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/11/sys/kern/kern_fork.c 369313 2021-02-19 21:50:26Z jamie $"); 39 40#include "opt_ktrace.h" 41#include "opt_kstack_pages.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/sysproto.h> 46#include <sys/eventhandler.h> 47#include <sys/fcntl.h> 48#include <sys/filedesc.h> 49#include <sys/jail.h> 50#include <sys/kernel.h> 51#include <sys/kthread.h> 52#include <sys/sysctl.h> 53#include <sys/lock.h> 54#include <sys/malloc.h> 55#include <sys/mutex.h> 56#include <sys/priv.h> 57#include <sys/proc.h> 58#include <sys/procdesc.h> 59#include <sys/pioctl.h> 60#include <sys/ptrace.h> 61#include <sys/racct.h> 62#include <sys/resourcevar.h> 63#include <sys/sched.h> 64#include <sys/syscall.h> 65#include <sys/vmmeter.h> 66#include <sys/vnode.h> 67#include <sys/acct.h> 68#include <sys/ktr.h> 69#include <sys/ktrace.h> 70#include <sys/unistd.h> 71#include <sys/sdt.h> 72#include <sys/sx.h> 73#include <sys/sysent.h> 74#include <sys/signalvar.h> 75 76#include <security/audit/audit.h> 77#include <security/mac/mac_framework.h> 78 79#include <vm/vm.h> 80#include <vm/pmap.h> 81#include <vm/vm_map.h> 82#include <vm/vm_extern.h> 83#include <vm/uma.h> 84#include <vm/vm_domain.h> 85 86#ifdef KDTRACE_HOOKS 87#include <sys/dtrace_bsd.h> 88dtrace_fork_func_t dtrace_fasttrap_fork; 89#endif 90 91SDT_PROVIDER_DECLARE(proc); 92SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); 93 94#ifndef _SYS_SYSPROTO_H_ 95struct fork_args { 96 int dummy; 97}; 98#endif 99 100EVENTHANDLER_LIST_DECLARE(process_fork); 101 102/* ARGSUSED */ 103int 104sys_fork(struct thread *td, struct fork_args *uap) 105{ 106 struct fork_req fr; 107 int error, pid; 108 109 bzero(&fr, sizeof(fr)); 110 fr.fr_flags = RFFDG | RFPROC; 111 fr.fr_pidp = &pid; 112 error = fork1(td, &fr); 113 if (error == 0) { 114 td->td_retval[0] = pid; 115 td->td_retval[1] = 0; 116 } 117 return (error); 118} 119 120/* ARGUSED */ 121int 122sys_pdfork(struct thread *td, struct pdfork_args *uap) 123{ 124 struct fork_req fr; 125 int error, fd, pid; 126 127 bzero(&fr, sizeof(fr)); 128 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC; 129 fr.fr_pidp = &pid; 130 fr.fr_pd_fd = &fd; 131 fr.fr_pd_flags = uap->flags; 132 /* 133 * It is necessary to return fd by reference because 0 is a valid file 134 * descriptor number, and the child needs to be able to distinguish 135 * itself from the parent using the return value. 136 */ 137 error = fork1(td, &fr); 138 if (error == 0) { 139 td->td_retval[0] = pid; 140 td->td_retval[1] = 0; 141 error = copyout(&fd, uap->fdp, sizeof(fd)); 142 } 143 return (error); 144} 145 146/* ARGSUSED */ 147int 148sys_vfork(struct thread *td, struct vfork_args *uap) 149{ 150 struct fork_req fr; 151 int error, pid; 152 153 bzero(&fr, sizeof(fr)); 154 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 155 fr.fr_pidp = &pid; 156 error = fork1(td, &fr); 157 if (error == 0) { 158 td->td_retval[0] = pid; 159 td->td_retval[1] = 0; 160 } 161 return (error); 162} 163 164int 165sys_rfork(struct thread *td, struct rfork_args *uap) 166{ 167 struct fork_req fr; 168 int error, pid; 169 170 /* Don't allow kernel-only flags. */ 171 if ((uap->flags & RFKERNELONLY) != 0) 172 return (EINVAL); 173 /* RFSPAWN must not appear with others */ 174 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN) 175 return (EINVAL); 176 177 AUDIT_ARG_FFLAGS(uap->flags); 178 bzero(&fr, sizeof(fr)); 179 if ((uap->flags & RFSPAWN) != 0) { 180 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 181 fr.fr_flags2 = FR2_DROPSIG_CAUGHT; 182 } else { 183 fr.fr_flags = uap->flags; 184 } 185 fr.fr_pidp = &pid; 186 error = fork1(td, &fr); 187 if (error == 0) { 188 td->td_retval[0] = pid; 189 td->td_retval[1] = 0; 190 } 191 return (error); 192} 193 194int nprocs = 1; /* process 0 */ 195int lastpid = 0; 196SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 197 "Last used PID"); 198 199/* 200 * Random component to lastpid generation. We mix in a random factor to make 201 * it a little harder to predict. We sanity check the modulus value to avoid 202 * doing it in critical paths. Don't let it be too small or we pointlessly 203 * waste randomness entropy, and don't let it be impossibly large. Using a 204 * modulus that is too big causes a LOT more process table scans and slows 205 * down fork processing as the pidchecked caching is defeated. 206 */ 207static int randompid = 0; 208 209static int 210sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 211{ 212 int error, pid; 213 214 error = sysctl_wire_old_buffer(req, sizeof(int)); 215 if (error != 0) 216 return(error); 217 sx_xlock(&allproc_lock); 218 pid = randompid; 219 error = sysctl_handle_int(oidp, &pid, 0, req); 220 if (error == 0 && req->newptr != NULL) { 221 if (pid == 0) 222 randompid = 0; 223 else if (pid == 1) 224 /* generate a random PID modulus between 100 and 1123 */ 225 randompid = 100 + arc4random() % 1024; 226 else if (pid < 0 || pid > pid_max - 100) 227 /* out of range */ 228 randompid = pid_max - 100; 229 else if (pid < 100) 230 /* Make it reasonable */ 231 randompid = 100; 232 else 233 randompid = pid; 234 } 235 sx_xunlock(&allproc_lock); 236 return (error); 237} 238 239SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 240 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value"); 241 242static int 243fork_findpid(int flags) 244{ 245 struct proc *p; 246 int trypid; 247 static int pidchecked = 0; 248 249 /* 250 * Requires allproc_lock in order to iterate over the list 251 * of processes, and proctree_lock to access p_pgrp. 252 */ 253 sx_assert(&allproc_lock, SX_LOCKED); 254 sx_assert(&proctree_lock, SX_LOCKED); 255 256 /* 257 * Find an unused process ID. We remember a range of unused IDs 258 * ready to use (from lastpid+1 through pidchecked-1). 259 * 260 * If RFHIGHPID is set (used during system boot), do not allocate 261 * low-numbered pids. 262 */ 263 trypid = lastpid + 1; 264 if (flags & RFHIGHPID) { 265 if (trypid < 10) 266 trypid = 10; 267 } else { 268 if (randompid) 269 trypid += arc4random() % randompid; 270 } 271retry: 272 /* 273 * If the process ID prototype has wrapped around, 274 * restart somewhat above 0, as the low-numbered procs 275 * tend to include daemons that don't exit. 276 */ 277 if (trypid >= pid_max) { 278 trypid = trypid % pid_max; 279 if (trypid < 100) 280 trypid += 100; 281 pidchecked = 0; 282 } 283 if (trypid >= pidchecked) { 284 int doingzomb = 0; 285 286 pidchecked = PID_MAX; 287 /* 288 * Scan the active and zombie procs to check whether this pid 289 * is in use. Remember the lowest pid that's greater 290 * than trypid, so we can avoid checking for a while. 291 * 292 * Avoid reuse of the process group id, session id or 293 * the reaper subtree id. Note that for process group 294 * and sessions, the amount of reserved pids is 295 * limited by process limit. For the subtree ids, the 296 * id is kept reserved only while there is a 297 * non-reaped process in the subtree, so amount of 298 * reserved pids is limited by process limit times 299 * two. 300 */ 301 p = LIST_FIRST(&allproc); 302again: 303 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 304 while (p->p_pid == trypid || 305 p->p_reapsubtree == trypid || 306 (p->p_pgrp != NULL && 307 (p->p_pgrp->pg_id == trypid || 308 (p->p_session != NULL && 309 p->p_session->s_sid == trypid)))) { 310 trypid++; 311 if (trypid >= pidchecked) 312 goto retry; 313 } 314 if (p->p_pid > trypid && pidchecked > p->p_pid) 315 pidchecked = p->p_pid; 316 if (p->p_pgrp != NULL) { 317 if (p->p_pgrp->pg_id > trypid && 318 pidchecked > p->p_pgrp->pg_id) 319 pidchecked = p->p_pgrp->pg_id; 320 if (p->p_session != NULL && 321 p->p_session->s_sid > trypid && 322 pidchecked > p->p_session->s_sid) 323 pidchecked = p->p_session->s_sid; 324 } 325 } 326 if (!doingzomb) { 327 doingzomb = 1; 328 p = LIST_FIRST(&zombproc); 329 goto again; 330 } 331 } 332 333 /* 334 * RFHIGHPID does not mess with the lastpid counter during boot. 335 */ 336 if (flags & RFHIGHPID) 337 pidchecked = 0; 338 else 339 lastpid = trypid; 340 341 return (trypid); 342} 343 344static int 345fork_norfproc(struct thread *td, int flags) 346{ 347 int error; 348 struct proc *p1; 349 350 KASSERT((flags & RFPROC) == 0, 351 ("fork_norfproc called with RFPROC set")); 352 p1 = td->td_proc; 353 354 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 355 (flags & (RFCFDG | RFFDG))) { 356 PROC_LOCK(p1); 357 if (thread_single(p1, SINGLE_BOUNDARY)) { 358 PROC_UNLOCK(p1); 359 return (ERESTART); 360 } 361 PROC_UNLOCK(p1); 362 } 363 364 error = vm_forkproc(td, NULL, NULL, NULL, flags); 365 if (error) 366 goto fail; 367 368 /* 369 * Close all file descriptors. 370 */ 371 if (flags & RFCFDG) { 372 struct filedesc *fdtmp; 373 fdtmp = fdinit(td->td_proc->p_fd, false); 374 fdescfree(td); 375 p1->p_fd = fdtmp; 376 } 377 378 /* 379 * Unshare file descriptors (from parent). 380 */ 381 if (flags & RFFDG) 382 fdunshare(td); 383 384fail: 385 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 386 (flags & (RFCFDG | RFFDG))) { 387 PROC_LOCK(p1); 388 thread_single_end(p1, SINGLE_BOUNDARY); 389 PROC_UNLOCK(p1); 390 } 391 return (error); 392} 393 394static void 395do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 396 struct vmspace *vm2, struct file *fp_procdesc) 397{ 398 struct proc *p1, *pptr; 399 int trypid; 400 struct filedesc *fd; 401 struct filedesc_to_leader *fdtol; 402 struct sigacts *newsigacts; 403 404 sx_assert(&proctree_lock, SX_SLOCKED); 405 sx_assert(&allproc_lock, SX_XLOCKED); 406 407 p1 = td->td_proc; 408 409 trypid = fork_findpid(fr->fr_flags); 410 411 sx_sunlock(&proctree_lock); 412 413 p2->p_state = PRS_NEW; /* protect against others */ 414 p2->p_pid = trypid; 415 AUDIT_ARG_PID(p2->p_pid); 416 LIST_INSERT_HEAD(&allproc, p2, p_list); 417 allproc_gen++; 418 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 419 tidhash_add(td2); 420 PROC_LOCK(p2); 421 PROC_LOCK(p1); 422 423 sx_xunlock(&allproc_lock); 424 425 bcopy(&p1->p_startcopy, &p2->p_startcopy, 426 __rangeof(struct proc, p_startcopy, p_endcopy)); 427 p2->p_elf_machine = p1->p_elf_machine; 428 p2->p_elf_flags = p1->p_elf_flags; 429 pargs_hold(p2->p_args); 430 431 PROC_UNLOCK(p1); 432 433 bzero(&p2->p_startzero, 434 __rangeof(struct proc, p_startzero, p_endzero)); 435 p2->p_ptevents = 0; 436 p2->p_pdeathsig = 0; 437 438 /* Tell the prison that we exist. */ 439 prison_proc_hold(p2->p_ucred->cr_prison); 440 441 PROC_UNLOCK(p2); 442 443 /* 444 * Malloc things while we don't hold any locks. 445 */ 446 if (fr->fr_flags & RFSIGSHARE) 447 newsigacts = NULL; 448 else 449 newsigacts = sigacts_alloc(); 450 451 /* 452 * Copy filedesc. 453 */ 454 if (fr->fr_flags & RFCFDG) { 455 fd = fdinit(p1->p_fd, false); 456 fdtol = NULL; 457 } else if (fr->fr_flags & RFFDG) { 458 fd = fdcopy(p1->p_fd); 459 fdtol = NULL; 460 } else { 461 fd = fdshare(p1->p_fd); 462 if (p1->p_fdtol == NULL) 463 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 464 p1->p_leader); 465 if ((fr->fr_flags & RFTHREAD) != 0) { 466 /* 467 * Shared file descriptor table, and shared 468 * process leaders. 469 */ 470 fdtol = p1->p_fdtol; 471 FILEDESC_XLOCK(p1->p_fd); 472 fdtol->fdl_refcount++; 473 FILEDESC_XUNLOCK(p1->p_fd); 474 } else { 475 /* 476 * Shared file descriptor table, and different 477 * process leaders. 478 */ 479 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 480 p1->p_fd, p2); 481 } 482 } 483 /* 484 * Make a proc table entry for the new process. 485 * Start by zeroing the section of proc that is zero-initialized, 486 * then copy the section that is copied directly from the parent. 487 */ 488 489 PROC_LOCK(p2); 490 PROC_LOCK(p1); 491 492 bzero(&td2->td_startzero, 493 __rangeof(struct thread, td_startzero, td_endzero)); 494 td2->td_sleeptimo = 0; 495 td2->td_vslock_sz = 0; 496 bzero(&td2->td_si, sizeof(td2->td_si)); 497 498 bcopy(&td->td_startcopy, &td2->td_startcopy, 499 __rangeof(struct thread, td_startcopy, td_endcopy)); 500 td2->td_sa = td->td_sa; 501 502 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 503 td2->td_sigstk = td->td_sigstk; 504 td2->td_flags = TDF_INMEM; 505 td2->td_lend_user_pri = PRI_MAX; 506 507#ifdef VIMAGE 508 td2->td_vnet = NULL; 509 td2->td_vnet_lpush = NULL; 510#endif 511 512 /* 513 * Allow the scheduler to initialize the child. 514 */ 515 thread_lock(td); 516 sched_fork(td, td2); 517 thread_unlock(td); 518 519 /* 520 * Duplicate sub-structures as needed. 521 * Increase reference counts on shared objects. 522 */ 523 p2->p_flag = P_INMEM; 524 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP | 525 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC); 526 p2->p_swtick = ticks; 527 if (p1->p_flag & P_PROFIL) 528 startprofclock(p2); 529 530 /* 531 * Whilst the proc lock is held, copy the VM domain data out 532 * using the VM domain method. 533 */ 534 vm_domain_policy_init(&p2->p_vm_dom_policy); 535 vm_domain_policy_localcopy(&p2->p_vm_dom_policy, 536 &p1->p_vm_dom_policy); 537 538 if (fr->fr_flags & RFSIGSHARE) { 539 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 540 } else { 541 sigacts_copy(newsigacts, p1->p_sigacts); 542 p2->p_sigacts = newsigacts; 543 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) { 544 mtx_lock(&p2->p_sigacts->ps_mtx); 545 sig_drop_caught(p2); 546 mtx_unlock(&p2->p_sigacts->ps_mtx); 547 } 548 } 549 550 if (fr->fr_flags & RFTSIGZMB) 551 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 552 else if (fr->fr_flags & RFLINUXTHPN) 553 p2->p_sigparent = SIGUSR1; 554 else 555 p2->p_sigparent = SIGCHLD; 556 557 p2->p_textvp = p1->p_textvp; 558 p2->p_fd = fd; 559 p2->p_fdtol = fdtol; 560 561 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 562 p2->p_flag |= P_PROTECTED; 563 p2->p_flag2 |= P2_INHERIT_PROTECTED; 564 } 565 566 /* 567 * p_limit is copy-on-write. Bump its refcount. 568 */ 569 lim_fork(p1, p2); 570 571 thread_cow_get_proc(td2, p2); 572 573 pstats_fork(p1->p_stats, p2->p_stats); 574 575 PROC_UNLOCK(p1); 576 PROC_UNLOCK(p2); 577 578 /* Bump references to the text vnode (for procfs). */ 579 if (p2->p_textvp) 580 vrefact(p2->p_textvp); 581 582 /* 583 * Set up linkage for kernel based threading. 584 */ 585 if ((fr->fr_flags & RFTHREAD) != 0) { 586 mtx_lock(&ppeers_lock); 587 p2->p_peers = p1->p_peers; 588 p1->p_peers = p2; 589 p2->p_leader = p1->p_leader; 590 mtx_unlock(&ppeers_lock); 591 PROC_LOCK(p1->p_leader); 592 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 593 PROC_UNLOCK(p1->p_leader); 594 /* 595 * The task leader is exiting, so process p1 is 596 * going to be killed shortly. Since p1 obviously 597 * isn't dead yet, we know that the leader is either 598 * sending SIGKILL's to all the processes in this 599 * task or is sleeping waiting for all the peers to 600 * exit. We let p1 complete the fork, but we need 601 * to go ahead and kill the new process p2 since 602 * the task leader may not get a chance to send 603 * SIGKILL to it. We leave it on the list so that 604 * the task leader will wait for this new process 605 * to commit suicide. 606 */ 607 PROC_LOCK(p2); 608 kern_psignal(p2, SIGKILL); 609 PROC_UNLOCK(p2); 610 } else 611 PROC_UNLOCK(p1->p_leader); 612 } else { 613 p2->p_peers = NULL; 614 p2->p_leader = p2; 615 } 616 617 sx_xlock(&proctree_lock); 618 PGRP_LOCK(p1->p_pgrp); 619 PROC_LOCK(p2); 620 PROC_LOCK(p1); 621 622 /* 623 * Preserve some more flags in subprocess. P_PROFIL has already 624 * been preserved. 625 */ 626 p2->p_flag |= p1->p_flag & P_SUGID; 627 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 628 SESS_LOCK(p1->p_session); 629 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 630 p2->p_flag |= P_CONTROLT; 631 SESS_UNLOCK(p1->p_session); 632 if (fr->fr_flags & RFPPWAIT) 633 p2->p_flag |= P_PPWAIT; 634 635 p2->p_pgrp = p1->p_pgrp; 636 LIST_INSERT_AFTER(p1, p2, p_pglist); 637 PGRP_UNLOCK(p1->p_pgrp); 638 LIST_INIT(&p2->p_children); 639 LIST_INIT(&p2->p_orphans); 640 641 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 642 643 /* 644 * If PF_FORK is set, the child process inherits the 645 * procfs ioctl flags from its parent. 646 */ 647 if (p1->p_pfsflags & PF_FORK) { 648 p2->p_stops = p1->p_stops; 649 p2->p_pfsflags = p1->p_pfsflags; 650 } 651 652 /* 653 * This begins the section where we must prevent the parent 654 * from being swapped. 655 */ 656 _PHOLD(p1); 657 PROC_UNLOCK(p1); 658 659 /* 660 * Attach the new process to its parent. 661 * 662 * If RFNOWAIT is set, the newly created process becomes a child 663 * of init. This effectively disassociates the child from the 664 * parent. 665 */ 666 if ((fr->fr_flags & RFNOWAIT) != 0) { 667 pptr = p1->p_reaper; 668 p2->p_reaper = pptr; 669 } else { 670 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 671 p1 : p1->p_reaper; 672 pptr = p1; 673 } 674 p2->p_pptr = pptr; 675 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 676 LIST_INIT(&p2->p_reaplist); 677 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 678 if (p2->p_reaper == p1) 679 p2->p_reapsubtree = p2->p_pid; 680 sx_xunlock(&proctree_lock); 681 682 /* Inform accounting that we have forked. */ 683 p2->p_acflag = AFORK; 684 PROC_UNLOCK(p2); 685 686#ifdef KTRACE 687 ktrprocfork(p1, p2); 688#endif 689 690 /* 691 * Finish creating the child process. It will return via a different 692 * execution path later. (ie: directly into user mode) 693 */ 694 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 695 696 if (fr->fr_flags == (RFFDG | RFPROC)) { 697 PCPU_INC(cnt.v_forks); 698 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 699 p2->p_vmspace->vm_ssize); 700 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 701 PCPU_INC(cnt.v_vforks); 702 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 703 p2->p_vmspace->vm_ssize); 704 } else if (p1 == &proc0) { 705 PCPU_INC(cnt.v_kthreads); 706 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 707 p2->p_vmspace->vm_ssize); 708 } else { 709 PCPU_INC(cnt.v_rforks); 710 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 711 p2->p_vmspace->vm_ssize); 712 } 713 714 /* 715 * Associate the process descriptor with the process before anything 716 * can happen that might cause that process to need the descriptor. 717 * However, don't do this until after fork(2) can no longer fail. 718 */ 719 if (fr->fr_flags & RFPROCDESC) 720 procdesc_new(p2, fr->fr_pd_flags); 721 722 /* 723 * Both processes are set up, now check if any loadable modules want 724 * to adjust anything. 725 */ 726 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 727 728 /* 729 * Set the child start time and mark the process as being complete. 730 */ 731 PROC_LOCK(p2); 732 PROC_LOCK(p1); 733 microuptime(&p2->p_stats->p_start); 734 PROC_SLOCK(p2); 735 p2->p_state = PRS_NORMAL; 736 PROC_SUNLOCK(p2); 737 738#ifdef KDTRACE_HOOKS 739 /* 740 * Tell the DTrace fasttrap provider about the new process so that any 741 * tracepoints inherited from the parent can be removed. We have to do 742 * this only after p_state is PRS_NORMAL since the fasttrap module will 743 * use pfind() later on. 744 */ 745 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 746 dtrace_fasttrap_fork(p1, p2); 747#endif 748 /* 749 * Hold the process so that it cannot exit after we make it runnable, 750 * but before we wait for the debugger. 751 */ 752 _PHOLD(p2); 753 if (fr->fr_flags & RFPPWAIT) { 754 td->td_pflags |= TDP_RFPPWAIT; 755 td->td_rfppwait_p = p2; 756 td->td_dbgflags |= TDB_VFORK; 757 } 758 PROC_UNLOCK(p2); 759 760 /* 761 * Now can be swapped. 762 */ 763 _PRELE(p1); 764 PROC_UNLOCK(p1); 765 766 /* 767 * Tell any interested parties about the new process. 768 */ 769 knote_fork(p1->p_klist, p2->p_pid); 770 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 771 772 if (fr->fr_flags & RFPROCDESC) { 773 procdesc_finit(p2->p_procdesc, fp_procdesc); 774 fdrop(fp_procdesc, td); 775 } 776 777 /* 778 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 779 * synced with forks in progress so it is OK if we miss it 780 * if being set atm. 781 */ 782 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 783 sx_xlock(&proctree_lock); 784 PROC_LOCK(p2); 785 786 /* 787 * p1->p_ptevents & p1->p_pptr are protected by both 788 * process and proctree locks for modifications, 789 * so owning proctree_lock allows the race-free read. 790 */ 791 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 792 /* 793 * Arrange for debugger to receive the fork event. 794 * 795 * We can report PL_FLAG_FORKED regardless of 796 * P_FOLLOWFORK settings, but it does not make a sense 797 * for runaway child. 798 */ 799 td->td_dbgflags |= TDB_FORK; 800 td->td_dbg_forked = p2->p_pid; 801 td2->td_dbgflags |= TDB_STOPATFORK; 802 proc_set_traced(p2, true); 803 CTR2(KTR_PTRACE, 804 "do_fork: attaching to new child pid %d: oppid %d", 805 p2->p_pid, p2->p_oppid); 806 proc_reparent(p2, p1->p_pptr); 807 } 808 PROC_UNLOCK(p2); 809 sx_xunlock(&proctree_lock); 810 } 811 812 if ((fr->fr_flags & RFSTOPPED) == 0) { 813 /* 814 * If RFSTOPPED not requested, make child runnable and 815 * add to run queue. 816 */ 817 thread_lock(td2); 818 TD_SET_CAN_RUN(td2); 819 sched_add(td2, SRQ_BORING); 820 thread_unlock(td2); 821 if (fr->fr_pidp != NULL) 822 *fr->fr_pidp = p2->p_pid; 823 } else { 824 *fr->fr_procp = p2; 825 } 826 827 PROC_LOCK(p2); 828 _PRELE(p2); 829 racct_proc_fork_done(p2); 830 PROC_UNLOCK(p2); 831} 832 833int 834fork1(struct thread *td, struct fork_req *fr) 835{ 836 struct proc *p1, *newproc; 837 struct thread *td2; 838 struct vmspace *vm2; 839 struct file *fp_procdesc; 840 vm_ooffset_t mem_charged; 841 int error, nprocs_new, ok; 842 static int curfail; 843 static struct timeval lastfail; 844 int flags, pages; 845 846 flags = fr->fr_flags; 847 pages = fr->fr_pages; 848 849 if ((flags & RFSTOPPED) != 0) 850 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 851 else 852 MPASS(fr->fr_procp == NULL); 853 854 /* Check for the undefined or unimplemented flags. */ 855 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 856 return (EINVAL); 857 858 /* Signal value requires RFTSIGZMB. */ 859 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 860 return (EINVAL); 861 862 /* Can't copy and clear. */ 863 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 864 return (EINVAL); 865 866 /* Check the validity of the signal number. */ 867 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 868 return (EINVAL); 869 870 if ((flags & RFPROCDESC) != 0) { 871 /* Can't not create a process yet get a process descriptor. */ 872 if ((flags & RFPROC) == 0) 873 return (EINVAL); 874 875 /* Must provide a place to put a procdesc if creating one. */ 876 if (fr->fr_pd_fd == NULL) 877 return (EINVAL); 878 879 /* Check if we are using supported flags. */ 880 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 881 return (EINVAL); 882 } 883 884 p1 = td->td_proc; 885 886 /* 887 * Here we don't create a new process, but we divorce 888 * certain parts of a process from itself. 889 */ 890 if ((flags & RFPROC) == 0) { 891 if (fr->fr_procp != NULL) 892 *fr->fr_procp = NULL; 893 else if (fr->fr_pidp != NULL) 894 *fr->fr_pidp = 0; 895 return (fork_norfproc(td, flags)); 896 } 897 898 fp_procdesc = NULL; 899 newproc = NULL; 900 vm2 = NULL; 901 902 /* 903 * Increment the nprocs resource before allocations occur. 904 * Although process entries are dynamically created, we still 905 * keep a global limit on the maximum number we will 906 * create. There are hard-limits as to the number of processes 907 * that can run, established by the KVA and memory usage for 908 * the process data. 909 * 910 * Don't allow a nonprivileged user to use the last ten 911 * processes; don't let root exceed the limit. 912 */ 913 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 914 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, 915 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { 916 error = EAGAIN; 917 sx_xlock(&allproc_lock); 918 if (ppsratecheck(&lastfail, &curfail, 1)) { 919 printf("maxproc limit exceeded by uid %u (pid %d); " 920 "see tuning(7) and login.conf(5)\n", 921 td->td_ucred->cr_ruid, p1->p_pid); 922 } 923 sx_xunlock(&allproc_lock); 924 goto fail2; 925 } 926 927 /* 928 * If required, create a process descriptor in the parent first; we 929 * will abandon it if something goes wrong. We don't finit() until 930 * later. 931 */ 932 if (flags & RFPROCDESC) { 933 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 934 fr->fr_pd_flags, fr->fr_pd_fcaps); 935 if (error != 0) 936 goto fail2; 937 } 938 939 mem_charged = 0; 940 if (pages == 0) 941 pages = kstack_pages; 942 /* Allocate new proc. */ 943 newproc = uma_zalloc(proc_zone, M_WAITOK); 944 td2 = FIRST_THREAD_IN_PROC(newproc); 945 if (td2 == NULL) { 946 td2 = thread_alloc(pages); 947 if (td2 == NULL) { 948 error = ENOMEM; 949 goto fail2; 950 } 951 proc_linkup(newproc, td2); 952 } else { 953 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 954 if (td2->td_kstack != 0) 955 vm_thread_dispose(td2); 956 if (!thread_alloc_stack(td2, pages)) { 957 error = ENOMEM; 958 goto fail2; 959 } 960 } 961 } 962 963 if ((flags & RFMEM) == 0) { 964 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 965 if (vm2 == NULL) { 966 error = ENOMEM; 967 goto fail2; 968 } 969 if (!swap_reserve(mem_charged)) { 970 /* 971 * The swap reservation failed. The accounting 972 * from the entries of the copied vm2 will be 973 * subtracted in vmspace_free(), so force the 974 * reservation there. 975 */ 976 swap_reserve_force(mem_charged); 977 error = ENOMEM; 978 goto fail2; 979 } 980 } else 981 vm2 = NULL; 982 983 /* 984 * XXX: This is ugly; when we copy resource usage, we need to bump 985 * per-cred resource counters. 986 */ 987 proc_set_cred_init(newproc, crhold(td->td_ucred)); 988 989 /* 990 * Initialize resource accounting for the child process. 991 */ 992 error = racct_proc_fork(p1, newproc); 993 if (error != 0) { 994 error = EAGAIN; 995 goto fail1; 996 } 997 998#ifdef MAC 999 mac_proc_init(newproc); 1000#endif 1001 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 1002 STAILQ_INIT(&newproc->p_ktr); 1003 1004 /* We have to lock the process tree while we look for a pid. */ 1005 sx_slock(&proctree_lock); 1006 sx_xlock(&allproc_lock); 1007 1008 /* 1009 * Increment the count of procs running with this uid. Don't allow 1010 * a nonprivileged user to exceed their current limit. 1011 * 1012 * XXXRW: Can we avoid privilege here if it's not needed? 1013 */ 1014 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 1015 if (error == 0) 1016 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 1017 else { 1018 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 1019 lim_cur(td, RLIMIT_NPROC)); 1020 } 1021 if (ok) { 1022 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 1023 return (0); 1024 } 1025 1026 error = EAGAIN; 1027 sx_sunlock(&proctree_lock); 1028 sx_xunlock(&allproc_lock); 1029#ifdef MAC 1030 mac_proc_destroy(newproc); 1031#endif 1032 racct_proc_exit(newproc); 1033fail1: 1034 crfree(newproc->p_ucred); 1035 newproc->p_ucred = NULL; 1036fail2: 1037 if (vm2 != NULL) 1038 vmspace_free(vm2); 1039 uma_zfree(proc_zone, newproc); 1040 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1041 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1042 fdrop(fp_procdesc, td); 1043 } 1044 atomic_add_int(&nprocs, -1); 1045 pause("fork", hz / 2); 1046 return (error); 1047} 1048 1049/* 1050 * Handle the return of a child process from fork1(). This function 1051 * is called from the MD fork_trampoline() entry point. 1052 */ 1053void 1054fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1055 struct trapframe *frame) 1056{ 1057 struct proc *p; 1058 struct thread *td; 1059 struct thread *dtd; 1060 1061 td = curthread; 1062 p = td->td_proc; 1063 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1064 1065 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1066 td, td_get_sched(td), p->p_pid, td->td_name); 1067 1068 sched_fork_exit(td); 1069 /* 1070 * Processes normally resume in mi_switch() after being 1071 * cpu_switch()'ed to, but when children start up they arrive here 1072 * instead, so we must do much the same things as mi_switch() would. 1073 */ 1074 if ((dtd = PCPU_GET(deadthread))) { 1075 PCPU_SET(deadthread, NULL); 1076 thread_stash(dtd); 1077 } 1078 thread_unlock(td); 1079 1080 /* 1081 * cpu_fork_kthread_handler intercepts this function call to 1082 * have this call a non-return function to stay in kernel mode. 1083 * initproc has its own fork handler, but it does return. 1084 */ 1085 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1086 callout(arg, frame); 1087 1088 /* 1089 * Check if a kernel thread misbehaved and returned from its main 1090 * function. 1091 */ 1092 if (p->p_flag & P_KPROC) { 1093 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1094 td->td_name, p->p_pid); 1095 kthread_exit(); 1096 } 1097 mtx_assert(&Giant, MA_NOTOWNED); 1098 1099 if (p->p_sysent->sv_schedtail != NULL) 1100 (p->p_sysent->sv_schedtail)(td); 1101 td->td_pflags &= ~TDP_FORKING; 1102} 1103 1104/* 1105 * Simplified back end of syscall(), used when returning from fork() 1106 * directly into user mode. This function is passed in to fork_exit() 1107 * as the first parameter and is called when returning to a new 1108 * userland process. 1109 */ 1110void 1111fork_return(struct thread *td, struct trapframe *frame) 1112{ 1113 struct proc *p; 1114 1115 p = td->td_proc; 1116 if (td->td_dbgflags & TDB_STOPATFORK) { 1117 PROC_LOCK(p); 1118 if ((p->p_flag & P_TRACED) != 0) { 1119 /* 1120 * Inform the debugger if one is still present. 1121 */ 1122 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1123 ptracestop(td, SIGSTOP, NULL); 1124 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1125 } else { 1126 /* 1127 * ... otherwise clear the request. 1128 */ 1129 td->td_dbgflags &= ~TDB_STOPATFORK; 1130 } 1131 PROC_UNLOCK(p); 1132 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1133 /* 1134 * This is the start of a new thread in a traced 1135 * process. Report a system call exit event. 1136 */ 1137 PROC_LOCK(p); 1138 td->td_dbgflags |= TDB_SCX; 1139 _STOPEVENT(p, S_SCX, td->td_sa.code); 1140 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1141 (td->td_dbgflags & TDB_BORN) != 0) 1142 ptracestop(td, SIGTRAP, NULL); 1143 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1144 PROC_UNLOCK(p); 1145 } 1146 1147 /* 1148 * If the prison was killed mid-fork, die along with it. 1149 */ 1150 if (td->td_ucred->cr_prison->pr_flags & PR_REMOVE) 1151 exit1(td, 0, SIGKILL); 1152 1153 userret(td, frame); 1154 1155#ifdef KTRACE 1156 if (KTRPOINT(td, KTR_SYSRET)) 1157 ktrsysret(SYS_fork, 0, 0); 1158#endif 1159} 1160