kern_fork.c revision 333162
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 333162 2018-05-02 07:57:36Z kib $"); 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); 173 174 AUDIT_ARG_FFLAGS(uap->flags); 175 bzero(&fr, sizeof(fr)); 176 fr.fr_flags = uap->flags; 177 fr.fr_pidp = &pid; 178 error = fork1(td, &fr); 179 if (error == 0) { 180 td->td_retval[0] = pid; 181 td->td_retval[1] = 0; 182 } 183 return (error); 184} 185 186int nprocs = 1; /* process 0 */ 187int lastpid = 0; 188SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 189 "Last used PID"); 190 191/* 192 * Random component to lastpid generation. We mix in a random factor to make 193 * it a little harder to predict. We sanity check the modulus value to avoid 194 * doing it in critical paths. Don't let it be too small or we pointlessly 195 * waste randomness entropy, and don't let it be impossibly large. Using a 196 * modulus that is too big causes a LOT more process table scans and slows 197 * down fork processing as the pidchecked caching is defeated. 198 */ 199static int randompid = 0; 200 201static int 202sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 203{ 204 int error, pid; 205 206 error = sysctl_wire_old_buffer(req, sizeof(int)); 207 if (error != 0) 208 return(error); 209 sx_xlock(&allproc_lock); 210 pid = randompid; 211 error = sysctl_handle_int(oidp, &pid, 0, req); 212 if (error == 0 && req->newptr != NULL) { 213 if (pid == 0) 214 randompid = 0; 215 else if (pid == 1) 216 /* generate a random PID modulus between 100 and 1123 */ 217 randompid = 100 + arc4random() % 1024; 218 else if (pid < 0 || pid > pid_max - 100) 219 /* out of range */ 220 randompid = pid_max - 100; 221 else if (pid < 100) 222 /* Make it reasonable */ 223 randompid = 100; 224 else 225 randompid = pid; 226 } 227 sx_xunlock(&allproc_lock); 228 return (error); 229} 230 231SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 232 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value"); 233 234static int 235fork_findpid(int flags) 236{ 237 struct proc *p; 238 int trypid; 239 static int pidchecked = 0; 240 241 /* 242 * Requires allproc_lock in order to iterate over the list 243 * of processes, and proctree_lock to access p_pgrp. 244 */ 245 sx_assert(&allproc_lock, SX_LOCKED); 246 sx_assert(&proctree_lock, SX_LOCKED); 247 248 /* 249 * Find an unused process ID. We remember a range of unused IDs 250 * ready to use (from lastpid+1 through pidchecked-1). 251 * 252 * If RFHIGHPID is set (used during system boot), do not allocate 253 * low-numbered pids. 254 */ 255 trypid = lastpid + 1; 256 if (flags & RFHIGHPID) { 257 if (trypid < 10) 258 trypid = 10; 259 } else { 260 if (randompid) 261 trypid += arc4random() % randompid; 262 } 263retry: 264 /* 265 * If the process ID prototype has wrapped around, 266 * restart somewhat above 0, as the low-numbered procs 267 * tend to include daemons that don't exit. 268 */ 269 if (trypid >= pid_max) { 270 trypid = trypid % pid_max; 271 if (trypid < 100) 272 trypid += 100; 273 pidchecked = 0; 274 } 275 if (trypid >= pidchecked) { 276 int doingzomb = 0; 277 278 pidchecked = PID_MAX; 279 /* 280 * Scan the active and zombie procs to check whether this pid 281 * is in use. Remember the lowest pid that's greater 282 * than trypid, so we can avoid checking for a while. 283 * 284 * Avoid reuse of the process group id, session id or 285 * the reaper subtree id. Note that for process group 286 * and sessions, the amount of reserved pids is 287 * limited by process limit. For the subtree ids, the 288 * id is kept reserved only while there is a 289 * non-reaped process in the subtree, so amount of 290 * reserved pids is limited by process limit times 291 * two. 292 */ 293 p = LIST_FIRST(&allproc); 294again: 295 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 296 while (p->p_pid == trypid || 297 p->p_reapsubtree == trypid || 298 (p->p_pgrp != NULL && 299 (p->p_pgrp->pg_id == trypid || 300 (p->p_session != NULL && 301 p->p_session->s_sid == trypid)))) { 302 trypid++; 303 if (trypid >= pidchecked) 304 goto retry; 305 } 306 if (p->p_pid > trypid && pidchecked > p->p_pid) 307 pidchecked = p->p_pid; 308 if (p->p_pgrp != NULL) { 309 if (p->p_pgrp->pg_id > trypid && 310 pidchecked > p->p_pgrp->pg_id) 311 pidchecked = p->p_pgrp->pg_id; 312 if (p->p_session != NULL && 313 p->p_session->s_sid > trypid && 314 pidchecked > p->p_session->s_sid) 315 pidchecked = p->p_session->s_sid; 316 } 317 } 318 if (!doingzomb) { 319 doingzomb = 1; 320 p = LIST_FIRST(&zombproc); 321 goto again; 322 } 323 } 324 325 /* 326 * RFHIGHPID does not mess with the lastpid counter during boot. 327 */ 328 if (flags & RFHIGHPID) 329 pidchecked = 0; 330 else 331 lastpid = trypid; 332 333 return (trypid); 334} 335 336static int 337fork_norfproc(struct thread *td, int flags) 338{ 339 int error; 340 struct proc *p1; 341 342 KASSERT((flags & RFPROC) == 0, 343 ("fork_norfproc called with RFPROC set")); 344 p1 = td->td_proc; 345 346 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 347 (flags & (RFCFDG | RFFDG))) { 348 PROC_LOCK(p1); 349 if (thread_single(p1, SINGLE_BOUNDARY)) { 350 PROC_UNLOCK(p1); 351 return (ERESTART); 352 } 353 PROC_UNLOCK(p1); 354 } 355 356 error = vm_forkproc(td, NULL, NULL, NULL, flags); 357 if (error) 358 goto fail; 359 360 /* 361 * Close all file descriptors. 362 */ 363 if (flags & RFCFDG) { 364 struct filedesc *fdtmp; 365 fdtmp = fdinit(td->td_proc->p_fd, false); 366 fdescfree(td); 367 p1->p_fd = fdtmp; 368 } 369 370 /* 371 * Unshare file descriptors (from parent). 372 */ 373 if (flags & RFFDG) 374 fdunshare(td); 375 376fail: 377 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 378 (flags & (RFCFDG | RFFDG))) { 379 PROC_LOCK(p1); 380 thread_single_end(p1, SINGLE_BOUNDARY); 381 PROC_UNLOCK(p1); 382 } 383 return (error); 384} 385 386static void 387do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 388 struct vmspace *vm2, struct file *fp_procdesc) 389{ 390 struct proc *p1, *pptr; 391 int trypid; 392 struct filedesc *fd; 393 struct filedesc_to_leader *fdtol; 394 struct sigacts *newsigacts; 395 396 sx_assert(&proctree_lock, SX_SLOCKED); 397 sx_assert(&allproc_lock, SX_XLOCKED); 398 399 p1 = td->td_proc; 400 401 trypid = fork_findpid(fr->fr_flags); 402 403 sx_sunlock(&proctree_lock); 404 405 p2->p_state = PRS_NEW; /* protect against others */ 406 p2->p_pid = trypid; 407 AUDIT_ARG_PID(p2->p_pid); 408 LIST_INSERT_HEAD(&allproc, p2, p_list); 409 allproc_gen++; 410 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 411 tidhash_add(td2); 412 PROC_LOCK(p2); 413 PROC_LOCK(p1); 414 415 sx_xunlock(&allproc_lock); 416 417 bcopy(&p1->p_startcopy, &p2->p_startcopy, 418 __rangeof(struct proc, p_startcopy, p_endcopy)); 419 p2->p_elf_machine = p1->p_elf_machine; 420 p2->p_elf_flags = p1->p_elf_flags; 421 pargs_hold(p2->p_args); 422 423 PROC_UNLOCK(p1); 424 425 bzero(&p2->p_startzero, 426 __rangeof(struct proc, p_startzero, p_endzero)); 427 p2->p_ptevents = 0; 428 p2->p_pdeathsig = 0; 429 430 /* Tell the prison that we exist. */ 431 prison_proc_hold(p2->p_ucred->cr_prison); 432 433 PROC_UNLOCK(p2); 434 435 /* 436 * Malloc things while we don't hold any locks. 437 */ 438 if (fr->fr_flags & RFSIGSHARE) 439 newsigacts = NULL; 440 else 441 newsigacts = sigacts_alloc(); 442 443 /* 444 * Copy filedesc. 445 */ 446 if (fr->fr_flags & RFCFDG) { 447 fd = fdinit(p1->p_fd, false); 448 fdtol = NULL; 449 } else if (fr->fr_flags & RFFDG) { 450 fd = fdcopy(p1->p_fd); 451 fdtol = NULL; 452 } else { 453 fd = fdshare(p1->p_fd); 454 if (p1->p_fdtol == NULL) 455 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 456 p1->p_leader); 457 if ((fr->fr_flags & RFTHREAD) != 0) { 458 /* 459 * Shared file descriptor table, and shared 460 * process leaders. 461 */ 462 fdtol = p1->p_fdtol; 463 FILEDESC_XLOCK(p1->p_fd); 464 fdtol->fdl_refcount++; 465 FILEDESC_XUNLOCK(p1->p_fd); 466 } else { 467 /* 468 * Shared file descriptor table, and different 469 * process leaders. 470 */ 471 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 472 p1->p_fd, p2); 473 } 474 } 475 /* 476 * Make a proc table entry for the new process. 477 * Start by zeroing the section of proc that is zero-initialized, 478 * then copy the section that is copied directly from the parent. 479 */ 480 481 PROC_LOCK(p2); 482 PROC_LOCK(p1); 483 484 bzero(&td2->td_startzero, 485 __rangeof(struct thread, td_startzero, td_endzero)); 486 td2->td_sleeptimo = 0; 487 td2->td_vslock_sz = 0; 488 bzero(&td2->td_si, sizeof(td2->td_si)); 489 490 bcopy(&td->td_startcopy, &td2->td_startcopy, 491 __rangeof(struct thread, td_startcopy, td_endcopy)); 492 td2->td_sa = td->td_sa; 493 494 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 495 td2->td_sigstk = td->td_sigstk; 496 td2->td_flags = TDF_INMEM; 497 td2->td_lend_user_pri = PRI_MAX; 498 499#ifdef VIMAGE 500 td2->td_vnet = NULL; 501 td2->td_vnet_lpush = NULL; 502#endif 503 504 /* 505 * Allow the scheduler to initialize the child. 506 */ 507 thread_lock(td); 508 sched_fork(td, td2); 509 thread_unlock(td); 510 511 /* 512 * Duplicate sub-structures as needed. 513 * Increase reference counts on shared objects. 514 */ 515 p2->p_flag = P_INMEM; 516 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP); 517 p2->p_swtick = ticks; 518 if (p1->p_flag & P_PROFIL) 519 startprofclock(p2); 520 521 /* 522 * Whilst the proc lock is held, copy the VM domain data out 523 * using the VM domain method. 524 */ 525 vm_domain_policy_init(&p2->p_vm_dom_policy); 526 vm_domain_policy_localcopy(&p2->p_vm_dom_policy, 527 &p1->p_vm_dom_policy); 528 529 if (fr->fr_flags & RFSIGSHARE) { 530 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 531 } else { 532 sigacts_copy(newsigacts, p1->p_sigacts); 533 p2->p_sigacts = newsigacts; 534 } 535 536 if (fr->fr_flags & RFTSIGZMB) 537 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 538 else if (fr->fr_flags & RFLINUXTHPN) 539 p2->p_sigparent = SIGUSR1; 540 else 541 p2->p_sigparent = SIGCHLD; 542 543 p2->p_textvp = p1->p_textvp; 544 p2->p_fd = fd; 545 p2->p_fdtol = fdtol; 546 547 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 548 p2->p_flag |= P_PROTECTED; 549 p2->p_flag2 |= P2_INHERIT_PROTECTED; 550 } 551 552 /* 553 * p_limit is copy-on-write. Bump its refcount. 554 */ 555 lim_fork(p1, p2); 556 557 thread_cow_get_proc(td2, p2); 558 559 pstats_fork(p1->p_stats, p2->p_stats); 560 561 PROC_UNLOCK(p1); 562 PROC_UNLOCK(p2); 563 564 /* Bump references to the text vnode (for procfs). */ 565 if (p2->p_textvp) 566 vrefact(p2->p_textvp); 567 568 /* 569 * Set up linkage for kernel based threading. 570 */ 571 if ((fr->fr_flags & RFTHREAD) != 0) { 572 mtx_lock(&ppeers_lock); 573 p2->p_peers = p1->p_peers; 574 p1->p_peers = p2; 575 p2->p_leader = p1->p_leader; 576 mtx_unlock(&ppeers_lock); 577 PROC_LOCK(p1->p_leader); 578 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 579 PROC_UNLOCK(p1->p_leader); 580 /* 581 * The task leader is exiting, so process p1 is 582 * going to be killed shortly. Since p1 obviously 583 * isn't dead yet, we know that the leader is either 584 * sending SIGKILL's to all the processes in this 585 * task or is sleeping waiting for all the peers to 586 * exit. We let p1 complete the fork, but we need 587 * to go ahead and kill the new process p2 since 588 * the task leader may not get a chance to send 589 * SIGKILL to it. We leave it on the list so that 590 * the task leader will wait for this new process 591 * to commit suicide. 592 */ 593 PROC_LOCK(p2); 594 kern_psignal(p2, SIGKILL); 595 PROC_UNLOCK(p2); 596 } else 597 PROC_UNLOCK(p1->p_leader); 598 } else { 599 p2->p_peers = NULL; 600 p2->p_leader = p2; 601 } 602 603 sx_xlock(&proctree_lock); 604 PGRP_LOCK(p1->p_pgrp); 605 PROC_LOCK(p2); 606 PROC_LOCK(p1); 607 608 /* 609 * Preserve some more flags in subprocess. P_PROFIL has already 610 * been preserved. 611 */ 612 p2->p_flag |= p1->p_flag & P_SUGID; 613 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 614 SESS_LOCK(p1->p_session); 615 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 616 p2->p_flag |= P_CONTROLT; 617 SESS_UNLOCK(p1->p_session); 618 if (fr->fr_flags & RFPPWAIT) 619 p2->p_flag |= P_PPWAIT; 620 621 p2->p_pgrp = p1->p_pgrp; 622 LIST_INSERT_AFTER(p1, p2, p_pglist); 623 PGRP_UNLOCK(p1->p_pgrp); 624 LIST_INIT(&p2->p_children); 625 LIST_INIT(&p2->p_orphans); 626 627 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 628 629 /* 630 * If PF_FORK is set, the child process inherits the 631 * procfs ioctl flags from its parent. 632 */ 633 if (p1->p_pfsflags & PF_FORK) { 634 p2->p_stops = p1->p_stops; 635 p2->p_pfsflags = p1->p_pfsflags; 636 } 637 638 /* 639 * This begins the section where we must prevent the parent 640 * from being swapped. 641 */ 642 _PHOLD(p1); 643 PROC_UNLOCK(p1); 644 645 /* 646 * Attach the new process to its parent. 647 * 648 * If RFNOWAIT is set, the newly created process becomes a child 649 * of init. This effectively disassociates the child from the 650 * parent. 651 */ 652 if ((fr->fr_flags & RFNOWAIT) != 0) { 653 pptr = p1->p_reaper; 654 p2->p_reaper = pptr; 655 } else { 656 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 657 p1 : p1->p_reaper; 658 pptr = p1; 659 } 660 p2->p_pptr = pptr; 661 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 662 LIST_INIT(&p2->p_reaplist); 663 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 664 if (p2->p_reaper == p1) 665 p2->p_reapsubtree = p2->p_pid; 666 sx_xunlock(&proctree_lock); 667 668 /* Inform accounting that we have forked. */ 669 p2->p_acflag = AFORK; 670 PROC_UNLOCK(p2); 671 672#ifdef KTRACE 673 ktrprocfork(p1, p2); 674#endif 675 676 /* 677 * Finish creating the child process. It will return via a different 678 * execution path later. (ie: directly into user mode) 679 */ 680 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 681 682 if (fr->fr_flags == (RFFDG | RFPROC)) { 683 PCPU_INC(cnt.v_forks); 684 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 685 p2->p_vmspace->vm_ssize); 686 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 687 PCPU_INC(cnt.v_vforks); 688 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 689 p2->p_vmspace->vm_ssize); 690 } else if (p1 == &proc0) { 691 PCPU_INC(cnt.v_kthreads); 692 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 693 p2->p_vmspace->vm_ssize); 694 } else { 695 PCPU_INC(cnt.v_rforks); 696 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 697 p2->p_vmspace->vm_ssize); 698 } 699 700 /* 701 * Associate the process descriptor with the process before anything 702 * can happen that might cause that process to need the descriptor. 703 * However, don't do this until after fork(2) can no longer fail. 704 */ 705 if (fr->fr_flags & RFPROCDESC) 706 procdesc_new(p2, fr->fr_pd_flags); 707 708 /* 709 * Both processes are set up, now check if any loadable modules want 710 * to adjust anything. 711 */ 712 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 713 714 /* 715 * Set the child start time and mark the process as being complete. 716 */ 717 PROC_LOCK(p2); 718 PROC_LOCK(p1); 719 microuptime(&p2->p_stats->p_start); 720 PROC_SLOCK(p2); 721 p2->p_state = PRS_NORMAL; 722 PROC_SUNLOCK(p2); 723 724#ifdef KDTRACE_HOOKS 725 /* 726 * Tell the DTrace fasttrap provider about the new process so that any 727 * tracepoints inherited from the parent can be removed. We have to do 728 * this only after p_state is PRS_NORMAL since the fasttrap module will 729 * use pfind() later on. 730 */ 731 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 732 dtrace_fasttrap_fork(p1, p2); 733#endif 734 /* 735 * Hold the process so that it cannot exit after we make it runnable, 736 * but before we wait for the debugger. 737 */ 738 _PHOLD(p2); 739 if (p1->p_ptevents & PTRACE_FORK) { 740 /* 741 * Arrange for debugger to receive the fork event. 742 * 743 * We can report PL_FLAG_FORKED regardless of 744 * P_FOLLOWFORK settings, but it does not make a sense 745 * for runaway child. 746 */ 747 td->td_dbgflags |= TDB_FORK; 748 td->td_dbg_forked = p2->p_pid; 749 td2->td_dbgflags |= TDB_STOPATFORK; 750 } 751 if (fr->fr_flags & RFPPWAIT) { 752 td->td_pflags |= TDP_RFPPWAIT; 753 td->td_rfppwait_p = p2; 754 td->td_dbgflags |= TDB_VFORK; 755 } 756 PROC_UNLOCK(p2); 757 758 /* 759 * Now can be swapped. 760 */ 761 _PRELE(p1); 762 PROC_UNLOCK(p1); 763 764 /* 765 * Tell any interested parties about the new process. 766 */ 767 knote_fork(p1->p_klist, p2->p_pid); 768 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 769 770 if (fr->fr_flags & RFPROCDESC) { 771 procdesc_finit(p2->p_procdesc, fp_procdesc); 772 fdrop(fp_procdesc, td); 773 } 774 775 if ((fr->fr_flags & RFSTOPPED) == 0) { 776 /* 777 * If RFSTOPPED not requested, make child runnable and 778 * add to run queue. 779 */ 780 thread_lock(td2); 781 TD_SET_CAN_RUN(td2); 782 sched_add(td2, SRQ_BORING); 783 thread_unlock(td2); 784 if (fr->fr_pidp != NULL) 785 *fr->fr_pidp = p2->p_pid; 786 } else { 787 *fr->fr_procp = p2; 788 } 789 790 PROC_LOCK(p2); 791 /* 792 * Wait until debugger is attached to child. 793 */ 794 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0) 795 cv_wait(&p2->p_dbgwait, &p2->p_mtx); 796 _PRELE(p2); 797 racct_proc_fork_done(p2); 798 PROC_UNLOCK(p2); 799} 800 801int 802fork1(struct thread *td, struct fork_req *fr) 803{ 804 struct proc *p1, *newproc; 805 struct thread *td2; 806 struct vmspace *vm2; 807 struct file *fp_procdesc; 808 vm_ooffset_t mem_charged; 809 int error, nprocs_new, ok; 810 static int curfail; 811 static struct timeval lastfail; 812 int flags, pages; 813 814 flags = fr->fr_flags; 815 pages = fr->fr_pages; 816 817 if ((flags & RFSTOPPED) != 0) 818 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 819 else 820 MPASS(fr->fr_procp == NULL); 821 822 /* Check for the undefined or unimplemented flags. */ 823 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 824 return (EINVAL); 825 826 /* Signal value requires RFTSIGZMB. */ 827 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 828 return (EINVAL); 829 830 /* Can't copy and clear. */ 831 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 832 return (EINVAL); 833 834 /* Check the validity of the signal number. */ 835 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 836 return (EINVAL); 837 838 if ((flags & RFPROCDESC) != 0) { 839 /* Can't not create a process yet get a process descriptor. */ 840 if ((flags & RFPROC) == 0) 841 return (EINVAL); 842 843 /* Must provide a place to put a procdesc if creating one. */ 844 if (fr->fr_pd_fd == NULL) 845 return (EINVAL); 846 847 /* Check if we are using supported flags. */ 848 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 849 return (EINVAL); 850 } 851 852 p1 = td->td_proc; 853 854 /* 855 * Here we don't create a new process, but we divorce 856 * certain parts of a process from itself. 857 */ 858 if ((flags & RFPROC) == 0) { 859 if (fr->fr_procp != NULL) 860 *fr->fr_procp = NULL; 861 else if (fr->fr_pidp != NULL) 862 *fr->fr_pidp = 0; 863 return (fork_norfproc(td, flags)); 864 } 865 866 fp_procdesc = NULL; 867 newproc = NULL; 868 vm2 = NULL; 869 870 /* 871 * Increment the nprocs resource before allocations occur. 872 * Although process entries are dynamically created, we still 873 * keep a global limit on the maximum number we will 874 * create. There are hard-limits as to the number of processes 875 * that can run, established by the KVA and memory usage for 876 * the process data. 877 * 878 * Don't allow a nonprivileged user to use the last ten 879 * processes; don't let root exceed the limit. 880 */ 881 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 882 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, 883 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { 884 error = EAGAIN; 885 sx_xlock(&allproc_lock); 886 if (ppsratecheck(&lastfail, &curfail, 1)) { 887 printf("maxproc limit exceeded by uid %u (pid %d); " 888 "see tuning(7) and login.conf(5)\n", 889 td->td_ucred->cr_ruid, p1->p_pid); 890 } 891 sx_xunlock(&allproc_lock); 892 goto fail2; 893 } 894 895 /* 896 * If required, create a process descriptor in the parent first; we 897 * will abandon it if something goes wrong. We don't finit() until 898 * later. 899 */ 900 if (flags & RFPROCDESC) { 901 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 902 fr->fr_pd_flags, fr->fr_pd_fcaps); 903 if (error != 0) 904 goto fail2; 905 } 906 907 mem_charged = 0; 908 if (pages == 0) 909 pages = kstack_pages; 910 /* Allocate new proc. */ 911 newproc = uma_zalloc(proc_zone, M_WAITOK); 912 td2 = FIRST_THREAD_IN_PROC(newproc); 913 if (td2 == NULL) { 914 td2 = thread_alloc(pages); 915 if (td2 == NULL) { 916 error = ENOMEM; 917 goto fail2; 918 } 919 proc_linkup(newproc, td2); 920 } else { 921 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 922 if (td2->td_kstack != 0) 923 vm_thread_dispose(td2); 924 if (!thread_alloc_stack(td2, pages)) { 925 error = ENOMEM; 926 goto fail2; 927 } 928 } 929 } 930 931 if ((flags & RFMEM) == 0) { 932 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 933 if (vm2 == NULL) { 934 error = ENOMEM; 935 goto fail2; 936 } 937 if (!swap_reserve(mem_charged)) { 938 /* 939 * The swap reservation failed. The accounting 940 * from the entries of the copied vm2 will be 941 * subtracted in vmspace_free(), so force the 942 * reservation there. 943 */ 944 swap_reserve_force(mem_charged); 945 error = ENOMEM; 946 goto fail2; 947 } 948 } else 949 vm2 = NULL; 950 951 /* 952 * XXX: This is ugly; when we copy resource usage, we need to bump 953 * per-cred resource counters. 954 */ 955 proc_set_cred_init(newproc, crhold(td->td_ucred)); 956 957 /* 958 * Initialize resource accounting for the child process. 959 */ 960 error = racct_proc_fork(p1, newproc); 961 if (error != 0) { 962 error = EAGAIN; 963 goto fail1; 964 } 965 966#ifdef MAC 967 mac_proc_init(newproc); 968#endif 969 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 970 STAILQ_INIT(&newproc->p_ktr); 971 972 /* We have to lock the process tree while we look for a pid. */ 973 sx_slock(&proctree_lock); 974 sx_xlock(&allproc_lock); 975 976 /* 977 * Increment the count of procs running with this uid. Don't allow 978 * a nonprivileged user to exceed their current limit. 979 * 980 * XXXRW: Can we avoid privilege here if it's not needed? 981 */ 982 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 983 if (error == 0) 984 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 985 else { 986 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 987 lim_cur(td, RLIMIT_NPROC)); 988 } 989 if (ok) { 990 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 991 return (0); 992 } 993 994 error = EAGAIN; 995 sx_sunlock(&proctree_lock); 996 sx_xunlock(&allproc_lock); 997#ifdef MAC 998 mac_proc_destroy(newproc); 999#endif 1000 racct_proc_exit(newproc); 1001fail1: 1002 crfree(newproc->p_ucred); 1003 newproc->p_ucred = NULL; 1004fail2: 1005 if (vm2 != NULL) 1006 vmspace_free(vm2); 1007 uma_zfree(proc_zone, newproc); 1008 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1009 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1010 fdrop(fp_procdesc, td); 1011 } 1012 atomic_add_int(&nprocs, -1); 1013 pause("fork", hz / 2); 1014 return (error); 1015} 1016 1017/* 1018 * Handle the return of a child process from fork1(). This function 1019 * is called from the MD fork_trampoline() entry point. 1020 */ 1021void 1022fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1023 struct trapframe *frame) 1024{ 1025 struct proc *p; 1026 struct thread *td; 1027 struct thread *dtd; 1028 1029 td = curthread; 1030 p = td->td_proc; 1031 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1032 1033 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1034 td, td_get_sched(td), p->p_pid, td->td_name); 1035 1036 sched_fork_exit(td); 1037 /* 1038 * Processes normally resume in mi_switch() after being 1039 * cpu_switch()'ed to, but when children start up they arrive here 1040 * instead, so we must do much the same things as mi_switch() would. 1041 */ 1042 if ((dtd = PCPU_GET(deadthread))) { 1043 PCPU_SET(deadthread, NULL); 1044 thread_stash(dtd); 1045 } 1046 thread_unlock(td); 1047 1048 /* 1049 * cpu_fork_kthread_handler intercepts this function call to 1050 * have this call a non-return function to stay in kernel mode. 1051 * initproc has its own fork handler, but it does return. 1052 */ 1053 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1054 callout(arg, frame); 1055 1056 /* 1057 * Check if a kernel thread misbehaved and returned from its main 1058 * function. 1059 */ 1060 if (p->p_flag & P_KPROC) { 1061 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1062 td->td_name, p->p_pid); 1063 kthread_exit(); 1064 } 1065 mtx_assert(&Giant, MA_NOTOWNED); 1066 1067 if (p->p_sysent->sv_schedtail != NULL) 1068 (p->p_sysent->sv_schedtail)(td); 1069 td->td_pflags &= ~TDP_FORKING; 1070} 1071 1072/* 1073 * Simplified back end of syscall(), used when returning from fork() 1074 * directly into user mode. This function is passed in to fork_exit() 1075 * as the first parameter and is called when returning to a new 1076 * userland process. 1077 */ 1078void 1079fork_return(struct thread *td, struct trapframe *frame) 1080{ 1081 struct proc *p, *dbg; 1082 1083 p = td->td_proc; 1084 if (td->td_dbgflags & TDB_STOPATFORK) { 1085 sx_xlock(&proctree_lock); 1086 PROC_LOCK(p); 1087 if (p->p_pptr->p_ptevents & PTRACE_FORK) { 1088 /* 1089 * If debugger still wants auto-attach for the 1090 * parent's children, do it now. 1091 */ 1092 dbg = p->p_pptr->p_pptr; 1093 proc_set_traced(p, true); 1094 CTR2(KTR_PTRACE, 1095 "fork_return: attaching to new child pid %d: oppid %d", 1096 p->p_pid, p->p_oppid); 1097 proc_reparent(p, dbg); 1098 sx_xunlock(&proctree_lock); 1099 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1100 ptracestop(td, SIGSTOP, NULL); 1101 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1102 } else { 1103 /* 1104 * ... otherwise clear the request. 1105 */ 1106 sx_xunlock(&proctree_lock); 1107 td->td_dbgflags &= ~TDB_STOPATFORK; 1108 cv_broadcast(&p->p_dbgwait); 1109 } 1110 PROC_UNLOCK(p); 1111 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1112 /* 1113 * This is the start of a new thread in a traced 1114 * process. Report a system call exit event. 1115 */ 1116 PROC_LOCK(p); 1117 td->td_dbgflags |= TDB_SCX; 1118 _STOPEVENT(p, S_SCX, td->td_sa.code); 1119 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1120 (td->td_dbgflags & TDB_BORN) != 0) 1121 ptracestop(td, SIGTRAP, NULL); 1122 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1123 PROC_UNLOCK(p); 1124 } 1125 1126 userret(td, frame); 1127 1128#ifdef KTRACE 1129 if (KTRPOINT(td, KTR_SYSRET)) 1130 ktrsysret(SYS_fork, 0, 0); 1131#endif 1132} 1133