1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ktrace.h" 43 #include "opt_kstack_pages.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/bitstring.h> 48 #include <sys/sysproto.h> 49 #include <sys/eventhandler.h> 50 #include <sys/fcntl.h> 51 #include <sys/filedesc.h> 52 #include <sys/jail.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/sysctl.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mutex.h> 59 #include <sys/priv.h> 60 #include <sys/proc.h> 61 #include <sys/procdesc.h> 62 #include <sys/ptrace.h> 63 #include <sys/racct.h> 64 #include <sys/resourcevar.h> 65 #include <sys/sched.h> 66 #include <sys/syscall.h> 67 #include <sys/vmmeter.h> 68 #include <sys/vnode.h> 69 #include <sys/acct.h> 70 #include <sys/ktr.h> 71 #include <sys/ktrace.h> 72 #include <sys/unistd.h> 73 #include <sys/sdt.h> 74 #include <sys/sx.h> 75 #include <sys/sysent.h> 76 #include <sys/signalvar.h> 77 78 #include <security/audit/audit.h> 79 #include <security/mac/mac_framework.h> 80 81 #include <vm/vm.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_extern.h> 85 #include <vm/uma.h> 86 87 #ifdef KDTRACE_HOOKS 88 #include <sys/dtrace_bsd.h> 89 dtrace_fork_func_t dtrace_fasttrap_fork; 90 #endif 91 92 SDT_PROVIDER_DECLARE(proc); 93 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); 94 95 #ifndef _SYS_SYSPROTO_H_ 96 struct fork_args { 97 int dummy; 98 }; 99 #endif 100 101 /* ARGSUSED */ 102 int 103 sys_fork(struct thread *td, struct fork_args *uap) 104 { 105 struct fork_req fr; 106 int error, pid; 107 108 bzero(&fr, sizeof(fr)); 109 fr.fr_flags = RFFDG | RFPROC; 110 fr.fr_pidp = &pid; 111 error = fork1(td, &fr); 112 if (error == 0) { 113 td->td_retval[0] = pid; 114 td->td_retval[1] = 0; 115 } 116 return (error); 117 } 118 119 /* ARGUSED */ 120 int 121 sys_pdfork(struct thread *td, struct pdfork_args *uap) 122 { 123 struct fork_req fr; 124 int error, fd, pid; 125 126 bzero(&fr, sizeof(fr)); 127 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC; 128 fr.fr_pidp = &pid; 129 fr.fr_pd_fd = &fd; 130 fr.fr_pd_flags = uap->flags; 131 AUDIT_ARG_FFLAGS(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 */ 147 int 148 sys_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 } 163 164 int 165 sys_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 } 193 194 int __exclusive_cache_line nprocs = 1; /* process 0 */ 195 int lastpid = 0; 196 SYSCTL_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 */ 207 static int randompid = 0; 208 209 static int 210 sysctl_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 } 238 239 SYSCTL_PROC(_kern, OID_AUTO, randompid, 240 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, 241 sysctl_kern_randompid, "I", 242 "Random PID modulus. Special values: 0: disable, 1: choose random value"); 243 244 extern bitstr_t proc_id_pidmap; 245 extern bitstr_t proc_id_grpidmap; 246 extern bitstr_t proc_id_sessidmap; 247 extern bitstr_t proc_id_reapmap; 248 249 /* 250 * Find an unused process ID 251 * 252 * If RFHIGHPID is set (used during system boot), do not allocate 253 * low-numbered pids. 254 */ 255 static int 256 fork_findpid(int flags) 257 { 258 pid_t result; 259 int trypid, random; 260 261 /* 262 * Avoid calling arc4random with procid_lock held. 263 */ 264 random = 0; 265 if (__predict_false(randompid)) 266 random = arc4random() % randompid; 267 268 mtx_lock(&procid_lock); 269 270 trypid = lastpid + 1; 271 if (flags & RFHIGHPID) { 272 if (trypid < 10) 273 trypid = 10; 274 } else { 275 trypid += random; 276 } 277 retry: 278 if (trypid >= pid_max) 279 trypid = 2; 280 281 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result); 282 if (result == -1) { 283 KASSERT(trypid != 2, ("unexpectedly ran out of IDs")); 284 trypid = 2; 285 goto retry; 286 } 287 if (bit_test(&proc_id_grpidmap, result) || 288 bit_test(&proc_id_sessidmap, result) || 289 bit_test(&proc_id_reapmap, result)) { 290 trypid = result + 1; 291 goto retry; 292 } 293 294 /* 295 * RFHIGHPID does not mess with the lastpid counter during boot. 296 */ 297 if ((flags & RFHIGHPID) == 0) 298 lastpid = result; 299 300 bit_set(&proc_id_pidmap, result); 301 mtx_unlock(&procid_lock); 302 303 return (result); 304 } 305 306 static int 307 fork_norfproc(struct thread *td, int flags) 308 { 309 int error; 310 struct proc *p1; 311 312 KASSERT((flags & RFPROC) == 0, 313 ("fork_norfproc called with RFPROC set")); 314 p1 = td->td_proc; 315 316 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 317 (flags & (RFCFDG | RFFDG))) { 318 PROC_LOCK(p1); 319 if (thread_single(p1, SINGLE_BOUNDARY)) { 320 PROC_UNLOCK(p1); 321 return (ERESTART); 322 } 323 PROC_UNLOCK(p1); 324 } 325 326 error = vm_forkproc(td, NULL, NULL, NULL, flags); 327 if (error) 328 goto fail; 329 330 /* 331 * Close all file descriptors. 332 */ 333 if (flags & RFCFDG) { 334 struct filedesc *fdtmp; 335 fdtmp = fdinit(td->td_proc->p_fd, false, NULL); 336 fdescfree(td); 337 p1->p_fd = fdtmp; 338 } 339 340 /* 341 * Unshare file descriptors (from parent). 342 */ 343 if (flags & RFFDG) 344 fdunshare(td); 345 346 fail: 347 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 348 (flags & (RFCFDG | RFFDG))) { 349 PROC_LOCK(p1); 350 thread_single_end(p1, SINGLE_BOUNDARY); 351 PROC_UNLOCK(p1); 352 } 353 return (error); 354 } 355 356 static void 357 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 358 struct vmspace *vm2, struct file *fp_procdesc) 359 { 360 struct proc *p1, *pptr; 361 struct filedesc *fd; 362 struct filedesc_to_leader *fdtol; 363 struct sigacts *newsigacts; 364 365 p1 = td->td_proc; 366 367 PROC_LOCK(p1); 368 bcopy(&p1->p_startcopy, &p2->p_startcopy, 369 __rangeof(struct proc, p_startcopy, p_endcopy)); 370 pargs_hold(p2->p_args); 371 PROC_UNLOCK(p1); 372 373 bzero(&p2->p_startzero, 374 __rangeof(struct proc, p_startzero, p_endzero)); 375 376 /* Tell the prison that we exist. */ 377 prison_proc_hold(p2->p_ucred->cr_prison); 378 379 p2->p_state = PRS_NEW; /* protect against others */ 380 p2->p_pid = fork_findpid(fr->fr_flags); 381 AUDIT_ARG_PID(p2->p_pid); 382 383 sx_xlock(&allproc_lock); 384 LIST_INSERT_HEAD(&allproc, p2, p_list); 385 allproc_gen++; 386 sx_xunlock(&allproc_lock); 387 388 sx_xlock(PIDHASHLOCK(p2->p_pid)); 389 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 390 sx_xunlock(PIDHASHLOCK(p2->p_pid)); 391 392 tidhash_add(td2); 393 394 /* 395 * Malloc things while we don't hold any locks. 396 */ 397 if (fr->fr_flags & RFSIGSHARE) 398 newsigacts = NULL; 399 else 400 newsigacts = sigacts_alloc(); 401 402 /* 403 * Copy filedesc. 404 */ 405 if (fr->fr_flags & RFCFDG) { 406 fd = fdinit(p1->p_fd, false, NULL); 407 fdtol = NULL; 408 } else if (fr->fr_flags & RFFDG) { 409 fd = fdcopy(p1->p_fd); 410 fdtol = NULL; 411 } else { 412 fd = fdshare(p1->p_fd); 413 if (p1->p_fdtol == NULL) 414 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 415 p1->p_leader); 416 if ((fr->fr_flags & RFTHREAD) != 0) { 417 /* 418 * Shared file descriptor table, and shared 419 * process leaders. 420 */ 421 fdtol = p1->p_fdtol; 422 FILEDESC_XLOCK(p1->p_fd); 423 fdtol->fdl_refcount++; 424 FILEDESC_XUNLOCK(p1->p_fd); 425 } else { 426 /* 427 * Shared file descriptor table, and different 428 * process leaders. 429 */ 430 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 431 p1->p_fd, p2); 432 } 433 } 434 /* 435 * Make a proc table entry for the new process. 436 * Start by zeroing the section of proc that is zero-initialized, 437 * then copy the section that is copied directly from the parent. 438 */ 439 440 PROC_LOCK(p2); 441 PROC_LOCK(p1); 442 443 bzero(&td2->td_startzero, 444 __rangeof(struct thread, td_startzero, td_endzero)); 445 446 bcopy(&td->td_startcopy, &td2->td_startcopy, 447 __rangeof(struct thread, td_startcopy, td_endcopy)); 448 449 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 450 td2->td_sigstk = td->td_sigstk; 451 td2->td_flags = TDF_INMEM; 452 td2->td_lend_user_pri = PRI_MAX; 453 454 #ifdef VIMAGE 455 td2->td_vnet = NULL; 456 td2->td_vnet_lpush = NULL; 457 #endif 458 459 /* 460 * Allow the scheduler to initialize the child. 461 */ 462 thread_lock(td); 463 sched_fork(td, td2); 464 thread_unlock(td); 465 466 /* 467 * Duplicate sub-structures as needed. 468 * Increase reference counts on shared objects. 469 */ 470 p2->p_flag = P_INMEM; 471 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 472 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 473 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP | 474 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC); 475 p2->p_swtick = ticks; 476 if (p1->p_flag & P_PROFIL) 477 startprofclock(p2); 478 479 if (fr->fr_flags & RFSIGSHARE) { 480 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 481 } else { 482 sigacts_copy(newsigacts, p1->p_sigacts); 483 p2->p_sigacts = newsigacts; 484 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) { 485 mtx_lock(&p2->p_sigacts->ps_mtx); 486 sig_drop_caught(p2); 487 mtx_unlock(&p2->p_sigacts->ps_mtx); 488 } 489 } 490 491 if (fr->fr_flags & RFTSIGZMB) 492 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 493 else if (fr->fr_flags & RFLINUXTHPN) 494 p2->p_sigparent = SIGUSR1; 495 else 496 p2->p_sigparent = SIGCHLD; 497 498 p2->p_textvp = p1->p_textvp; 499 p2->p_fd = fd; 500 p2->p_fdtol = fdtol; 501 502 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 503 p2->p_flag |= P_PROTECTED; 504 p2->p_flag2 |= P2_INHERIT_PROTECTED; 505 } 506 507 /* 508 * p_limit is copy-on-write. Bump its refcount. 509 */ 510 lim_fork(p1, p2); 511 512 thread_cow_get_proc(td2, p2); 513 514 pstats_fork(p1->p_stats, p2->p_stats); 515 516 PROC_UNLOCK(p1); 517 PROC_UNLOCK(p2); 518 519 /* Bump references to the text vnode (for procfs). */ 520 if (p2->p_textvp) 521 vrefact(p2->p_textvp); 522 523 /* 524 * Set up linkage for kernel based threading. 525 */ 526 if ((fr->fr_flags & RFTHREAD) != 0) { 527 mtx_lock(&ppeers_lock); 528 p2->p_peers = p1->p_peers; 529 p1->p_peers = p2; 530 p2->p_leader = p1->p_leader; 531 mtx_unlock(&ppeers_lock); 532 PROC_LOCK(p1->p_leader); 533 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 534 PROC_UNLOCK(p1->p_leader); 535 /* 536 * The task leader is exiting, so process p1 is 537 * going to be killed shortly. Since p1 obviously 538 * isn't dead yet, we know that the leader is either 539 * sending SIGKILL's to all the processes in this 540 * task or is sleeping waiting for all the peers to 541 * exit. We let p1 complete the fork, but we need 542 * to go ahead and kill the new process p2 since 543 * the task leader may not get a chance to send 544 * SIGKILL to it. We leave it on the list so that 545 * the task leader will wait for this new process 546 * to commit suicide. 547 */ 548 PROC_LOCK(p2); 549 kern_psignal(p2, SIGKILL); 550 PROC_UNLOCK(p2); 551 } else 552 PROC_UNLOCK(p1->p_leader); 553 } else { 554 p2->p_peers = NULL; 555 p2->p_leader = p2; 556 } 557 558 sx_xlock(&proctree_lock); 559 PGRP_LOCK(p1->p_pgrp); 560 PROC_LOCK(p2); 561 PROC_LOCK(p1); 562 563 /* 564 * Preserve some more flags in subprocess. P_PROFIL has already 565 * been preserved. 566 */ 567 p2->p_flag |= p1->p_flag & P_SUGID; 568 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | 569 TDP_SIGFASTBLOCK)) | TDP_FORKING; 570 SESS_LOCK(p1->p_session); 571 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 572 p2->p_flag |= P_CONTROLT; 573 SESS_UNLOCK(p1->p_session); 574 if (fr->fr_flags & RFPPWAIT) 575 p2->p_flag |= P_PPWAIT; 576 577 p2->p_pgrp = p1->p_pgrp; 578 LIST_INSERT_AFTER(p1, p2, p_pglist); 579 PGRP_UNLOCK(p1->p_pgrp); 580 LIST_INIT(&p2->p_children); 581 LIST_INIT(&p2->p_orphans); 582 583 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 584 585 /* 586 * This begins the section where we must prevent the parent 587 * from being swapped. 588 */ 589 _PHOLD(p1); 590 PROC_UNLOCK(p1); 591 592 /* 593 * Attach the new process to its parent. 594 * 595 * If RFNOWAIT is set, the newly created process becomes a child 596 * of init. This effectively disassociates the child from the 597 * parent. 598 */ 599 if ((fr->fr_flags & RFNOWAIT) != 0) { 600 pptr = p1->p_reaper; 601 p2->p_reaper = pptr; 602 } else { 603 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 604 p1 : p1->p_reaper; 605 pptr = p1; 606 } 607 p2->p_pptr = pptr; 608 p2->p_oppid = pptr->p_pid; 609 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 610 LIST_INIT(&p2->p_reaplist); 611 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 612 if (p2->p_reaper == p1 && p1 != initproc) { 613 p2->p_reapsubtree = p2->p_pid; 614 proc_id_set_cond(PROC_ID_REAP, p2->p_pid); 615 } 616 sx_xunlock(&proctree_lock); 617 618 /* Inform accounting that we have forked. */ 619 p2->p_acflag = AFORK; 620 PROC_UNLOCK(p2); 621 622 #ifdef KTRACE 623 ktrprocfork(p1, p2); 624 #endif 625 626 /* 627 * Finish creating the child process. It will return via a different 628 * execution path later. (ie: directly into user mode) 629 */ 630 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 631 632 if (fr->fr_flags == (RFFDG | RFPROC)) { 633 VM_CNT_INC(v_forks); 634 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 635 p2->p_vmspace->vm_ssize); 636 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 637 VM_CNT_INC(v_vforks); 638 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 639 p2->p_vmspace->vm_ssize); 640 } else if (p1 == &proc0) { 641 VM_CNT_INC(v_kthreads); 642 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 643 p2->p_vmspace->vm_ssize); 644 } else { 645 VM_CNT_INC(v_rforks); 646 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 647 p2->p_vmspace->vm_ssize); 648 } 649 650 /* 651 * Associate the process descriptor with the process before anything 652 * can happen that might cause that process to need the descriptor. 653 * However, don't do this until after fork(2) can no longer fail. 654 */ 655 if (fr->fr_flags & RFPROCDESC) 656 procdesc_new(p2, fr->fr_pd_flags); 657 658 /* 659 * Both processes are set up, now check if any loadable modules want 660 * to adjust anything. 661 */ 662 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 663 664 /* 665 * Set the child start time and mark the process as being complete. 666 */ 667 PROC_LOCK(p2); 668 PROC_LOCK(p1); 669 microuptime(&p2->p_stats->p_start); 670 PROC_SLOCK(p2); 671 p2->p_state = PRS_NORMAL; 672 PROC_SUNLOCK(p2); 673 674 #ifdef KDTRACE_HOOKS 675 /* 676 * Tell the DTrace fasttrap provider about the new process so that any 677 * tracepoints inherited from the parent can be removed. We have to do 678 * this only after p_state is PRS_NORMAL since the fasttrap module will 679 * use pfind() later on. 680 */ 681 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 682 dtrace_fasttrap_fork(p1, p2); 683 #endif 684 if (fr->fr_flags & RFPPWAIT) { 685 td->td_pflags |= TDP_RFPPWAIT; 686 td->td_rfppwait_p = p2; 687 td->td_dbgflags |= TDB_VFORK; 688 } 689 PROC_UNLOCK(p2); 690 691 /* 692 * Tell any interested parties about the new process. 693 */ 694 knote_fork(p1->p_klist, p2->p_pid); 695 696 /* 697 * Now can be swapped. 698 */ 699 _PRELE(p1); 700 PROC_UNLOCK(p1); 701 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 702 703 if (fr->fr_flags & RFPROCDESC) { 704 procdesc_finit(p2->p_procdesc, fp_procdesc); 705 fdrop(fp_procdesc, td); 706 } 707 708 /* 709 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 710 * synced with forks in progress so it is OK if we miss it 711 * if being set atm. 712 */ 713 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 714 sx_xlock(&proctree_lock); 715 PROC_LOCK(p2); 716 717 /* 718 * p1->p_ptevents & p1->p_pptr are protected by both 719 * process and proctree locks for modifications, 720 * so owning proctree_lock allows the race-free read. 721 */ 722 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 723 /* 724 * Arrange for debugger to receive the fork event. 725 * 726 * We can report PL_FLAG_FORKED regardless of 727 * P_FOLLOWFORK settings, but it does not make a sense 728 * for runaway child. 729 */ 730 td->td_dbgflags |= TDB_FORK; 731 td->td_dbg_forked = p2->p_pid; 732 td2->td_dbgflags |= TDB_STOPATFORK; 733 proc_set_traced(p2, true); 734 CTR2(KTR_PTRACE, 735 "do_fork: attaching to new child pid %d: oppid %d", 736 p2->p_pid, p2->p_oppid); 737 proc_reparent(p2, p1->p_pptr, false); 738 } 739 PROC_UNLOCK(p2); 740 sx_xunlock(&proctree_lock); 741 } 742 743 racct_proc_fork_done(p2); 744 745 if ((fr->fr_flags & RFSTOPPED) == 0) { 746 if (fr->fr_pidp != NULL) 747 *fr->fr_pidp = p2->p_pid; 748 /* 749 * If RFSTOPPED not requested, make child runnable and 750 * add to run queue. 751 */ 752 thread_lock(td2); 753 TD_SET_CAN_RUN(td2); 754 sched_add(td2, SRQ_BORING); 755 } else { 756 *fr->fr_procp = p2; 757 } 758 } 759 760 void 761 fork_rfppwait(struct thread *td) 762 { 763 struct proc *p, *p2; 764 765 MPASS(td->td_pflags & TDP_RFPPWAIT); 766 767 p = td->td_proc; 768 /* 769 * Preserve synchronization semantics of vfork. If 770 * waiting for child to exec or exit, fork set 771 * P_PPWAIT on child, and there we sleep on our proc 772 * (in case of exit). 773 * 774 * Do it after the ptracestop() above is finished, to 775 * not block our debugger until child execs or exits 776 * to finish vfork wait. 777 */ 778 td->td_pflags &= ~TDP_RFPPWAIT; 779 p2 = td->td_rfppwait_p; 780 again: 781 PROC_LOCK(p2); 782 while (p2->p_flag & P_PPWAIT) { 783 PROC_LOCK(p); 784 if (thread_suspend_check_needed()) { 785 PROC_UNLOCK(p2); 786 thread_suspend_check(0); 787 PROC_UNLOCK(p); 788 goto again; 789 } else { 790 PROC_UNLOCK(p); 791 } 792 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz); 793 } 794 PROC_UNLOCK(p2); 795 796 if (td->td_dbgflags & TDB_VFORK) { 797 PROC_LOCK(p); 798 if (p->p_ptevents & PTRACE_VFORK) 799 ptracestop(td, SIGTRAP, NULL); 800 td->td_dbgflags &= ~TDB_VFORK; 801 PROC_UNLOCK(p); 802 } 803 } 804 805 int 806 fork1(struct thread *td, struct fork_req *fr) 807 { 808 struct proc *p1, *newproc; 809 struct thread *td2; 810 struct vmspace *vm2; 811 struct ucred *cred; 812 struct file *fp_procdesc; 813 vm_ooffset_t mem_charged; 814 int error, nprocs_new; 815 static int curfail; 816 static struct timeval lastfail; 817 int flags, pages; 818 819 flags = fr->fr_flags; 820 pages = fr->fr_pages; 821 822 if ((flags & RFSTOPPED) != 0) 823 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 824 else 825 MPASS(fr->fr_procp == NULL); 826 827 /* Check for the undefined or unimplemented flags. */ 828 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 829 return (EINVAL); 830 831 /* Signal value requires RFTSIGZMB. */ 832 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 833 return (EINVAL); 834 835 /* Can't copy and clear. */ 836 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 837 return (EINVAL); 838 839 /* Check the validity of the signal number. */ 840 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 841 return (EINVAL); 842 843 if ((flags & RFPROCDESC) != 0) { 844 /* Can't not create a process yet get a process descriptor. */ 845 if ((flags & RFPROC) == 0) 846 return (EINVAL); 847 848 /* Must provide a place to put a procdesc if creating one. */ 849 if (fr->fr_pd_fd == NULL) 850 return (EINVAL); 851 852 /* Check if we are using supported flags. */ 853 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 854 return (EINVAL); 855 } 856 857 p1 = td->td_proc; 858 859 /* 860 * Here we don't create a new process, but we divorce 861 * certain parts of a process from itself. 862 */ 863 if ((flags & RFPROC) == 0) { 864 if (fr->fr_procp != NULL) 865 *fr->fr_procp = NULL; 866 else if (fr->fr_pidp != NULL) 867 *fr->fr_pidp = 0; 868 return (fork_norfproc(td, flags)); 869 } 870 871 fp_procdesc = NULL; 872 newproc = NULL; 873 vm2 = NULL; 874 875 /* 876 * Increment the nprocs resource before allocations occur. 877 * Although process entries are dynamically created, we still 878 * keep a global limit on the maximum number we will 879 * create. There are hard-limits as to the number of processes 880 * that can run, established by the KVA and memory usage for 881 * the process data. 882 * 883 * Don't allow a nonprivileged user to use the last ten 884 * processes; don't let root exceed the limit. 885 */ 886 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 887 if (nprocs_new >= maxproc - 10) { 888 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 || 889 nprocs_new >= maxproc) { 890 error = EAGAIN; 891 sx_xlock(&allproc_lock); 892 if (ppsratecheck(&lastfail, &curfail, 1)) { 893 printf("maxproc limit exceeded by uid %u " 894 "(pid %d); see tuning(7) and " 895 "login.conf(5)\n", 896 td->td_ucred->cr_ruid, p1->p_pid); 897 } 898 sx_xunlock(&allproc_lock); 899 goto fail2; 900 } 901 } 902 903 /* 904 * If required, create a process descriptor in the parent first; we 905 * will abandon it if something goes wrong. We don't finit() until 906 * later. 907 */ 908 if (flags & RFPROCDESC) { 909 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 910 fr->fr_pd_flags, fr->fr_pd_fcaps); 911 if (error != 0) 912 goto fail2; 913 AUDIT_ARG_FD(*fr->fr_pd_fd); 914 } 915 916 mem_charged = 0; 917 if (pages == 0) 918 pages = kstack_pages; 919 /* Allocate new proc. */ 920 newproc = uma_zalloc(proc_zone, M_WAITOK); 921 td2 = FIRST_THREAD_IN_PROC(newproc); 922 if (td2 == NULL) { 923 td2 = thread_alloc(pages); 924 if (td2 == NULL) { 925 error = ENOMEM; 926 goto fail2; 927 } 928 proc_linkup(newproc, td2); 929 } else { 930 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 931 if (td2->td_kstack != 0) 932 vm_thread_dispose(td2); 933 if (!thread_alloc_stack(td2, pages)) { 934 error = ENOMEM; 935 goto fail2; 936 } 937 } 938 } 939 940 if ((flags & RFMEM) == 0) { 941 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 942 if (vm2 == NULL) { 943 error = ENOMEM; 944 goto fail2; 945 } 946 if (!swap_reserve(mem_charged)) { 947 /* 948 * The swap reservation failed. The accounting 949 * from the entries of the copied vm2 will be 950 * subtracted in vmspace_free(), so force the 951 * reservation there. 952 */ 953 swap_reserve_force(mem_charged); 954 error = ENOMEM; 955 goto fail2; 956 } 957 } else 958 vm2 = NULL; 959 960 /* 961 * XXX: This is ugly; when we copy resource usage, we need to bump 962 * per-cred resource counters. 963 */ 964 proc_set_cred_init(newproc, td->td_ucred); 965 966 /* 967 * Initialize resource accounting for the child process. 968 */ 969 error = racct_proc_fork(p1, newproc); 970 if (error != 0) { 971 error = EAGAIN; 972 goto fail1; 973 } 974 975 #ifdef MAC 976 mac_proc_init(newproc); 977 #endif 978 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 979 STAILQ_INIT(&newproc->p_ktr); 980 981 /* 982 * Increment the count of procs running with this uid. Don't allow 983 * a nonprivileged user to exceed their current limit. 984 */ 985 cred = td->td_ucred; 986 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) { 987 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0) 988 goto fail0; 989 chgproccnt(cred->cr_ruidinfo, 1, 0); 990 } 991 992 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 993 return (0); 994 fail0: 995 error = EAGAIN; 996 #ifdef MAC 997 mac_proc_destroy(newproc); 998 #endif 999 racct_proc_exit(newproc); 1000 fail1: 1001 proc_unset_cred(newproc); 1002 fail2: 1003 if (vm2 != NULL) 1004 vmspace_free(vm2); 1005 uma_zfree(proc_zone, newproc); 1006 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1007 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1008 fdrop(fp_procdesc, td); 1009 } 1010 atomic_add_int(&nprocs, -1); 1011 pause("fork", hz / 2); 1012 return (error); 1013 } 1014 1015 /* 1016 * Handle the return of a child process from fork1(). This function 1017 * is called from the MD fork_trampoline() entry point. 1018 */ 1019 void 1020 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1021 struct trapframe *frame) 1022 { 1023 struct proc *p; 1024 struct thread *td; 1025 struct thread *dtd; 1026 1027 td = curthread; 1028 p = td->td_proc; 1029 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1030 1031 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1032 td, td_get_sched(td), p->p_pid, td->td_name); 1033 1034 sched_fork_exit(td); 1035 /* 1036 * Processes normally resume in mi_switch() after being 1037 * cpu_switch()'ed to, but when children start up they arrive here 1038 * instead, so we must do much the same things as mi_switch() would. 1039 */ 1040 if ((dtd = PCPU_GET(deadthread))) { 1041 PCPU_SET(deadthread, NULL); 1042 thread_stash(dtd); 1043 } 1044 thread_unlock(td); 1045 1046 /* 1047 * cpu_fork_kthread_handler intercepts this function call to 1048 * have this call a non-return function to stay in kernel mode. 1049 * initproc has its own fork handler, but it does return. 1050 */ 1051 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1052 callout(arg, frame); 1053 1054 /* 1055 * Check if a kernel thread misbehaved and returned from its main 1056 * function. 1057 */ 1058 if (p->p_flag & P_KPROC) { 1059 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1060 td->td_name, p->p_pid); 1061 kthread_exit(); 1062 } 1063 mtx_assert(&Giant, MA_NOTOWNED); 1064 1065 if (p->p_sysent->sv_schedtail != NULL) 1066 (p->p_sysent->sv_schedtail)(td); 1067 td->td_pflags &= ~TDP_FORKING; 1068 } 1069 1070 /* 1071 * Simplified back end of syscall(), used when returning from fork() 1072 * directly into user mode. This function is passed in to fork_exit() 1073 * as the first parameter and is called when returning to a new 1074 * userland process. 1075 */ 1076 void 1077 fork_return(struct thread *td, struct trapframe *frame) 1078 { 1079 struct proc *p; 1080 1081 p = td->td_proc; 1082 if (td->td_dbgflags & TDB_STOPATFORK) { 1083 PROC_LOCK(p); 1084 if ((p->p_flag & P_TRACED) != 0) { 1085 /* 1086 * Inform the debugger if one is still present. 1087 */ 1088 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1089 ptracestop(td, SIGSTOP, NULL); 1090 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1091 } else { 1092 /* 1093 * ... otherwise clear the request. 1094 */ 1095 td->td_dbgflags &= ~TDB_STOPATFORK; 1096 } 1097 PROC_UNLOCK(p); 1098 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1099 /* 1100 * This is the start of a new thread in a traced 1101 * process. Report a system call exit event. 1102 */ 1103 PROC_LOCK(p); 1104 td->td_dbgflags |= TDB_SCX; 1105 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1106 (td->td_dbgflags & TDB_BORN) != 0) 1107 ptracestop(td, SIGTRAP, NULL); 1108 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1109 PROC_UNLOCK(p); 1110 } 1111 1112 userret(td, frame); 1113 1114 #ifdef KTRACE 1115 if (KTRPOINT(td, KTR_SYSRET)) 1116 ktrsysret(SYS_fork, 0, 0); 1117 #endif 1118 } 1119