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