p_flag2 |= P2_INHERIT_PROTECTED; 554 } 555 556 /* 557 * p_limit is copy-on-write. Bump its refcount. 558 */ 559 lim_fork(p1, p2); 560 561 thread_cow_get_proc(td2, p2); 562 563 pstats_fork(p1->p_stats, p2->p_stats); 564 565 PROC_UNLOCK(p1); 566 PROC_UNLOCK(p2); 567 568 /* Bump references to the text vnode (for procfs). */ 569 if (p2->p_textvp) 570 vrefact(p2->p_textvp); 571 572 /* 573 * Set up linkage for kernel based threading. 574 */ 575 if ((fr->fr_flags & RFTHREAD) != 0) { 576 mtx_lock(&ppeers_lock); 577 p2->p_peers = p1->p_peers; 578 p1->p_peers = p2; 579 p2->p_leader = p1->p_leader; 580 mtx_unlock(&ppeers_lock); 581 PROC_LOCK(p1->p_leader); 582 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 583 PROC_UNLOCK(p1->p_leader); 584 /* 585 * The task leader is exiting, so process p1 is 586 * going to be killed shortly. Since p1 obviously 587 * isn't dead yet, we know that the leader is either 588 * sending SIGKILL's to all the processes in this 589 * task or is sleeping waiting for all the peers to 590 * exit. We let p1 complete the fork, but we need 591 * to go ahead and kill the new process p2 since 592 * the task leader may not get a chance to send 593 * SIGKILL to it. We leave it on the list so that 594 * the task leader will wait for this new process 595 * to commit suicide. 596 */ 597 PROC_LOCK(p2); 598 kern_psignal(p2, SIGKILL); 599 PROC_UNLOCK(p2); 600 } else 601 PROC_UNLOCK(p1->p_leader); 602 } else { 603 p2->p_peers = NULL; 604 p2->p_leader = p2; 605 } 606 607 sx_xlock(&proctree_lock); 608 PGRP_LOCK(p1->p_pgrp); 609 PROC_LOCK(p2); 610 PROC_LOCK(p1); 611 612 /* 613 * Preserve some more flags in subprocess. P_PROFIL has already 614 * been preserved. 615 */ 616 p2->p_flag |= p1->p_flag & P_SUGID; 617 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 618 SESS_LOCK(p1->p_session); 619 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 620 p2->p_flag |= P_CONTROLT; 621 SESS_UNLOCK(p1->p_session); 622 if (fr->fr_flags & RFPPWAIT) 623 p2->p_flag |= P_PPWAIT; 624 625 p2->p_pgrp = p1->p_pgrp; 626 LIST_INSERT_AFTER(p1, p2, p_pglist); 627 PGRP_UNLOCK(p1->p_pgrp); 628 LIST_INIT(&p2->p_children); 629 LIST_INIT(&p2->p_orphans); 630 631 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 632 633 /* 634 * If PF_FORK is set, the child process inherits the 635 * procfs ioctl flags from its parent. 636 */ 637 if (p1->p_pfsflags & PF_FORK) { 638 p2->p_stops = p1->p_stops; 639 p2->p_pfsflags = p1->p_pfsflags; 640 } 641 642 /* 643 * This begins the section where we must prevent the parent 644 * from being swapped. 645 */ 646 _PHOLD(p1); 647 PROC_UNLOCK(p1); 648 649 /* 650 * Attach the new process to its parent. 651 * 652 * If RFNOWAIT is set, the newly created process becomes a child 653 * of init. This effectively disassociates the child from the 654 * parent. 655 */ 656 if ((fr->fr_flags & RFNOWAIT) != 0) { 657 pptr = p1->p_reaper; 658 p2->p_reaper = pptr; 659 } else { 660 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 661 p1 : p1->p_reaper; 662 pptr = p1; 663 } 664 p2->p_pptr = pptr; 665 p2->p_oppid = pptr->p_pid; 666 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 667 LIST_INIT(&p2->p_reaplist); 668 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 669 if (p2->p_reaper == p1) 670 p2->p_reapsubtree = p2->p_pid; 671 sx_xunlock(&proctree_lock); 672 673 /* Inform accounting that we have forked. */ 674 p2->p_acflag = AFORK; 675 PROC_UNLOCK(p2); 676 677#ifdef KTRACE 678 ktrprocfork(p1, p2); 679#endif 680 681 /* 682 * Finish creating the child process. It will return via a different 683 * execution path later. (ie: directly into user mode) 684 */ 685 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 686 687 if (fr->fr_flags == (RFFDG | RFPROC)) { 688 VM_CNT_INC(v_forks); 689 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 690 p2->p_vmspace->vm_ssize); 691 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 692 VM_CNT_INC(v_vforks); 693 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 694 p2->p_vmspace->vm_ssize); 695 } else if (p1 == &proc0) { 696 VM_CNT_INC(v_kthreads); 697 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 698 p2->p_vmspace->vm_ssize); 699 } else { 700 VM_CNT_INC(v_rforks); 701 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 702 p2->p_vmspace->vm_ssize); 703 } 704 705 /* 706 * Associate the process descriptor with the process before anything 707 * can happen that might cause that process to need the descriptor. 708 * However, don't do this until after fork(2) can no longer fail. 709 */ 710 if (fr->fr_flags & RFPROCDESC) 711 procdesc_new(p2, fr->fr_pd_flags); 712 713 /* 714 * Both processes are set up, now check if any loadable modules want 715 * to adjust anything. 716 */ 717 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 718 719 /* 720 * Set the child start time and mark the process as being complete. 721 */ 722 PROC_LOCK(p2); 723 PROC_LOCK(p1); 724 microuptime(&p2->p_stats->p_start); 725 PROC_SLOCK(p2); 726 p2->p_state = PRS_NORMAL; 727 PROC_SUNLOCK(p2); 728 729#ifdef KDTRACE_HOOKS 730 /* 731 * Tell the DTrace fasttrap provider about the new process so that any 732 * tracepoints inherited from the parent can be removed. We have to do 733 * this only after p_state is PRS_NORMAL since the fasttrap module will 734 * use pfind() later on. 735 */ 736 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 737 dtrace_fasttrap_fork(p1, p2); 738#endif 739 /* 740 * Hold the process so that it cannot exit after we make it runnable, 741 * but before we wait for the debugger. 742 */ 743 _PHOLD(p2); 744 if (fr->fr_flags & RFPPWAIT) { 745 td->td_pflags |= TDP_RFPPWAIT; 746 td->td_rfppwait_p = p2; 747 td->td_dbgflags |= TDB_VFORK; 748 } 749 PROC_UNLOCK(p2); 750 751 /* 752 * Now can be swapped. 753 */ 754 _PRELE(p1); 755 PROC_UNLOCK(p1); 756 757 /* 758 * Tell any interested parties about the new process. 759 */ 760 knote_fork(p1->p_klist, p2->p_pid); 761 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 762 763 if (fr->fr_flags & RFPROCDESC) { 764 procdesc_finit(p2->p_procdesc, fp_procdesc); 765 fdrop(fp_procdesc, td); 766 } 767 768 /* 769 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 770 * synced with forks in progress so it is OK if we miss it 771 * if being set atm. 772 */ 773 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 774 sx_xlock(&proctree_lock); 775 PROC_LOCK(p2); 776 777 /* 778 * p1->p_ptevents & p1->p_pptr are protected by both 779 * process and proctree locks for modifications, 780 * so owning proctree_lock allows the race-free read. 781 */ 782 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 783 /* 784 * Arrange for debugger to receive the fork event. 785 * 786 * We can report PL_FLAG_FORKED regardless of 787 * P_FOLLOWFORK settings, but it does not make a sense 788 * for runaway child. 789 */ 790 td->td_dbgflags |= TDB_FORK; 791 td->td_dbg_forked = p2->p_pid; 792 td2->td_dbgflags |= TDB_STOPATFORK; 793 proc_set_traced(p2, true); 794 CTR2(KTR_PTRACE, 795 "do_fork: attaching to new child pid %d: oppid %d", 796 p2->p_pid, p2->p_oppid); 797 proc_reparent(p2, p1->p_pptr, false); 798 } 799 PROC_UNLOCK(p2); 800 sx_xunlock(&proctree_lock); 801 } 802 803 if ((fr->fr_flags & RFSTOPPED) == 0) { 804 /* 805 * If RFSTOPPED not requested, make child runnable and 806 * add to run queue. 807 */ 808 thread_lock(td2); 809 TD_SET_CAN_RUN(td2); 810 sched_add(td2, SRQ_BORING); 811 thread_unlock(td2); 812 if (fr->fr_pidp != NULL) 813 *fr->fr_pidp = p2->p_pid; 814 } else { 815 *fr->fr_procp = p2; 816 } 817 818 PROC_LOCK(p2); 819 _PRELE(p2); 820 racct_proc_fork_done(p2); 821 PROC_UNLOCK(p2); 822} 823 824int 825fork1(struct thread *td, struct fork_req *fr) 826{ 827 struct proc *p1, *newproc; 828 struct thread *td2; 829 struct vmspace *vm2; 830 struct file *fp_procdesc; 831 vm_ooffset_t mem_charged; 832 int error, nprocs_new, ok; 833 static int curfail; 834 static struct timeval lastfail; 835 int flags, pages; 836 837 flags = fr->fr_flags; 838 pages = fr->fr_pages; 839 840 if ((flags & RFSTOPPED) != 0) 841 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 842 else 843 MPASS(fr->fr_procp == NULL); 844 845 /* Check for the undefined or unimplemented flags. */ 846 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 847 return (EINVAL); 848 849 /* Signal value requires RFTSIGZMB. */ 850 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 851 return (EINVAL); 852 853 /* Can't copy and clear. */ 854 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 855 return (EINVAL); 856 857 /* Check the validity of the signal number. */ 858 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 859 return (EINVAL); 860 861 if ((flags & RFPROCDESC) != 0) { 862 /* Can't not create a process yet get a process descriptor. */ 863 if ((flags & RFPROC) == 0) 864 return (EINVAL); 865 866 /* Must provide a place to put a procdesc if creating one. */ 867 if (fr->fr_pd_fd == NULL) 868 return (EINVAL); 869 870 /* Check if we are using supported flags. */ 871 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 872 return (EINVAL); 873 } 874 875 p1 = td->td_proc; 876 877 /* 878 * Here we don't create a new process, but we divorce 879 * certain parts of a process from itself. 880 */ 881 if ((flags & RFPROC) == 0) { 882 if (fr->fr_procp != NULL) 883 *fr->fr_procp = NULL; 884 else if (fr->fr_pidp != NULL) 885 *fr->fr_pidp = 0; 886 return (fork_norfproc(td, flags)); 887 } 888 889 fp_procdesc = NULL; 890 newproc = NULL; 891 vm2 = NULL; 892 893 /* 894 * Increment the nprocs resource before allocations occur. 895 * Although process entries are dynamically created, we still 896 * keep a global limit on the maximum number we will 897 * create. There are hard-limits as to the number of processes 898 * that can run, established by the KVA and memory usage for 899 * the process data. 900 * 901 * Don't allow a nonprivileged user to use the last ten 902 * processes; don't let root exceed the limit. 903 */ 904 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 905 if (nprocs_new >= maxproc - 10) { 906 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC, 0) != 0 || 907 nprocs_new >= maxproc) { 908 error = EAGAIN; 909 sx_xlock(&allproc_lock); 910 if (ppsratecheck(&lastfail, &curfail, 1)) { 911 printf("maxproc limit exceeded by uid %u " 912 "(pid %d); see tuning(7) and " 913 "login.conf(5)\n", 914 td->td_ucred->cr_ruid, p1->p_pid); 915 } 916 sx_xunlock(&allproc_lock); 917 goto fail2; 918 } 919 } 920 921 /* 922 * If required, create a process descriptor in the parent first; we 923 * will abandon it if something goes wrong. We don't finit() until 924 * later. 925 */ 926 if (flags & RFPROCDESC) { 927 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 928 fr->fr_pd_flags, fr->fr_pd_fcaps); 929 if (error != 0) 930 goto fail2; 931 } 932 933 mem_charged = 0; 934 if (pages == 0) 935 pages = kstack_pages; 936 /* Allocate new proc. */ 937 newproc = uma_zalloc(proc_zone, M_WAITOK); 938 td2 = FIRST_THREAD_IN_PROC(newproc); 939 if (td2 == NULL) { 940 td2 = thread_alloc(pages); 941 if (td2 == NULL) { 942 error = ENOMEM; 943 goto fail2; 944 } 945 proc_linkup(newproc, td2); 946 } else { 947 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 948 if (td2->td_kstack != 0) 949 vm_thread_dispose(td2); 950 if (!thread_alloc_stack(td2, pages)) { 951 error = ENOMEM; 952 goto fail2; 953 } 954 } 955 } 956 957 if ((flags & RFMEM) == 0) { 958 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 959 if (vm2 == NULL) { 960 error = ENOMEM; 961 goto fail2; 962 } 963 if (!swap_reserve(mem_charged)) { 964 /* 965 * The swap reservation failed. The accounting 966 * from the entries of the copied vm2 will be 967 * subtracted in vmspace_free(), so force the 968 * reservation there. 969 */ 970 swap_reserve_force(mem_charged); 971 error = ENOMEM; 972 goto fail2; 973 } 974 } else 975 vm2 = NULL; 976 977 /* 978 * XXX: This is ugly; when we copy resource usage, we need to bump 979 * per-cred resource counters. 980 */ 981 proc_set_cred_init(newproc, crhold(td->td_ucred)); 982 983 /* 984 * Initialize resource accounting for the child process. 985 */ 986 error = racct_proc_fork(p1, newproc); 987 if (error != 0) { 988 error = EAGAIN; 989 goto fail1; 990 } 991 992#ifdef MAC 993 mac_proc_init(newproc); 994#endif 995 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 996 STAILQ_INIT(&newproc->p_ktr); 997 998 /* We have to lock the process tree while we look for a pid. */ 999 sx_xlock(&proctree_lock); 1000 sx_xlock(&allproc_lock); 1001 1002 /* 1003 * Increment the count of procs running with this uid. Don't allow 1004 * a nonprivileged user to exceed their current limit. 1005 * 1006 * XXXRW: Can we avoid privilege here if it's not needed? 1007 */ 1008 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 1009 if (error == 0) 1010 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 1011 else { 1012 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 1013 lim_cur(td, RLIMIT_NPROC)); 1014 } 1015 if (ok) { 1016 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 1017 return (0); 1018 } 1019 1020 error = EAGAIN; 1021 sx_xunlock(&allproc_lock); 1022 sx_xunlock(&proctree_lock); 1023#ifdef MAC 1024 mac_proc_destroy(newproc); 1025#endif 1026 racct_proc_exit(newproc); 1027fail1: 1028 crfree(newproc->p_ucred); 1029 newproc->p_ucred = NULL; 1030fail2: 1031 if (vm2 != NULL) 1032 vmspace_free(vm2); 1033 uma_zfree(proc_zone, newproc); 1034 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1035 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1036 fdrop(fp_procdesc, td); 1037 } 1038 atomic_add_int(&nprocs, -1); 1039 pause("fork", hz / 2); 1040 return (error); 1041} 1042 1043/* 1044 * Handle the return of a child process from fork1(). This function 1045 * is called from the MD fork_trampoline() entry point. 1046 */ 1047void 1048fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1049 struct trapframe *frame) 1050{ 1051 struct proc *p; 1052 struct thread *td; 1053 struct thread *dtd; 1054 1055 td = curthread; 1056 p = td->td_proc; 1057 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1058 1059 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1060 td, td_get_sched(td), p->p_pid, td->td_name); 1061 1062 sched_fork_exit(td); 1063 /* 1064 * Processes normally resume in mi_switch() after being 1065 * cpu_switch()'ed to, but when children start up they arrive here 1066 * instead, so we must do much the same things as mi_switch() would. 1067 */ 1068 if ((dtd = PCPU_GET(deadthread))) { 1069 PCPU_SET(deadthread, NULL); 1070 thread_stash(dtd); 1071 } 1072 thread_unlock(td); 1073 1074 /* 1075 * cpu_fork_kthread_handler intercepts this function call to 1076 * have this call a non-return function to stay in kernel mode. 1077 * initproc has its own fork handler, but it does return. 1078 */ 1079 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1080 callout(arg, frame); 1081 1082 /* 1083 * Check if a kernel thread misbehaved and returned from its main 1084 * function. 1085 */ 1086 if (p->p_flag & P_KPROC) { 1087 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1088 td->td_name, p->p_pid); 1089 kthread_exit(); 1090 } 1091 mtx_assert(&Giant, MA_NOTOWNED); 1092 1093 if (p->p_sysent->sv_schedtail != NULL) 1094 (p->p_sysent->sv_schedtail)(td); 1095 td->td_pflags &= ~TDP_FORKING; 1096} 1097 1098/* 1099 * Simplified back end of syscall(), used when returning from fork() 1100 * directly into user mode. This function is passed in to fork_exit() 1101 * as the first parameter and is called when returning to a new 1102 * userland process. 1103 */ 1104void 1105fork_return(struct thread *td, struct trapframe *frame) 1106{ 1107 struct proc *p; 1108 1109 p = td->td_proc; 1110 if (td->td_dbgflags & TDB_STOPATFORK) { 1111 PROC_LOCK(p); 1112 if ((p->p_flag & P_TRACED) != 0) { 1113 /* 1114 * Inform the debugger if one is still present. 1115 */ 1116 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1117 ptracestop(td, SIGSTOP, NULL); 1118 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1119 } else { 1120 /* 1121 * ... otherwise clear the request. 1122 */ 1123 td->td_dbgflags &= ~TDB_STOPATFORK; 1124 } 1125 PROC_UNLOCK(p); 1126 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1127 /* 1128 * This is the start of a new thread in a traced 1129 * process. Report a system call exit event. 1130 */ 1131 PROC_LOCK(p); 1132 td->td_dbgflags |= TDB_SCX; 1133 _STOPEVENT(p, S_SCX, td->td_sa.code); 1134 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1135 (td->td_dbgflags & TDB_BORN) != 0) 1136 ptracestop(td, SIGTRAP, NULL); 1137 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1138 PROC_UNLOCK(p); 1139 } 1140 1141 /* 1142 * If the prison was killed mid-fork, die along with it. 1143 */ 1144 if (td->td_ucred->cr_prison->pr_flags & PR_REMOVE) 1145 exit1(td, 0, SIGKILL); 1146 1147 userret(td, frame); 1148 1149#ifdef KTRACE 1150 if (KTRPOINT(td, KTR_SYSRET)) 1151 ktrsysret(SYS_fork, 0, 0); 1152#endif 1153} 1154