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