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/msan.h> 59 #include <sys/mutex.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/procdesc.h> 63 #include <sys/ptrace.h> 64 #include <sys/racct.h> 65 #include <sys/resourcevar.h> 66 #include <sys/sched.h> 67 #include <sys/syscall.h> 68 #include <sys/vmmeter.h> 69 #include <sys/vnode.h> 70 #include <sys/acct.h> 71 #include <sys/ktr.h> 72 #include <sys/ktrace.h> 73 #include <sys/unistd.h> 74 #include <sys/sdt.h> 75 #include <sys/sx.h> 76 #include <sys/sysent.h> 77 #include <sys/signalvar.h> 78 79 #include <security/audit/audit.h> 80 #include <security/mac/mac_framework.h> 81 82 #include <vm/vm.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_extern.h> 86 #include <vm/uma.h> 87 88 #ifdef KDTRACE_HOOKS 89 #include <sys/dtrace_bsd.h> 90 dtrace_fork_func_t dtrace_fasttrap_fork; 91 #endif 92 93 SDT_PROVIDER_DECLARE(proc); 94 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); 95 96 #ifndef _SYS_SYSPROTO_H_ 97 struct fork_args { 98 int dummy; 99 }; 100 #endif 101 102 /* ARGSUSED */ 103 int 104 sys_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 */ 121 int 122 sys_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 AUDIT_ARG_FFLAGS(uap->flags); 133 /* 134 * It is necessary to return fd by reference because 0 is a valid file 135 * descriptor number, and the child needs to be able to distinguish 136 * itself from the parent using the return value. 137 */ 138 error = fork1(td, &fr); 139 if (error == 0) { 140 td->td_retval[0] = pid; 141 td->td_retval[1] = 0; 142 error = copyout(&fd, uap->fdp, sizeof(fd)); 143 } 144 return (error); 145 } 146 147 /* ARGSUSED */ 148 int 149 sys_vfork(struct thread *td, struct vfork_args *uap) 150 { 151 struct fork_req fr; 152 int error, pid; 153 154 bzero(&fr, sizeof(fr)); 155 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 156 fr.fr_pidp = &pid; 157 error = fork1(td, &fr); 158 if (error == 0) { 159 td->td_retval[0] = pid; 160 td->td_retval[1] = 0; 161 } 162 return (error); 163 } 164 165 int 166 sys_rfork(struct thread *td, struct rfork_args *uap) 167 { 168 struct fork_req fr; 169 int error, pid; 170 171 /* Don't allow kernel-only flags. */ 172 if ((uap->flags & RFKERNELONLY) != 0) 173 return (EINVAL); 174 /* RFSPAWN must not appear with others */ 175 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN) 176 return (EINVAL); 177 178 AUDIT_ARG_FFLAGS(uap->flags); 179 bzero(&fr, sizeof(fr)); 180 if ((uap->flags & RFSPAWN) != 0) { 181 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 182 fr.fr_flags2 = FR2_DROPSIG_CAUGHT; 183 } else { 184 fr.fr_flags = uap->flags; 185 } 186 fr.fr_pidp = &pid; 187 error = fork1(td, &fr); 188 if (error == 0) { 189 td->td_retval[0] = pid; 190 td->td_retval[1] = 0; 191 } 192 return (error); 193 } 194 195 int __exclusive_cache_line nprocs = 1; /* process 0 */ 196 int lastpid = 0; 197 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 198 "Last used PID"); 199 200 /* 201 * Random component to lastpid generation. We mix in a random factor to make 202 * it a little harder to predict. We sanity check the modulus value to avoid 203 * doing it in critical paths. Don't let it be too small or we pointlessly 204 * waste randomness entropy, and don't let it be impossibly large. Using a 205 * modulus that is too big causes a LOT more process table scans and slows 206 * down fork processing as the pidchecked caching is defeated. 207 */ 208 static int randompid = 0; 209 210 static int 211 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 212 { 213 int error, pid; 214 215 error = sysctl_wire_old_buffer(req, sizeof(int)); 216 if (error != 0) 217 return(error); 218 sx_xlock(&allproc_lock); 219 pid = randompid; 220 error = sysctl_handle_int(oidp, &pid, 0, req); 221 if (error == 0 && req->newptr != NULL) { 222 if (pid == 0) 223 randompid = 0; 224 else if (pid == 1) 225 /* generate a random PID modulus between 100 and 1123 */ 226 randompid = 100 + arc4random() % 1024; 227 else if (pid < 0 || pid > pid_max - 100) 228 /* out of range */ 229 randompid = pid_max - 100; 230 else if (pid < 100) 231 /* Make it reasonable */ 232 randompid = 100; 233 else 234 randompid = pid; 235 } 236 sx_xunlock(&allproc_lock); 237 return (error); 238 } 239 240 SYSCTL_PROC(_kern, OID_AUTO, randompid, 241 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, 242 sysctl_kern_randompid, "I", 243 "Random PID modulus. Special values: 0: disable, 1: choose random value"); 244 245 extern bitstr_t proc_id_pidmap; 246 extern bitstr_t proc_id_grpidmap; 247 extern bitstr_t proc_id_sessidmap; 248 extern bitstr_t proc_id_reapmap; 249 250 /* 251 * Find an unused process ID 252 * 253 * If RFHIGHPID is set (used during system boot), do not allocate 254 * low-numbered pids. 255 */ 256 static int 257 fork_findpid(int flags) 258 { 259 pid_t result; 260 int trypid, random; 261 262 /* 263 * Avoid calling arc4random with procid_lock held. 264 */ 265 random = 0; 266 if (__predict_false(randompid)) 267 random = arc4random() % randompid; 268 269 mtx_lock(&procid_lock); 270 271 trypid = lastpid + 1; 272 if (flags & RFHIGHPID) { 273 if (trypid < 10) 274 trypid = 10; 275 } else { 276 trypid += random; 277 } 278 retry: 279 if (trypid >= pid_max) 280 trypid = 2; 281 282 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result); 283 if (result == -1) { 284 KASSERT(trypid != 2, ("unexpectedly ran out of IDs")); 285 trypid = 2; 286 goto retry; 287 } 288 if (bit_test(&proc_id_grpidmap, result) || 289 bit_test(&proc_id_sessidmap, result) || 290 bit_test(&proc_id_reapmap, result)) { 291 trypid = result + 1; 292 goto retry; 293 } 294 295 /* 296 * RFHIGHPID does not mess with the lastpid counter during boot. 297 */ 298 if ((flags & RFHIGHPID) == 0) 299 lastpid = result; 300 301 bit_set(&proc_id_pidmap, result); 302 mtx_unlock(&procid_lock); 303 304 return (result); 305 } 306 307 static int 308 fork_norfproc(struct thread *td, int flags) 309 { 310 int error; 311 struct proc *p1; 312 313 KASSERT((flags & RFPROC) == 0, 314 ("fork_norfproc called with RFPROC set")); 315 p1 = td->td_proc; 316 317 /* 318 * Quiesce other threads if necessary. If RFMEM is not specified we 319 * must ensure that other threads do not concurrently create a second 320 * process sharing the vmspace, see vmspace_unshare(). 321 */ 322 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS && 323 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) { 324 PROC_LOCK(p1); 325 if (thread_single(p1, SINGLE_BOUNDARY)) { 326 PROC_UNLOCK(p1); 327 return (ERESTART); 328 } 329 PROC_UNLOCK(p1); 330 } 331 332 error = vm_forkproc(td, NULL, NULL, NULL, flags); 333 if (error) 334 goto fail; 335 336 /* 337 * Close all file descriptors. 338 */ 339 if (flags & RFCFDG) { 340 struct filedesc *fdtmp; 341 struct pwddesc *pdtmp; 342 pdtmp = pdinit(td->td_proc->p_pd, false); 343 fdtmp = fdinit(td->td_proc->p_fd, false, NULL); 344 pdescfree(td); 345 fdescfree(td); 346 p1->p_fd = fdtmp; 347 p1->p_pd = pdtmp; 348 } 349 350 /* 351 * Unshare file descriptors (from parent). 352 */ 353 if (flags & RFFDG) { 354 fdunshare(td); 355 pdunshare(td); 356 } 357 358 fail: 359 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS && 360 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) { 361 PROC_LOCK(p1); 362 thread_single_end(p1, SINGLE_BOUNDARY); 363 PROC_UNLOCK(p1); 364 } 365 return (error); 366 } 367 368 static void 369 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 370 struct vmspace *vm2, struct file *fp_procdesc) 371 { 372 struct proc *p1, *pptr; 373 struct filedesc *fd; 374 struct filedesc_to_leader *fdtol; 375 struct pwddesc *pd; 376 struct sigacts *newsigacts; 377 378 p1 = td->td_proc; 379 380 PROC_LOCK(p1); 381 bcopy(&p1->p_startcopy, &p2->p_startcopy, 382 __rangeof(struct proc, p_startcopy, p_endcopy)); 383 pargs_hold(p2->p_args); 384 PROC_UNLOCK(p1); 385 386 bzero(&p2->p_startzero, 387 __rangeof(struct proc, p_startzero, p_endzero)); 388 389 /* Tell the prison that we exist. */ 390 prison_proc_hold(p2->p_ucred->cr_prison); 391 392 p2->p_state = PRS_NEW; /* protect against others */ 393 p2->p_pid = fork_findpid(fr->fr_flags); 394 AUDIT_ARG_PID(p2->p_pid); 395 TSFORK(p2->p_pid, p1->p_pid); 396 397 sx_xlock(&allproc_lock); 398 LIST_INSERT_HEAD(&allproc, p2, p_list); 399 allproc_gen++; 400 sx_xunlock(&allproc_lock); 401 402 sx_xlock(PIDHASHLOCK(p2->p_pid)); 403 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 404 sx_xunlock(PIDHASHLOCK(p2->p_pid)); 405 406 tidhash_add(td2); 407 408 /* 409 * Malloc things while we don't hold any locks. 410 */ 411 if (fr->fr_flags & RFSIGSHARE) 412 newsigacts = NULL; 413 else 414 newsigacts = sigacts_alloc(); 415 416 /* 417 * Copy filedesc. 418 */ 419 if (fr->fr_flags & RFCFDG) { 420 pd = pdinit(p1->p_pd, false); 421 fd = fdinit(p1->p_fd, false, NULL); 422 fdtol = NULL; 423 } else if (fr->fr_flags & RFFDG) { 424 if (fr->fr_flags2 & FR2_SHARE_PATHS) 425 pd = pdshare(p1->p_pd); 426 else 427 pd = pdcopy(p1->p_pd); 428 fd = fdcopy(p1->p_fd); 429 fdtol = NULL; 430 } else { 431 if (fr->fr_flags2 & FR2_SHARE_PATHS) 432 pd = pdcopy(p1->p_pd); 433 else 434 pd = pdshare(p1->p_pd); 435 fd = fdshare(p1->p_fd); 436 if (p1->p_fdtol == NULL) 437 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 438 p1->p_leader); 439 if ((fr->fr_flags & RFTHREAD) != 0) { 440 /* 441 * Shared file descriptor table, and shared 442 * process leaders. 443 */ 444 fdtol = p1->p_fdtol; 445 FILEDESC_XLOCK(p1->p_fd); 446 fdtol->fdl_refcount++; 447 FILEDESC_XUNLOCK(p1->p_fd); 448 } else { 449 /* 450 * Shared file descriptor table, and different 451 * process leaders. 452 */ 453 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 454 p1->p_fd, p2); 455 } 456 } 457 /* 458 * Make a proc table entry for the new process. 459 * Start by zeroing the section of proc that is zero-initialized, 460 * then copy the section that is copied directly from the parent. 461 */ 462 463 PROC_LOCK(p2); 464 PROC_LOCK(p1); 465 466 bzero(&td2->td_startzero, 467 __rangeof(struct thread, td_startzero, td_endzero)); 468 469 bcopy(&td->td_startcopy, &td2->td_startcopy, 470 __rangeof(struct thread, td_startcopy, td_endcopy)); 471 472 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 473 td2->td_sigstk = td->td_sigstk; 474 td2->td_flags = TDF_INMEM; 475 td2->td_lend_user_pri = PRI_MAX; 476 477 #ifdef VIMAGE 478 td2->td_vnet = NULL; 479 td2->td_vnet_lpush = NULL; 480 #endif 481 482 /* 483 * Allow the scheduler to initialize the child. 484 */ 485 thread_lock(td); 486 sched_fork(td, td2); 487 /* 488 * Request AST to check for TDP_RFPPWAIT. Do it here 489 * to avoid calling thread_lock() again. 490 */ 491 if ((fr->fr_flags & RFPPWAIT) != 0) 492 td->td_flags |= TDF_ASTPENDING; 493 thread_unlock(td); 494 495 /* 496 * Duplicate sub-structures as needed. 497 * Increase reference counts on shared objects. 498 */ 499 p2->p_flag = P_INMEM; 500 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 501 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 502 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP | 503 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS | 504 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 505 p2->p_swtick = ticks; 506 if (p1->p_flag & P_PROFIL) 507 startprofclock(p2); 508 509 if (fr->fr_flags & RFSIGSHARE) { 510 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 511 } else { 512 sigacts_copy(newsigacts, p1->p_sigacts); 513 p2->p_sigacts = newsigacts; 514 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) { 515 mtx_lock(&p2->p_sigacts->ps_mtx); 516 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) 517 sig_drop_caught(p2); 518 if ((fr->fr_flags2 & FR2_KPROC) != 0) 519 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT; 520 mtx_unlock(&p2->p_sigacts->ps_mtx); 521 } 522 } 523 524 if (fr->fr_flags & RFTSIGZMB) 525 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 526 else if (fr->fr_flags & RFLINUXTHPN) 527 p2->p_sigparent = SIGUSR1; 528 else 529 p2->p_sigparent = SIGCHLD; 530 531 if ((fr->fr_flags2 & FR2_KPROC) != 0) { 532 p2->p_flag |= P_SYSTEM | P_KPROC; 533 td2->td_pflags |= TDP_KTHREAD; 534 } 535 536 p2->p_textvp = p1->p_textvp; 537 p2->p_textdvp = p1->p_textdvp; 538 p2->p_fd = fd; 539 p2->p_fdtol = fdtol; 540 p2->p_pd = pd; 541 542 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 543 p2->p_flag |= P_PROTECTED; 544 p2->p_flag2 |= P2_INHERIT_PROTECTED; 545 } 546 547 /* 548 * p_limit is copy-on-write. Bump its refcount. 549 */ 550 lim_fork(p1, p2); 551 552 thread_cow_get_proc(td2, p2); 553 554 pstats_fork(p1->p_stats, p2->p_stats); 555 556 PROC_UNLOCK(p1); 557 PROC_UNLOCK(p2); 558 559 /* 560 * Bump references to the text vnode and directory, and copy 561 * the hardlink name. 562 */ 563 if (p2->p_textvp != NULL) 564 vrefact(p2->p_textvp); 565 if (p2->p_textdvp != NULL) 566 vrefact(p2->p_textdvp); 567 p2->p_binname = p1->p_binname == NULL ? NULL : 568 strdup(p1->p_binname, M_PARGS); 569 570 /* 571 * Set up linkage for kernel based threading. 572 */ 573 if ((fr->fr_flags & RFTHREAD) != 0) { 574 mtx_lock(&ppeers_lock); 575 p2->p_peers = p1->p_peers; 576 p1->p_peers = p2; 577 p2->p_leader = p1->p_leader; 578 mtx_unlock(&ppeers_lock); 579 PROC_LOCK(p1->p_leader); 580 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 581 PROC_UNLOCK(p1->p_leader); 582 /* 583 * The task leader is exiting, so process p1 is 584 * going to be killed shortly. Since p1 obviously 585 * isn't dead yet, we know that the leader is either 586 * sending SIGKILL's to all the processes in this 587 * task or is sleeping waiting for all the peers to 588 * exit. We let p1 complete the fork, but we need 589 * to go ahead and kill the new process p2 since 590 * the task leader may not get a chance to send 591 * SIGKILL to it. We leave it on the list so that 592 * the task leader will wait for this new process 593 * to commit suicide. 594 */ 595 PROC_LOCK(p2); 596 kern_psignal(p2, SIGKILL); 597 PROC_UNLOCK(p2); 598 } else 599 PROC_UNLOCK(p1->p_leader); 600 } else { 601 p2->p_peers = NULL; 602 p2->p_leader = p2; 603 } 604 605 sx_xlock(&proctree_lock); 606 PGRP_LOCK(p1->p_pgrp); 607 PROC_LOCK(p2); 608 PROC_LOCK(p1); 609 610 /* 611 * Preserve some more flags in subprocess. P_PROFIL has already 612 * been preserved. 613 */ 614 p2->p_flag |= p1->p_flag & P_SUGID; 615 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK)); 616 SESS_LOCK(p1->p_session); 617 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 618 p2->p_flag |= P_CONTROLT; 619 SESS_UNLOCK(p1->p_session); 620 if (fr->fr_flags & RFPPWAIT) 621 p2->p_flag |= P_PPWAIT; 622 623 p2->p_pgrp = p1->p_pgrp; 624 LIST_INSERT_AFTER(p1, p2, p_pglist); 625 PGRP_UNLOCK(p1->p_pgrp); 626 LIST_INIT(&p2->p_children); 627 LIST_INIT(&p2->p_orphans); 628 629 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 630 TAILQ_INIT(&p2->p_kqtim_stop); 631 632 /* 633 * This begins the section where we must prevent the parent 634 * from being swapped. 635 */ 636 _PHOLD(p1); 637 PROC_UNLOCK(p1); 638 639 /* 640 * Attach the new process to its parent. 641 * 642 * If RFNOWAIT is set, the newly created process becomes a child 643 * of init. This effectively disassociates the child from the 644 * parent. 645 */ 646 if ((fr->fr_flags & RFNOWAIT) != 0) { 647 pptr = p1->p_reaper; 648 p2->p_reaper = pptr; 649 } else { 650 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 651 p1 : p1->p_reaper; 652 pptr = p1; 653 } 654 p2->p_pptr = pptr; 655 p2->p_oppid = pptr->p_pid; 656 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 657 LIST_INIT(&p2->p_reaplist); 658 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 659 if (p2->p_reaper == p1 && p1 != initproc) { 660 p2->p_reapsubtree = p2->p_pid; 661 proc_id_set_cond(PROC_ID_REAP, p2->p_pid); 662 } 663 sx_xunlock(&proctree_lock); 664 665 /* Inform accounting that we have forked. */ 666 p2->p_acflag = AFORK; 667 PROC_UNLOCK(p2); 668 669 #ifdef KTRACE 670 ktrprocfork(p1, p2); 671 #endif 672 673 /* 674 * Finish creating the child process. It will return via a different 675 * execution path later. (ie: directly into user mode) 676 */ 677 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 678 679 if (fr->fr_flags == (RFFDG | RFPROC)) { 680 VM_CNT_INC(v_forks); 681 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 682 p2->p_vmspace->vm_ssize); 683 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 684 VM_CNT_INC(v_vforks); 685 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 686 p2->p_vmspace->vm_ssize); 687 } else if (p1 == &proc0) { 688 VM_CNT_INC(v_kthreads); 689 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 690 p2->p_vmspace->vm_ssize); 691 } else { 692 VM_CNT_INC(v_rforks); 693 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 694 p2->p_vmspace->vm_ssize); 695 } 696 697 /* 698 * Associate the process descriptor with the process before anything 699 * can happen that might cause that process to need the descriptor. 700 * However, don't do this until after fork(2) can no longer fail. 701 */ 702 if (fr->fr_flags & RFPROCDESC) 703 procdesc_new(p2, fr->fr_pd_flags); 704 705 /* 706 * Both processes are set up, now check if any loadable modules want 707 * to adjust anything. 708 */ 709 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 710 711 /* 712 * Set the child start time and mark the process as being complete. 713 */ 714 PROC_LOCK(p2); 715 PROC_LOCK(p1); 716 microuptime(&p2->p_stats->p_start); 717 PROC_SLOCK(p2); 718 p2->p_state = PRS_NORMAL; 719 PROC_SUNLOCK(p2); 720 721 #ifdef KDTRACE_HOOKS 722 /* 723 * Tell the DTrace fasttrap provider about the new process so that any 724 * tracepoints inherited from the parent can be removed. We have to do 725 * this only after p_state is PRS_NORMAL since the fasttrap module will 726 * use pfind() later on. 727 */ 728 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 729 dtrace_fasttrap_fork(p1, p2); 730 #endif 731 if (fr->fr_flags & RFPPWAIT) { 732 td->td_pflags |= TDP_RFPPWAIT; 733 td->td_rfppwait_p = p2; 734 td->td_dbgflags |= TDB_VFORK; 735 } 736 PROC_UNLOCK(p2); 737 738 /* 739 * Tell any interested parties about the new process. 740 */ 741 knote_fork(p1->p_klist, p2->p_pid); 742 743 /* 744 * Now can be swapped. 745 */ 746 _PRELE(p1); 747 PROC_UNLOCK(p1); 748 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 749 750 if (fr->fr_flags & RFPROCDESC) { 751 procdesc_finit(p2->p_procdesc, fp_procdesc); 752 fdrop(fp_procdesc, td); 753 } 754 755 /* 756 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 757 * synced with forks in progress so it is OK if we miss it 758 * if being set atm. 759 */ 760 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 761 sx_xlock(&proctree_lock); 762 PROC_LOCK(p2); 763 764 /* 765 * p1->p_ptevents & p1->p_pptr are protected by both 766 * process and proctree locks for modifications, 767 * so owning proctree_lock allows the race-free read. 768 */ 769 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 770 /* 771 * Arrange for debugger to receive the fork event. 772 * 773 * We can report PL_FLAG_FORKED regardless of 774 * P_FOLLOWFORK settings, but it does not make a sense 775 * for runaway child. 776 */ 777 td->td_dbgflags |= TDB_FORK; 778 td->td_dbg_forked = p2->p_pid; 779 td2->td_dbgflags |= TDB_STOPATFORK; 780 proc_set_traced(p2, true); 781 CTR2(KTR_PTRACE, 782 "do_fork: attaching to new child pid %d: oppid %d", 783 p2->p_pid, p2->p_oppid); 784 proc_reparent(p2, p1->p_pptr, false); 785 } 786 PROC_UNLOCK(p2); 787 sx_xunlock(&proctree_lock); 788 } 789 790 racct_proc_fork_done(p2); 791 792 if ((fr->fr_flags & RFSTOPPED) == 0) { 793 if (fr->fr_pidp != NULL) 794 *fr->fr_pidp = p2->p_pid; 795 /* 796 * If RFSTOPPED not requested, make child runnable and 797 * add to run queue. 798 */ 799 thread_lock(td2); 800 TD_SET_CAN_RUN(td2); 801 sched_add(td2, SRQ_BORING); 802 } else { 803 *fr->fr_procp = p2; 804 } 805 } 806 807 void 808 fork_rfppwait(struct thread *td) 809 { 810 struct proc *p, *p2; 811 812 MPASS(td->td_pflags & TDP_RFPPWAIT); 813 814 p = td->td_proc; 815 /* 816 * Preserve synchronization semantics of vfork. If 817 * waiting for child to exec or exit, fork set 818 * P_PPWAIT on child, and there we sleep on our proc 819 * (in case of exit). 820 * 821 * Do it after the ptracestop() above is finished, to 822 * not block our debugger until child execs or exits 823 * to finish vfork wait. 824 */ 825 td->td_pflags &= ~TDP_RFPPWAIT; 826 p2 = td->td_rfppwait_p; 827 again: 828 PROC_LOCK(p2); 829 while (p2->p_flag & P_PPWAIT) { 830 PROC_LOCK(p); 831 if (thread_suspend_check_needed()) { 832 PROC_UNLOCK(p2); 833 thread_suspend_check(0); 834 PROC_UNLOCK(p); 835 goto again; 836 } else { 837 PROC_UNLOCK(p); 838 } 839 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz); 840 } 841 PROC_UNLOCK(p2); 842 843 if (td->td_dbgflags & TDB_VFORK) { 844 PROC_LOCK(p); 845 if (p->p_ptevents & PTRACE_VFORK) 846 ptracestop(td, SIGTRAP, NULL); 847 td->td_dbgflags &= ~TDB_VFORK; 848 PROC_UNLOCK(p); 849 } 850 } 851 852 int 853 fork1(struct thread *td, struct fork_req *fr) 854 { 855 struct proc *p1, *newproc; 856 struct thread *td2; 857 struct vmspace *vm2; 858 struct ucred *cred; 859 struct file *fp_procdesc; 860 vm_ooffset_t mem_charged; 861 int error, nprocs_new; 862 static int curfail; 863 static struct timeval lastfail; 864 int flags, pages; 865 866 flags = fr->fr_flags; 867 pages = fr->fr_pages; 868 869 if ((flags & RFSTOPPED) != 0) 870 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 871 else 872 MPASS(fr->fr_procp == NULL); 873 874 /* Check for the undefined or unimplemented flags. */ 875 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 876 return (EINVAL); 877 878 /* Signal value requires RFTSIGZMB. */ 879 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 880 return (EINVAL); 881 882 /* Can't copy and clear. */ 883 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 884 return (EINVAL); 885 886 /* Check the validity of the signal number. */ 887 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 888 return (EINVAL); 889 890 if ((flags & RFPROCDESC) != 0) { 891 /* Can't not create a process yet get a process descriptor. */ 892 if ((flags & RFPROC) == 0) 893 return (EINVAL); 894 895 /* Must provide a place to put a procdesc if creating one. */ 896 if (fr->fr_pd_fd == NULL) 897 return (EINVAL); 898 899 /* Check if we are using supported flags. */ 900 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 901 return (EINVAL); 902 } 903 904 p1 = td->td_proc; 905 906 /* 907 * Here we don't create a new process, but we divorce 908 * certain parts of a process from itself. 909 */ 910 if ((flags & RFPROC) == 0) { 911 if (fr->fr_procp != NULL) 912 *fr->fr_procp = NULL; 913 else if (fr->fr_pidp != NULL) 914 *fr->fr_pidp = 0; 915 return (fork_norfproc(td, flags)); 916 } 917 918 fp_procdesc = NULL; 919 newproc = NULL; 920 vm2 = NULL; 921 922 /* 923 * Increment the nprocs resource before allocations occur. 924 * Although process entries are dynamically created, we still 925 * keep a global limit on the maximum number we will 926 * create. There are hard-limits as to the number of processes 927 * that can run, established by the KVA and memory usage for 928 * the process data. 929 * 930 * Don't allow a nonprivileged user to use the last ten 931 * processes; don't let root exceed the limit. 932 */ 933 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 934 if (nprocs_new >= maxproc - 10) { 935 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 || 936 nprocs_new >= maxproc) { 937 error = EAGAIN; 938 sx_xlock(&allproc_lock); 939 if (ppsratecheck(&lastfail, &curfail, 1)) { 940 printf("maxproc limit exceeded by uid %u " 941 "(pid %d); see tuning(7) and " 942 "login.conf(5)\n", 943 td->td_ucred->cr_ruid, p1->p_pid); 944 } 945 sx_xunlock(&allproc_lock); 946 goto fail2; 947 } 948 } 949 950 /* 951 * If required, create a process descriptor in the parent first; we 952 * will abandon it if something goes wrong. We don't finit() until 953 * later. 954 */ 955 if (flags & RFPROCDESC) { 956 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 957 fr->fr_pd_flags, fr->fr_pd_fcaps); 958 if (error != 0) 959 goto fail2; 960 AUDIT_ARG_FD(*fr->fr_pd_fd); 961 } 962 963 mem_charged = 0; 964 if (pages == 0) 965 pages = kstack_pages; 966 /* Allocate new proc. */ 967 newproc = uma_zalloc(proc_zone, M_WAITOK); 968 td2 = FIRST_THREAD_IN_PROC(newproc); 969 if (td2 == NULL) { 970 td2 = thread_alloc(pages); 971 if (td2 == NULL) { 972 error = ENOMEM; 973 goto fail2; 974 } 975 proc_linkup(newproc, td2); 976 } else { 977 kmsan_thread_alloc(td2); 978 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 979 if (td2->td_kstack != 0) 980 vm_thread_dispose(td2); 981 if (!thread_alloc_stack(td2, pages)) { 982 error = ENOMEM; 983 goto fail2; 984 } 985 } 986 } 987 988 if ((flags & RFMEM) == 0) { 989 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 990 if (vm2 == NULL) { 991 error = ENOMEM; 992 goto fail2; 993 } 994 if (!swap_reserve(mem_charged)) { 995 /* 996 * The swap reservation failed. The accounting 997 * from the entries of the copied vm2 will be 998 * subtracted in vmspace_free(), so force the 999 * reservation there. 1000 */ 1001 swap_reserve_force(mem_charged); 1002 error = ENOMEM; 1003 goto fail2; 1004 } 1005 } else 1006 vm2 = NULL; 1007 1008 /* 1009 * XXX: This is ugly; when we copy resource usage, we need to bump 1010 * per-cred resource counters. 1011 */ 1012 proc_set_cred_init(newproc, td->td_ucred); 1013 1014 /* 1015 * Initialize resource accounting for the child process. 1016 */ 1017 error = racct_proc_fork(p1, newproc); 1018 if (error != 0) { 1019 error = EAGAIN; 1020 goto fail1; 1021 } 1022 1023 #ifdef MAC 1024 mac_proc_init(newproc); 1025 #endif 1026 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 1027 STAILQ_INIT(&newproc->p_ktr); 1028 1029 /* 1030 * Increment the count of procs running with this uid. Don't allow 1031 * a nonprivileged user to exceed their current limit. 1032 */ 1033 cred = td->td_ucred; 1034 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) { 1035 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0) 1036 goto fail0; 1037 chgproccnt(cred->cr_ruidinfo, 1, 0); 1038 } 1039 1040 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 1041 return (0); 1042 fail0: 1043 error = EAGAIN; 1044 #ifdef MAC 1045 mac_proc_destroy(newproc); 1046 #endif 1047 racct_proc_exit(newproc); 1048 fail1: 1049 proc_unset_cred(newproc); 1050 fail2: 1051 if (vm2 != NULL) 1052 vmspace_free(vm2); 1053 uma_zfree(proc_zone, newproc); 1054 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1055 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1056 fdrop(fp_procdesc, td); 1057 } 1058 atomic_add_int(&nprocs, -1); 1059 pause("fork", hz / 2); 1060 return (error); 1061 } 1062 1063 /* 1064 * Handle the return of a child process from fork1(). This function 1065 * is called from the MD fork_trampoline() entry point. 1066 */ 1067 void 1068 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1069 struct trapframe *frame) 1070 { 1071 struct proc *p; 1072 struct thread *td; 1073 struct thread *dtd; 1074 1075 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED); 1076 1077 td = curthread; 1078 p = td->td_proc; 1079 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1080 1081 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1082 td, td_get_sched(td), p->p_pid, td->td_name); 1083 1084 sched_fork_exit(td); 1085 /* 1086 * Processes normally resume in mi_switch() after being 1087 * cpu_switch()'ed to, but when children start up they arrive here 1088 * instead, so we must do much the same things as mi_switch() would. 1089 */ 1090 if ((dtd = PCPU_GET(deadthread))) { 1091 PCPU_SET(deadthread, NULL); 1092 thread_stash(dtd); 1093 } 1094 thread_unlock(td); 1095 1096 /* 1097 * cpu_fork_kthread_handler intercepts this function call to 1098 * have this call a non-return function to stay in kernel mode. 1099 * initproc has its own fork handler, but it does return. 1100 */ 1101 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1102 callout(arg, frame); 1103 1104 /* 1105 * Check if a kernel thread misbehaved and returned from its main 1106 * function. 1107 */ 1108 if (p->p_flag & P_KPROC) { 1109 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1110 td->td_name, p->p_pid); 1111 kthread_exit(); 1112 } 1113 mtx_assert(&Giant, MA_NOTOWNED); 1114 1115 if (p->p_sysent->sv_schedtail != NULL) 1116 (p->p_sysent->sv_schedtail)(td); 1117 } 1118 1119 /* 1120 * Simplified back end of syscall(), used when returning from fork() 1121 * directly into user mode. This function is passed in to fork_exit() 1122 * as the first parameter and is called when returning to a new 1123 * userland process. 1124 */ 1125 void 1126 fork_return(struct thread *td, struct trapframe *frame) 1127 { 1128 struct proc *p; 1129 1130 p = td->td_proc; 1131 if (td->td_dbgflags & TDB_STOPATFORK) { 1132 PROC_LOCK(p); 1133 if ((p->p_flag & P_TRACED) != 0) { 1134 /* 1135 * Inform the debugger if one is still present. 1136 */ 1137 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1138 ptracestop(td, SIGSTOP, NULL); 1139 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1140 } else { 1141 /* 1142 * ... otherwise clear the request. 1143 */ 1144 td->td_dbgflags &= ~TDB_STOPATFORK; 1145 } 1146 PROC_UNLOCK(p); 1147 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1148 /* 1149 * This is the start of a new thread in a traced 1150 * process. Report a system call exit event. 1151 */ 1152 PROC_LOCK(p); 1153 td->td_dbgflags |= TDB_SCX; 1154 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1155 (td->td_dbgflags & TDB_BORN) != 0) 1156 ptracestop(td, SIGTRAP, NULL); 1157 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1158 PROC_UNLOCK(p); 1159 } 1160 1161 /* 1162 * If the prison was killed mid-fork, die along with it. 1163 */ 1164 if (!prison_isalive(td->td_ucred->cr_prison)) 1165 exit1(td, 0, SIGKILL); 1166 1167 userret(td, frame); 1168 1169 #ifdef KTRACE 1170 if (KTRPOINT(td, KTR_SYSRET)) 1171 ktrsysret(SYS_fork, 0, 0); 1172 #endif 1173 } 1174