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 struct proc *p1; 311 int error; 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 != 0) 334 goto fail; 335 336 /* 337 * Close all file descriptors. 338 */ 339 if ((flags & RFCFDG) != 0) { 340 struct filedesc *fdtmp; 341 struct pwddesc *pdtmp; 342 343 pdtmp = pdinit(td->td_proc->p_pd, false); 344 fdtmp = fdinit(); 345 pdescfree(td); 346 fdescfree(td); 347 p1->p_fd = fdtmp; 348 p1->p_pd = pdtmp; 349 } 350 351 /* 352 * Unshare file descriptors (from parent). 353 */ 354 if ((flags & RFFDG) != 0) { 355 fdunshare(td); 356 pdunshare(td); 357 } 358 359 fail: 360 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS && 361 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) { 362 PROC_LOCK(p1); 363 thread_single_end(p1, SINGLE_BOUNDARY); 364 PROC_UNLOCK(p1); 365 } 366 return (error); 367 } 368 369 static void 370 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 371 struct vmspace *vm2, struct file *fp_procdesc) 372 { 373 struct proc *p1, *pptr; 374 struct filedesc *fd; 375 struct filedesc_to_leader *fdtol; 376 struct pwddesc *pd; 377 struct sigacts *newsigacts; 378 379 p1 = td->td_proc; 380 381 PROC_LOCK(p1); 382 bcopy(&p1->p_startcopy, &p2->p_startcopy, 383 __rangeof(struct proc, p_startcopy, p_endcopy)); 384 pargs_hold(p2->p_args); 385 PROC_UNLOCK(p1); 386 387 bzero(&p2->p_startzero, 388 __rangeof(struct proc, p_startzero, p_endzero)); 389 390 /* Tell the prison that we exist. */ 391 prison_proc_hold(p2->p_ucred->cr_prison); 392 393 p2->p_state = PRS_NEW; /* protect against others */ 394 p2->p_pid = fork_findpid(fr->fr_flags); 395 AUDIT_ARG_PID(p2->p_pid); 396 TSFORK(p2->p_pid, p1->p_pid); 397 398 sx_xlock(&allproc_lock); 399 LIST_INSERT_HEAD(&allproc, p2, p_list); 400 allproc_gen++; 401 prison_proc_link(p2->p_ucred->cr_prison, p2); 402 sx_xunlock(&allproc_lock); 403 404 sx_xlock(PIDHASHLOCK(p2->p_pid)); 405 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 406 sx_xunlock(PIDHASHLOCK(p2->p_pid)); 407 408 tidhash_add(td2); 409 410 /* 411 * Malloc things while we don't hold any locks. 412 */ 413 if (fr->fr_flags & RFSIGSHARE) 414 newsigacts = NULL; 415 else 416 newsigacts = sigacts_alloc(); 417 418 /* 419 * Copy filedesc. 420 */ 421 if (fr->fr_flags & RFCFDG) { 422 pd = pdinit(p1->p_pd, false); 423 fd = fdinit(); 424 fdtol = NULL; 425 } else if (fr->fr_flags & RFFDG) { 426 if (fr->fr_flags2 & FR2_SHARE_PATHS) 427 pd = pdshare(p1->p_pd); 428 else 429 pd = pdcopy(p1->p_pd); 430 fd = fdcopy(p1->p_fd); 431 fdtol = NULL; 432 } else { 433 if (fr->fr_flags2 & FR2_SHARE_PATHS) 434 pd = pdcopy(p1->p_pd); 435 else 436 pd = pdshare(p1->p_pd); 437 fd = fdshare(p1->p_fd); 438 if (p1->p_fdtol == NULL) 439 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 440 p1->p_leader); 441 if ((fr->fr_flags & RFTHREAD) != 0) { 442 /* 443 * Shared file descriptor table, and shared 444 * process leaders. 445 */ 446 fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd); 447 } else { 448 /* 449 * Shared file descriptor table, and different 450 * process leaders. 451 */ 452 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 453 p1->p_fd, p2); 454 } 455 } 456 /* 457 * Make a proc table entry for the new process. 458 * Start by zeroing the section of proc that is zero-initialized, 459 * then copy the section that is copied directly from the parent. 460 */ 461 462 PROC_LOCK(p2); 463 PROC_LOCK(p1); 464 465 bzero(&td2->td_startzero, 466 __rangeof(struct thread, td_startzero, td_endzero)); 467 468 bcopy(&td->td_startcopy, &td2->td_startcopy, 469 __rangeof(struct thread, td_startcopy, td_endcopy)); 470 471 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 472 td2->td_sigstk = td->td_sigstk; 473 td2->td_flags = TDF_INMEM; 474 td2->td_lend_user_pri = PRI_MAX; 475 476 #ifdef VIMAGE 477 td2->td_vnet = NULL; 478 td2->td_vnet_lpush = NULL; 479 #endif 480 481 /* 482 * Allow the scheduler to initialize the child. 483 */ 484 thread_lock(td); 485 sched_fork(td, td2); 486 /* 487 * Request AST to check for TDP_RFPPWAIT. Do it here 488 * to avoid calling thread_lock() again. 489 */ 490 if ((fr->fr_flags & RFPPWAIT) != 0) 491 ast_sched_locked(td, TDA_VFORK); 492 thread_unlock(td); 493 494 /* 495 * Duplicate sub-structures as needed. 496 * Increase reference counts on shared objects. 497 */ 498 p2->p_flag = P_INMEM; 499 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 500 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 501 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP | 502 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS | 503 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 504 p2->p_swtick = ticks; 505 if (p1->p_flag & P_PROFIL) 506 startprofclock(p2); 507 508 if (fr->fr_flags & RFSIGSHARE) { 509 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 510 } else { 511 sigacts_copy(newsigacts, p1->p_sigacts); 512 p2->p_sigacts = newsigacts; 513 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) { 514 mtx_lock(&p2->p_sigacts->ps_mtx); 515 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) 516 sig_drop_caught(p2); 517 if ((fr->fr_flags2 & FR2_KPROC) != 0) 518 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT; 519 mtx_unlock(&p2->p_sigacts->ps_mtx); 520 } 521 } 522 523 if (fr->fr_flags & RFTSIGZMB) 524 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 525 else if (fr->fr_flags & RFLINUXTHPN) 526 p2->p_sigparent = SIGUSR1; 527 else 528 p2->p_sigparent = SIGCHLD; 529 530 if ((fr->fr_flags2 & FR2_KPROC) != 0) { 531 p2->p_flag |= P_SYSTEM | P_KPROC; 532 td2->td_pflags |= TDP_KTHREAD; 533 } 534 535 p2->p_textvp = p1->p_textvp; 536 p2->p_textdvp = p1->p_textdvp; 537 p2->p_fd = fd; 538 p2->p_fdtol = fdtol; 539 p2->p_pd = pd; 540 541 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 542 p2->p_flag |= P_PROTECTED; 543 p2->p_flag2 |= P2_INHERIT_PROTECTED; 544 } 545 546 /* 547 * p_limit is copy-on-write. Bump its refcount. 548 */ 549 lim_fork(p1, p2); 550 551 thread_cow_get_proc(td2, p2); 552 553 pstats_fork(p1->p_stats, p2->p_stats); 554 555 PROC_UNLOCK(p1); 556 PROC_UNLOCK(p2); 557 558 /* 559 * Bump references to the text vnode and directory, and copy 560 * the hardlink name. 561 */ 562 if (p2->p_textvp != NULL) 563 vrefact(p2->p_textvp); 564 if (p2->p_textdvp != NULL) 565 vrefact(p2->p_textdvp); 566 p2->p_binname = p1->p_binname == NULL ? NULL : 567 strdup(p1->p_binname, M_PARGS); 568 569 /* 570 * Set up linkage for kernel based threading. 571 */ 572 if ((fr->fr_flags & RFTHREAD) != 0) { 573 mtx_lock(&ppeers_lock); 574 p2->p_peers = p1->p_peers; 575 p1->p_peers = p2; 576 p2->p_leader = p1->p_leader; 577 mtx_unlock(&ppeers_lock); 578 PROC_LOCK(p1->p_leader); 579 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 580 PROC_UNLOCK(p1->p_leader); 581 /* 582 * The task leader is exiting, so process p1 is 583 * going to be killed shortly. Since p1 obviously 584 * isn't dead yet, we know that the leader is either 585 * sending SIGKILL's to all the processes in this 586 * task or is sleeping waiting for all the peers to 587 * exit. We let p1 complete the fork, but we need 588 * to go ahead and kill the new process p2 since 589 * the task leader may not get a chance to send 590 * SIGKILL to it. We leave it on the list so that 591 * the task leader will wait for this new process 592 * to commit suicide. 593 */ 594 PROC_LOCK(p2); 595 kern_psignal(p2, SIGKILL); 596 PROC_UNLOCK(p2); 597 } else 598 PROC_UNLOCK(p1->p_leader); 599 } else { 600 p2->p_peers = NULL; 601 p2->p_leader = p2; 602 } 603 604 sx_xlock(&proctree_lock); 605 PGRP_LOCK(p1->p_pgrp); 606 PROC_LOCK(p2); 607 PROC_LOCK(p1); 608 609 /* 610 * Preserve some more flags in subprocess. P_PROFIL has already 611 * been preserved. 612 */ 613 p2->p_flag |= p1->p_flag & P_SUGID; 614 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK)); 615 SESS_LOCK(p1->p_session); 616 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 617 p2->p_flag |= P_CONTROLT; 618 SESS_UNLOCK(p1->p_session); 619 if (fr->fr_flags & RFPPWAIT) 620 p2->p_flag |= P_PPWAIT; 621 622 p2->p_pgrp = p1->p_pgrp; 623 LIST_INSERT_AFTER(p1, p2, p_pglist); 624 PGRP_UNLOCK(p1->p_pgrp); 625 LIST_INIT(&p2->p_children); 626 LIST_INIT(&p2->p_orphans); 627 628 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 629 TAILQ_INIT(&p2->p_kqtim_stop); 630 631 /* 632 * This begins the section where we must prevent the parent 633 * from being swapped. 634 */ 635 _PHOLD(p1); 636 PROC_UNLOCK(p1); 637 638 /* 639 * Attach the new process to its parent. 640 * 641 * If RFNOWAIT is set, the newly created process becomes a child 642 * of init. This effectively disassociates the child from the 643 * parent. 644 */ 645 if ((fr->fr_flags & RFNOWAIT) != 0) { 646 pptr = p1->p_reaper; 647 p2->p_reaper = pptr; 648 } else { 649 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 650 p1 : p1->p_reaper; 651 pptr = p1; 652 } 653 p2->p_pptr = pptr; 654 p2->p_oppid = pptr->p_pid; 655 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 656 LIST_INIT(&p2->p_reaplist); 657 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 658 if (p2->p_reaper == p1 && p1 != initproc) { 659 p2->p_reapsubtree = p2->p_pid; 660 proc_id_set_cond(PROC_ID_REAP, p2->p_pid); 661 } 662 sx_xunlock(&proctree_lock); 663 664 /* Inform accounting that we have forked. */ 665 p2->p_acflag = AFORK; 666 PROC_UNLOCK(p2); 667 668 #ifdef KTRACE 669 ktrprocfork(p1, p2); 670 #endif 671 672 /* 673 * Finish creating the child process. It will return via a different 674 * execution path later. (ie: directly into user mode) 675 */ 676 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 677 678 if (fr->fr_flags == (RFFDG | RFPROC)) { 679 VM_CNT_INC(v_forks); 680 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 681 p2->p_vmspace->vm_ssize); 682 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 683 VM_CNT_INC(v_vforks); 684 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 685 p2->p_vmspace->vm_ssize); 686 } else if (p1 == &proc0) { 687 VM_CNT_INC(v_kthreads); 688 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 689 p2->p_vmspace->vm_ssize); 690 } else { 691 VM_CNT_INC(v_rforks); 692 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 693 p2->p_vmspace->vm_ssize); 694 } 695 696 /* 697 * Associate the process descriptor with the process before anything 698 * can happen that might cause that process to need the descriptor. 699 * However, don't do this until after fork(2) can no longer fail. 700 */ 701 if (fr->fr_flags & RFPROCDESC) 702 procdesc_new(p2, fr->fr_pd_flags); 703 704 /* 705 * Both processes are set up, now check if any loadable modules want 706 * to adjust anything. 707 */ 708 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 709 710 /* 711 * Set the child start time and mark the process as being complete. 712 */ 713 PROC_LOCK(p2); 714 PROC_LOCK(p1); 715 microuptime(&p2->p_stats->p_start); 716 PROC_SLOCK(p2); 717 p2->p_state = PRS_NORMAL; 718 PROC_SUNLOCK(p2); 719 720 #ifdef KDTRACE_HOOKS 721 /* 722 * Tell the DTrace fasttrap provider about the new process so that any 723 * tracepoints inherited from the parent can be removed. We have to do 724 * this only after p_state is PRS_NORMAL since the fasttrap module will 725 * use pfind() later on. 726 */ 727 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 728 dtrace_fasttrap_fork(p1, p2); 729 #endif 730 if (fr->fr_flags & RFPPWAIT) { 731 td->td_pflags |= TDP_RFPPWAIT; 732 td->td_rfppwait_p = p2; 733 td->td_dbgflags |= TDB_VFORK; 734 } 735 PROC_UNLOCK(p2); 736 737 /* 738 * Tell any interested parties about the new process. 739 */ 740 knote_fork(p1->p_klist, p2->p_pid); 741 742 /* 743 * Now can be swapped. 744 */ 745 _PRELE(p1); 746 PROC_UNLOCK(p1); 747 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 748 749 if (fr->fr_flags & RFPROCDESC) { 750 procdesc_finit(p2->p_procdesc, fp_procdesc); 751 fdrop(fp_procdesc, td); 752 } 753 754 /* 755 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 756 * synced with forks in progress so it is OK if we miss it 757 * if being set atm. 758 */ 759 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 760 sx_xlock(&proctree_lock); 761 PROC_LOCK(p2); 762 763 /* 764 * p1->p_ptevents & p1->p_pptr are protected by both 765 * process and proctree locks for modifications, 766 * so owning proctree_lock allows the race-free read. 767 */ 768 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 769 /* 770 * Arrange for debugger to receive the fork event. 771 * 772 * We can report PL_FLAG_FORKED regardless of 773 * P_FOLLOWFORK settings, but it does not make a sense 774 * for runaway child. 775 */ 776 td->td_dbgflags |= TDB_FORK; 777 td->td_dbg_forked = p2->p_pid; 778 td2->td_dbgflags |= TDB_STOPATFORK; 779 proc_set_traced(p2, true); 780 CTR2(KTR_PTRACE, 781 "do_fork: attaching to new child pid %d: oppid %d", 782 p2->p_pid, p2->p_oppid); 783 proc_reparent(p2, p1->p_pptr, false); 784 } 785 PROC_UNLOCK(p2); 786 sx_xunlock(&proctree_lock); 787 } 788 789 racct_proc_fork_done(p2); 790 791 if ((fr->fr_flags & RFSTOPPED) == 0) { 792 if (fr->fr_pidp != NULL) 793 *fr->fr_pidp = p2->p_pid; 794 /* 795 * If RFSTOPPED not requested, make child runnable and 796 * add to run queue. 797 */ 798 thread_lock(td2); 799 TD_SET_CAN_RUN(td2); 800 sched_add(td2, SRQ_BORING); 801 } else { 802 *fr->fr_procp = p2; 803 } 804 } 805 806 static void 807 ast_vfork(struct thread *td, int tda __unused) 808 { 809 struct proc *p, *p2; 810 811 MPASS(td->td_pflags & TDP_RFPPWAIT); 812 813 p = td->td_proc; 814 /* 815 * Preserve synchronization semantics of vfork. If 816 * waiting for child to exec or exit, fork set 817 * P_PPWAIT on child, and there we sleep on our proc 818 * (in case of exit). 819 * 820 * Do it after the ptracestop() above is finished, to 821 * not block our debugger until child execs or exits 822 * to finish vfork wait. 823 */ 824 td->td_pflags &= ~TDP_RFPPWAIT; 825 p2 = td->td_rfppwait_p; 826 again: 827 PROC_LOCK(p2); 828 while (p2->p_flag & P_PPWAIT) { 829 PROC_LOCK(p); 830 if (thread_suspend_check_needed()) { 831 PROC_UNLOCK(p2); 832 thread_suspend_check(0); 833 PROC_UNLOCK(p); 834 goto again; 835 } else { 836 PROC_UNLOCK(p); 837 } 838 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz); 839 } 840 PROC_UNLOCK(p2); 841 842 if (td->td_dbgflags & TDB_VFORK) { 843 PROC_LOCK(p); 844 if (p->p_ptevents & PTRACE_VFORK) 845 ptracestop(td, SIGTRAP, NULL); 846 td->td_dbgflags &= ~TDB_VFORK; 847 PROC_UNLOCK(p); 848 } 849 } 850 851 int 852 fork1(struct thread *td, struct fork_req *fr) 853 { 854 struct proc *p1, *newproc; 855 struct thread *td2; 856 struct vmspace *vm2; 857 struct ucred *cred; 858 struct file *fp_procdesc; 859 struct pgrp *pg; 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 bool killsx_locked; 866 867 flags = fr->fr_flags; 868 pages = fr->fr_pages; 869 870 if ((flags & RFSTOPPED) != 0) 871 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 872 else 873 MPASS(fr->fr_procp == NULL); 874 875 /* Check for the undefined or unimplemented flags. */ 876 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 877 return (EINVAL); 878 879 /* Signal value requires RFTSIGZMB. */ 880 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 881 return (EINVAL); 882 883 /* Can't copy and clear. */ 884 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 885 return (EINVAL); 886 887 /* Check the validity of the signal number. */ 888 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 889 return (EINVAL); 890 891 if ((flags & RFPROCDESC) != 0) { 892 /* Can't not create a process yet get a process descriptor. */ 893 if ((flags & RFPROC) == 0) 894 return (EINVAL); 895 896 /* Must provide a place to put a procdesc if creating one. */ 897 if (fr->fr_pd_fd == NULL) 898 return (EINVAL); 899 900 /* Check if we are using supported flags. */ 901 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 902 return (EINVAL); 903 } 904 905 p1 = td->td_proc; 906 907 /* 908 * Here we don't create a new process, but we divorce 909 * certain parts of a process from itself. 910 */ 911 if ((flags & RFPROC) == 0) { 912 if (fr->fr_procp != NULL) 913 *fr->fr_procp = NULL; 914 else if (fr->fr_pidp != NULL) 915 *fr->fr_pidp = 0; 916 return (fork_norfproc(td, flags)); 917 } 918 919 fp_procdesc = NULL; 920 newproc = NULL; 921 vm2 = NULL; 922 killsx_locked = false; 923 924 /* 925 * Increment the nprocs resource before allocations occur. 926 * Although process entries are dynamically created, we still 927 * keep a global limit on the maximum number we will 928 * create. There are hard-limits as to the number of processes 929 * that can run, established by the KVA and memory usage for 930 * the process data. 931 * 932 * Don't allow a nonprivileged user to use the last ten 933 * processes; don't let root exceed the limit. 934 */ 935 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 936 if (nprocs_new >= maxproc - 10) { 937 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 || 938 nprocs_new >= maxproc) { 939 error = EAGAIN; 940 sx_xlock(&allproc_lock); 941 if (ppsratecheck(&lastfail, &curfail, 1)) { 942 printf("maxproc limit exceeded by uid %u " 943 "(pid %d); see tuning(7) and " 944 "login.conf(5)\n", 945 td->td_ucred->cr_ruid, p1->p_pid); 946 } 947 sx_xunlock(&allproc_lock); 948 goto fail2; 949 } 950 } 951 952 /* 953 * Atomically check for signals and block threads from sending 954 * a signal to our process group until the child is visible. 955 */ 956 pg = p1->p_pgrp; 957 if (sx_slock_sig(&pg->pg_killsx) != 0) { 958 error = ERESTART; 959 goto fail2; 960 } else if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0 || 961 atomic_load_int(&p1->p_killpg_cnt) != 0)) { 962 /* 963 * Either the process was moved to other process 964 * group, or there is pending signal. sx_slock_sig() 965 * does not check for signals if not sleeping for the 966 * lock. 967 */ 968 sx_sunlock(&pg->pg_killsx); 969 error = ERESTART; 970 goto fail2; 971 } else { 972 killsx_locked = true; 973 } 974 975 /* 976 * If required, create a process descriptor in the parent first; we 977 * will abandon it if something goes wrong. We don't finit() until 978 * later. 979 */ 980 if (flags & RFPROCDESC) { 981 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 982 fr->fr_pd_flags, fr->fr_pd_fcaps); 983 if (error != 0) 984 goto fail2; 985 AUDIT_ARG_FD(*fr->fr_pd_fd); 986 } 987 988 mem_charged = 0; 989 if (pages == 0) 990 pages = kstack_pages; 991 /* Allocate new proc. */ 992 newproc = uma_zalloc(proc_zone, M_WAITOK); 993 td2 = FIRST_THREAD_IN_PROC(newproc); 994 if (td2 == NULL) { 995 td2 = thread_alloc(pages); 996 if (td2 == NULL) { 997 error = ENOMEM; 998 goto fail2; 999 } 1000 proc_linkup(newproc, td2); 1001 } else { 1002 kmsan_thread_alloc(td2); 1003 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 1004 if (td2->td_kstack != 0) 1005 vm_thread_dispose(td2); 1006 if (!thread_alloc_stack(td2, pages)) { 1007 error = ENOMEM; 1008 goto fail2; 1009 } 1010 } 1011 } 1012 1013 if ((flags & RFMEM) == 0) { 1014 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 1015 if (vm2 == NULL) { 1016 error = ENOMEM; 1017 goto fail2; 1018 } 1019 if (!swap_reserve(mem_charged)) { 1020 /* 1021 * The swap reservation failed. The accounting 1022 * from the entries of the copied vm2 will be 1023 * subtracted in vmspace_free(), so force the 1024 * reservation there. 1025 */ 1026 swap_reserve_force(mem_charged); 1027 error = ENOMEM; 1028 goto fail2; 1029 } 1030 } else 1031 vm2 = NULL; 1032 1033 /* 1034 * XXX: This is ugly; when we copy resource usage, we need to bump 1035 * per-cred resource counters. 1036 */ 1037 proc_set_cred_init(newproc, td->td_ucred); 1038 1039 /* 1040 * Initialize resource accounting for the child process. 1041 */ 1042 error = racct_proc_fork(p1, newproc); 1043 if (error != 0) { 1044 error = EAGAIN; 1045 goto fail1; 1046 } 1047 1048 #ifdef MAC 1049 mac_proc_init(newproc); 1050 #endif 1051 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 1052 STAILQ_INIT(&newproc->p_ktr); 1053 1054 /* 1055 * Increment the count of procs running with this uid. Don't allow 1056 * a nonprivileged user to exceed their current limit. 1057 */ 1058 cred = td->td_ucred; 1059 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) { 1060 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0) 1061 goto fail0; 1062 chgproccnt(cred->cr_ruidinfo, 1, 0); 1063 } 1064 1065 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 1066 sx_sunlock(&pg->pg_killsx); 1067 return (0); 1068 fail0: 1069 error = EAGAIN; 1070 #ifdef MAC 1071 mac_proc_destroy(newproc); 1072 #endif 1073 racct_proc_exit(newproc); 1074 fail1: 1075 proc_unset_cred(newproc); 1076 fail2: 1077 if (vm2 != NULL) 1078 vmspace_free(vm2); 1079 uma_zfree(proc_zone, newproc); 1080 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1081 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1082 fdrop(fp_procdesc, td); 1083 } 1084 atomic_add_int(&nprocs, -1); 1085 if (killsx_locked) 1086 sx_sunlock(&pg->pg_killsx); 1087 pause("fork", hz / 2); 1088 return (error); 1089 } 1090 1091 /* 1092 * Handle the return of a child process from fork1(). This function 1093 * is called from the MD fork_trampoline() entry point. 1094 */ 1095 void 1096 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1097 struct trapframe *frame) 1098 { 1099 struct proc *p; 1100 struct thread *td; 1101 struct thread *dtd; 1102 1103 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED); 1104 1105 td = curthread; 1106 p = td->td_proc; 1107 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1108 1109 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1110 td, td_get_sched(td), p->p_pid, td->td_name); 1111 1112 sched_fork_exit(td); 1113 1114 /* 1115 * Processes normally resume in mi_switch() after being 1116 * cpu_switch()'ed to, but when children start up they arrive here 1117 * instead, so we must do much the same things as mi_switch() would. 1118 */ 1119 if ((dtd = PCPU_GET(deadthread))) { 1120 PCPU_SET(deadthread, NULL); 1121 thread_stash(dtd); 1122 } 1123 thread_unlock(td); 1124 1125 /* 1126 * cpu_fork_kthread_handler intercepts this function call to 1127 * have this call a non-return function to stay in kernel mode. 1128 * initproc has its own fork handler, but it does return. 1129 */ 1130 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1131 callout(arg, frame); 1132 1133 /* 1134 * Check if a kernel thread misbehaved and returned from its main 1135 * function. 1136 */ 1137 if (p->p_flag & P_KPROC) { 1138 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1139 td->td_name, p->p_pid); 1140 kthread_exit(); 1141 } 1142 mtx_assert(&Giant, MA_NOTOWNED); 1143 1144 if (p->p_sysent->sv_schedtail != NULL) 1145 (p->p_sysent->sv_schedtail)(td); 1146 } 1147 1148 /* 1149 * Simplified back end of syscall(), used when returning from fork() 1150 * directly into user mode. This function is passed in to fork_exit() 1151 * as the first parameter and is called when returning to a new 1152 * userland process. 1153 */ 1154 void 1155 fork_return(struct thread *td, struct trapframe *frame) 1156 { 1157 struct proc *p; 1158 1159 p = td->td_proc; 1160 if (td->td_dbgflags & TDB_STOPATFORK) { 1161 PROC_LOCK(p); 1162 if ((p->p_flag & P_TRACED) != 0) { 1163 /* 1164 * Inform the debugger if one is still present. 1165 */ 1166 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1167 ptracestop(td, SIGSTOP, NULL); 1168 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1169 } else { 1170 /* 1171 * ... otherwise clear the request. 1172 */ 1173 td->td_dbgflags &= ~TDB_STOPATFORK; 1174 } 1175 PROC_UNLOCK(p); 1176 } else if (p->p_flag & P_TRACED) { 1177 /* 1178 * This is the start of a new thread in a traced 1179 * process. Report a system call exit event. 1180 */ 1181 PROC_LOCK(p); 1182 td->td_dbgflags |= TDB_SCX; 1183 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1184 (td->td_dbgflags & TDB_BORN) != 0) 1185 ptracestop(td, SIGTRAP, NULL); 1186 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1187 PROC_UNLOCK(p); 1188 } 1189 1190 /* 1191 * If the prison was killed mid-fork, die along with it. 1192 */ 1193 if (!prison_isalive(td->td_ucred->cr_prison)) 1194 exit1(td, 0, SIGKILL); 1195 1196 userret(td, frame); 1197 1198 #ifdef KTRACE 1199 if (KTRPOINT(td, KTR_SYSRET)) 1200 ktrsysret(td->td_sa.code, 0, 0); 1201 #endif 1202 } 1203 1204 static void 1205 fork_init(void *arg __unused) 1206 { 1207 ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT, 1208 ast_vfork); 1209 } 1210 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL); 1211