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 again: 323 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS && 324 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) { 325 PROC_LOCK(p1); 326 while (p1->p_singlethr > 0) { 327 error = msleep(&p1->p_singlethr, &p1->p_mtx, 328 PWAIT | PCATCH, "rfork1t", 0); 329 if (error != 0) { 330 PROC_UNLOCK(p1); 331 return (ERESTART); 332 } 333 goto again; 334 } 335 if (thread_single(p1, SINGLE_BOUNDARY)) { 336 PROC_UNLOCK(p1); 337 return (ERESTART); 338 } 339 PROC_UNLOCK(p1); 340 } 341 342 error = vm_forkproc(td, NULL, NULL, NULL, flags); 343 if (error) 344 goto fail; 345 346 /* 347 * Close all file descriptors. 348 */ 349 if (flags & RFCFDG) { 350 struct filedesc *fdtmp; 351 struct pwddesc *pdtmp; 352 pdtmp = pdinit(td->td_proc->p_pd, false); 353 fdtmp = fdinit(); 354 pdescfree(td); 355 fdescfree(td); 356 p1->p_fd = fdtmp; 357 p1->p_pd = pdtmp; 358 } 359 360 /* 361 * Unshare file descriptors (from parent). 362 */ 363 if (flags & RFFDG) { 364 fdunshare(td); 365 pdunshare(td); 366 } 367 368 fail: 369 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS && 370 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) { 371 PROC_LOCK(p1); 372 thread_single_end(p1, SINGLE_BOUNDARY); 373 PROC_UNLOCK(p1); 374 } 375 return (error); 376 } 377 378 static void 379 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 380 struct vmspace *vm2, struct file *fp_procdesc) 381 { 382 struct proc *p1, *pptr; 383 struct filedesc *fd; 384 struct filedesc_to_leader *fdtol; 385 struct pwddesc *pd; 386 struct sigacts *newsigacts; 387 388 p1 = td->td_proc; 389 390 PROC_LOCK(p1); 391 bcopy(&p1->p_startcopy, &p2->p_startcopy, 392 __rangeof(struct proc, p_startcopy, p_endcopy)); 393 pargs_hold(p2->p_args); 394 PROC_UNLOCK(p1); 395 396 bzero(&p2->p_startzero, 397 __rangeof(struct proc, p_startzero, p_endzero)); 398 399 /* Tell the prison that we exist. */ 400 prison_proc_hold(p2->p_ucred->cr_prison); 401 402 p2->p_state = PRS_NEW; /* protect against others */ 403 p2->p_pid = fork_findpid(fr->fr_flags); 404 AUDIT_ARG_PID(p2->p_pid); 405 TSFORK(p2->p_pid, p1->p_pid); 406 407 sx_xlock(&allproc_lock); 408 LIST_INSERT_HEAD(&allproc, p2, p_list); 409 allproc_gen++; 410 sx_xunlock(&allproc_lock); 411 412 sx_xlock(PIDHASHLOCK(p2->p_pid)); 413 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 414 sx_xunlock(PIDHASHLOCK(p2->p_pid)); 415 416 tidhash_add(td2); 417 418 /* 419 * Malloc things while we don't hold any locks. 420 */ 421 if (fr->fr_flags & RFSIGSHARE) 422 newsigacts = NULL; 423 else 424 newsigacts = sigacts_alloc(); 425 426 /* 427 * Copy filedesc. 428 */ 429 if (fr->fr_flags & RFCFDG) { 430 pd = pdinit(p1->p_pd, false); 431 fd = fdinit(); 432 fdtol = NULL; 433 } else if (fr->fr_flags & RFFDG) { 434 if (fr->fr_flags2 & FR2_SHARE_PATHS) 435 pd = pdshare(p1->p_pd); 436 else 437 pd = pdcopy(p1->p_pd); 438 fd = fdcopy(p1->p_fd); 439 fdtol = NULL; 440 } else { 441 if (fr->fr_flags2 & FR2_SHARE_PATHS) 442 pd = pdcopy(p1->p_pd); 443 else 444 pd = pdshare(p1->p_pd); 445 fd = fdshare(p1->p_fd); 446 if (p1->p_fdtol == NULL) 447 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 448 p1->p_leader); 449 if ((fr->fr_flags & RFTHREAD) != 0) { 450 /* 451 * Shared file descriptor table, and shared 452 * process leaders. 453 */ 454 fdtol = p1->p_fdtol; 455 FILEDESC_XLOCK(p1->p_fd); 456 fdtol->fdl_refcount++; 457 FILEDESC_XUNLOCK(p1->p_fd); 458 } else { 459 /* 460 * Shared file descriptor table, and different 461 * process leaders. 462 */ 463 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 464 p1->p_fd, p2); 465 } 466 } 467 /* 468 * Make a proc table entry for the new process. 469 * Start by zeroing the section of proc that is zero-initialized, 470 * then copy the section that is copied directly from the parent. 471 */ 472 473 PROC_LOCK(p2); 474 PROC_LOCK(p1); 475 476 bzero(&td2->td_startzero, 477 __rangeof(struct thread, td_startzero, td_endzero)); 478 479 bcopy(&td->td_startcopy, &td2->td_startcopy, 480 __rangeof(struct thread, td_startcopy, td_endcopy)); 481 482 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 483 td2->td_sigstk = td->td_sigstk; 484 td2->td_flags = TDF_INMEM; 485 td2->td_lend_user_pri = PRI_MAX; 486 487 #ifdef VIMAGE 488 td2->td_vnet = NULL; 489 td2->td_vnet_lpush = NULL; 490 #endif 491 492 /* 493 * Allow the scheduler to initialize the child. 494 */ 495 thread_lock(td); 496 sched_fork(td, td2); 497 /* 498 * Request AST to check for TDP_RFPPWAIT. Do it here 499 * to avoid calling thread_lock() again. 500 */ 501 if ((fr->fr_flags & RFPPWAIT) != 0) 502 td->td_flags |= TDF_ASTPENDING; 503 thread_unlock(td); 504 505 /* 506 * Duplicate sub-structures as needed. 507 * Increase reference counts on shared objects. 508 */ 509 p2->p_flag = P_INMEM; 510 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 511 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 512 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP | 513 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS | 514 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 515 p2->p_swtick = ticks; 516 if (p1->p_flag & P_PROFIL) 517 startprofclock(p2); 518 519 if (fr->fr_flags & RFSIGSHARE) { 520 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 521 } else { 522 sigacts_copy(newsigacts, p1->p_sigacts); 523 p2->p_sigacts = newsigacts; 524 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) { 525 mtx_lock(&p2->p_sigacts->ps_mtx); 526 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) 527 sig_drop_caught(p2); 528 if ((fr->fr_flags2 & FR2_KPROC) != 0) 529 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT; 530 mtx_unlock(&p2->p_sigacts->ps_mtx); 531 } 532 } 533 534 if (fr->fr_flags & RFTSIGZMB) 535 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 536 else if (fr->fr_flags & RFLINUXTHPN) 537 p2->p_sigparent = SIGUSR1; 538 else 539 p2->p_sigparent = SIGCHLD; 540 541 if ((fr->fr_flags2 & FR2_KPROC) != 0) { 542 p2->p_flag |= P_SYSTEM | P_KPROC; 543 td2->td_pflags |= TDP_KTHREAD; 544 } 545 546 p2->p_textvp = p1->p_textvp; 547 p2->p_textdvp = p1->p_textdvp; 548 p2->p_fd = fd; 549 p2->p_fdtol = fdtol; 550 p2->p_pd = pd; 551 552 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 553 p2->p_flag |= P_PROTECTED; 554 p2->p_flag2 |= P2_INHERIT_PROTECTED; 555 } 556 557 /* 558 * p_limit is copy-on-write. Bump its refcount. 559 */ 560 lim_fork(p1, p2); 561 562 thread_cow_get_proc(td2, p2); 563 564 pstats_fork(p1->p_stats, p2->p_stats); 565 566 PROC_UNLOCK(p1); 567 PROC_UNLOCK(p2); 568 569 /* 570 * Bump references to the text vnode and directory, and copy 571 * the hardlink name. 572 */ 573 if (p2->p_textvp != NULL) 574 vrefact(p2->p_textvp); 575 if (p2->p_textdvp != NULL) 576 vrefact(p2->p_textdvp); 577 p2->p_binname = p1->p_binname == NULL ? NULL : 578 strdup(p1->p_binname, M_PARGS); 579 580 /* 581 * Set up linkage for kernel based threading. 582 */ 583 if ((fr->fr_flags & RFTHREAD) != 0) { 584 mtx_lock(&ppeers_lock); 585 p2->p_peers = p1->p_peers; 586 p1->p_peers = p2; 587 p2->p_leader = p1->p_leader; 588 mtx_unlock(&ppeers_lock); 589 PROC_LOCK(p1->p_leader); 590 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 591 PROC_UNLOCK(p1->p_leader); 592 /* 593 * The task leader is exiting, so process p1 is 594 * going to be killed shortly. Since p1 obviously 595 * isn't dead yet, we know that the leader is either 596 * sending SIGKILL's to all the processes in this 597 * task or is sleeping waiting for all the peers to 598 * exit. We let p1 complete the fork, but we need 599 * to go ahead and kill the new process p2 since 600 * the task leader may not get a chance to send 601 * SIGKILL to it. We leave it on the list so that 602 * the task leader will wait for this new process 603 * to commit suicide. 604 */ 605 PROC_LOCK(p2); 606 kern_psignal(p2, SIGKILL); 607 PROC_UNLOCK(p2); 608 } else 609 PROC_UNLOCK(p1->p_leader); 610 } else { 611 p2->p_peers = NULL; 612 p2->p_leader = p2; 613 } 614 615 sx_xlock(&proctree_lock); 616 PGRP_LOCK(p1->p_pgrp); 617 PROC_LOCK(p2); 618 PROC_LOCK(p1); 619 620 /* 621 * Preserve some more flags in subprocess. P_PROFIL has already 622 * been preserved. 623 */ 624 p2->p_flag |= p1->p_flag & P_SUGID; 625 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK)); 626 SESS_LOCK(p1->p_session); 627 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 628 p2->p_flag |= P_CONTROLT; 629 SESS_UNLOCK(p1->p_session); 630 if (fr->fr_flags & RFPPWAIT) 631 p2->p_flag |= P_PPWAIT; 632 633 p2->p_pgrp = p1->p_pgrp; 634 LIST_INSERT_AFTER(p1, p2, p_pglist); 635 PGRP_UNLOCK(p1->p_pgrp); 636 LIST_INIT(&p2->p_children); 637 LIST_INIT(&p2->p_orphans); 638 639 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 640 TAILQ_INIT(&p2->p_kqtim_stop); 641 642 /* 643 * This begins the section where we must prevent the parent 644 * from being swapped. 645 */ 646 _PHOLD(p1); 647 PROC_UNLOCK(p1); 648 649 /* 650 * Attach the new process to its parent. 651 * 652 * If RFNOWAIT is set, the newly created process becomes a child 653 * of init. This effectively disassociates the child from the 654 * parent. 655 */ 656 if ((fr->fr_flags & RFNOWAIT) != 0) { 657 pptr = p1->p_reaper; 658 p2->p_reaper = pptr; 659 } else { 660 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 661 p1 : p1->p_reaper; 662 pptr = p1; 663 } 664 p2->p_pptr = pptr; 665 p2->p_oppid = pptr->p_pid; 666 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 667 LIST_INIT(&p2->p_reaplist); 668 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 669 if (p2->p_reaper == p1 && p1 != initproc) { 670 p2->p_reapsubtree = p2->p_pid; 671 proc_id_set_cond(PROC_ID_REAP, p2->p_pid); 672 } 673 sx_xunlock(&proctree_lock); 674 675 /* Inform accounting that we have forked. */ 676 p2->p_acflag = AFORK; 677 PROC_UNLOCK(p2); 678 679 #ifdef KTRACE 680 ktrprocfork(p1, p2); 681 #endif 682 683 /* 684 * Finish creating the child process. It will return via a different 685 * execution path later. (ie: directly into user mode) 686 */ 687 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 688 689 if (fr->fr_flags == (RFFDG | RFPROC)) { 690 VM_CNT_INC(v_forks); 691 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 692 p2->p_vmspace->vm_ssize); 693 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 694 VM_CNT_INC(v_vforks); 695 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 696 p2->p_vmspace->vm_ssize); 697 } else if (p1 == &proc0) { 698 VM_CNT_INC(v_kthreads); 699 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 700 p2->p_vmspace->vm_ssize); 701 } else { 702 VM_CNT_INC(v_rforks); 703 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 704 p2->p_vmspace->vm_ssize); 705 } 706 707 /* 708 * Associate the process descriptor with the process before anything 709 * can happen that might cause that process to need the descriptor. 710 * However, don't do this until after fork(2) can no longer fail. 711 */ 712 if (fr->fr_flags & RFPROCDESC) 713 procdesc_new(p2, fr->fr_pd_flags); 714 715 /* 716 * Both processes are set up, now check if any loadable modules want 717 * to adjust anything. 718 */ 719 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 720 721 /* 722 * Set the child start time and mark the process as being complete. 723 */ 724 PROC_LOCK(p2); 725 PROC_LOCK(p1); 726 microuptime(&p2->p_stats->p_start); 727 PROC_SLOCK(p2); 728 p2->p_state = PRS_NORMAL; 729 PROC_SUNLOCK(p2); 730 731 #ifdef KDTRACE_HOOKS 732 /* 733 * Tell the DTrace fasttrap provider about the new process so that any 734 * tracepoints inherited from the parent can be removed. We have to do 735 * this only after p_state is PRS_NORMAL since the fasttrap module will 736 * use pfind() later on. 737 */ 738 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 739 dtrace_fasttrap_fork(p1, p2); 740 #endif 741 if (fr->fr_flags & RFPPWAIT) { 742 td->td_pflags |= TDP_RFPPWAIT; 743 td->td_rfppwait_p = p2; 744 td->td_dbgflags |= TDB_VFORK; 745 } 746 PROC_UNLOCK(p2); 747 748 /* 749 * Tell any interested parties about the new process. 750 */ 751 knote_fork(p1->p_klist, p2->p_pid); 752 753 /* 754 * Now can be swapped. 755 */ 756 _PRELE(p1); 757 PROC_UNLOCK(p1); 758 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 759 760 if (fr->fr_flags & RFPROCDESC) { 761 procdesc_finit(p2->p_procdesc, fp_procdesc); 762 fdrop(fp_procdesc, td); 763 } 764 765 /* 766 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 767 * synced with forks in progress so it is OK if we miss it 768 * if being set atm. 769 */ 770 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 771 sx_xlock(&proctree_lock); 772 PROC_LOCK(p2); 773 774 /* 775 * p1->p_ptevents & p1->p_pptr are protected by both 776 * process and proctree locks for modifications, 777 * so owning proctree_lock allows the race-free read. 778 */ 779 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 780 /* 781 * Arrange for debugger to receive the fork event. 782 * 783 * We can report PL_FLAG_FORKED regardless of 784 * P_FOLLOWFORK settings, but it does not make a sense 785 * for runaway child. 786 */ 787 td->td_dbgflags |= TDB_FORK; 788 td->td_dbg_forked = p2->p_pid; 789 td2->td_dbgflags |= TDB_STOPATFORK; 790 proc_set_traced(p2, true); 791 CTR2(KTR_PTRACE, 792 "do_fork: attaching to new child pid %d: oppid %d", 793 p2->p_pid, p2->p_oppid); 794 proc_reparent(p2, p1->p_pptr, false); 795 } 796 PROC_UNLOCK(p2); 797 sx_xunlock(&proctree_lock); 798 } 799 800 racct_proc_fork_done(p2); 801 802 if ((fr->fr_flags & RFSTOPPED) == 0) { 803 if (fr->fr_pidp != NULL) 804 *fr->fr_pidp = p2->p_pid; 805 /* 806 * If RFSTOPPED not requested, make child runnable and 807 * add to run queue. 808 */ 809 thread_lock(td2); 810 TD_SET_CAN_RUN(td2); 811 sched_add(td2, SRQ_BORING); 812 } else { 813 *fr->fr_procp = p2; 814 } 815 } 816 817 void 818 fork_rfppwait(struct thread *td) 819 { 820 struct proc *p, *p2; 821 822 MPASS(td->td_pflags & TDP_RFPPWAIT); 823 824 p = td->td_proc; 825 /* 826 * Preserve synchronization semantics of vfork. If 827 * waiting for child to exec or exit, fork set 828 * P_PPWAIT on child, and there we sleep on our proc 829 * (in case of exit). 830 * 831 * Do it after the ptracestop() above is finished, to 832 * not block our debugger until child execs or exits 833 * to finish vfork wait. 834 */ 835 td->td_pflags &= ~TDP_RFPPWAIT; 836 p2 = td->td_rfppwait_p; 837 again: 838 PROC_LOCK(p2); 839 while (p2->p_flag & P_PPWAIT) { 840 PROC_LOCK(p); 841 if (thread_suspend_check_needed()) { 842 PROC_UNLOCK(p2); 843 thread_suspend_check(0); 844 PROC_UNLOCK(p); 845 goto again; 846 } else { 847 PROC_UNLOCK(p); 848 } 849 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz); 850 } 851 PROC_UNLOCK(p2); 852 853 if (td->td_dbgflags & TDB_VFORK) { 854 PROC_LOCK(p); 855 if (p->p_ptevents & PTRACE_VFORK) 856 ptracestop(td, SIGTRAP, NULL); 857 td->td_dbgflags &= ~TDB_VFORK; 858 PROC_UNLOCK(p); 859 } 860 } 861 862 int 863 fork1(struct thread *td, struct fork_req *fr) 864 { 865 struct proc *p1, *newproc; 866 struct thread *td2; 867 struct vmspace *vm2; 868 struct ucred *cred; 869 struct file *fp_procdesc; 870 vm_ooffset_t mem_charged; 871 int error, nprocs_new; 872 static int curfail; 873 static struct timeval lastfail; 874 int flags, pages; 875 876 flags = fr->fr_flags; 877 pages = fr->fr_pages; 878 879 if ((flags & RFSTOPPED) != 0) 880 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 881 else 882 MPASS(fr->fr_procp == NULL); 883 884 /* Check for the undefined or unimplemented flags. */ 885 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 886 return (EINVAL); 887 888 /* Signal value requires RFTSIGZMB. */ 889 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 890 return (EINVAL); 891 892 /* Can't copy and clear. */ 893 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 894 return (EINVAL); 895 896 /* Check the validity of the signal number. */ 897 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 898 return (EINVAL); 899 900 if ((flags & RFPROCDESC) != 0) { 901 /* Can't not create a process yet get a process descriptor. */ 902 if ((flags & RFPROC) == 0) 903 return (EINVAL); 904 905 /* Must provide a place to put a procdesc if creating one. */ 906 if (fr->fr_pd_fd == NULL) 907 return (EINVAL); 908 909 /* Check if we are using supported flags. */ 910 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 911 return (EINVAL); 912 } 913 914 p1 = td->td_proc; 915 916 /* 917 * Here we don't create a new process, but we divorce 918 * certain parts of a process from itself. 919 */ 920 if ((flags & RFPROC) == 0) { 921 if (fr->fr_procp != NULL) 922 *fr->fr_procp = NULL; 923 else if (fr->fr_pidp != NULL) 924 *fr->fr_pidp = 0; 925 return (fork_norfproc(td, flags)); 926 } 927 928 fp_procdesc = NULL; 929 newproc = NULL; 930 vm2 = NULL; 931 932 /* 933 * Increment the nprocs resource before allocations occur. 934 * Although process entries are dynamically created, we still 935 * keep a global limit on the maximum number we will 936 * create. There are hard-limits as to the number of processes 937 * that can run, established by the KVA and memory usage for 938 * the process data. 939 * 940 * Don't allow a nonprivileged user to use the last ten 941 * processes; don't let root exceed the limit. 942 */ 943 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 944 if (nprocs_new >= maxproc - 10) { 945 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 || 946 nprocs_new >= maxproc) { 947 error = EAGAIN; 948 sx_xlock(&allproc_lock); 949 if (ppsratecheck(&lastfail, &curfail, 1)) { 950 printf("maxproc limit exceeded by uid %u " 951 "(pid %d); see tuning(7) and " 952 "login.conf(5)\n", 953 td->td_ucred->cr_ruid, p1->p_pid); 954 } 955 sx_xunlock(&allproc_lock); 956 goto fail2; 957 } 958 } 959 960 /* 961 * If required, create a process descriptor in the parent first; we 962 * will abandon it if something goes wrong. We don't finit() until 963 * later. 964 */ 965 if (flags & RFPROCDESC) { 966 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 967 fr->fr_pd_flags, fr->fr_pd_fcaps); 968 if (error != 0) 969 goto fail2; 970 AUDIT_ARG_FD(*fr->fr_pd_fd); 971 } 972 973 mem_charged = 0; 974 if (pages == 0) 975 pages = kstack_pages; 976 /* Allocate new proc. */ 977 newproc = uma_zalloc(proc_zone, M_WAITOK); 978 td2 = FIRST_THREAD_IN_PROC(newproc); 979 if (td2 == NULL) { 980 td2 = thread_alloc(pages); 981 if (td2 == NULL) { 982 error = ENOMEM; 983 goto fail2; 984 } 985 proc_linkup(newproc, td2); 986 } else { 987 kmsan_thread_alloc(td2); 988 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 989 if (td2->td_kstack != 0) 990 vm_thread_dispose(td2); 991 if (!thread_alloc_stack(td2, pages)) { 992 error = ENOMEM; 993 goto fail2; 994 } 995 } 996 } 997 998 if ((flags & RFMEM) == 0) { 999 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 1000 if (vm2 == NULL) { 1001 error = ENOMEM; 1002 goto fail2; 1003 } 1004 if (!swap_reserve(mem_charged)) { 1005 /* 1006 * The swap reservation failed. The accounting 1007 * from the entries of the copied vm2 will be 1008 * subtracted in vmspace_free(), so force the 1009 * reservation there. 1010 */ 1011 swap_reserve_force(mem_charged); 1012 error = ENOMEM; 1013 goto fail2; 1014 } 1015 } else 1016 vm2 = NULL; 1017 1018 /* 1019 * XXX: This is ugly; when we copy resource usage, we need to bump 1020 * per-cred resource counters. 1021 */ 1022 proc_set_cred_init(newproc, td->td_ucred); 1023 1024 /* 1025 * Initialize resource accounting for the child process. 1026 */ 1027 error = racct_proc_fork(p1, newproc); 1028 if (error != 0) { 1029 error = EAGAIN; 1030 goto fail1; 1031 } 1032 1033 #ifdef MAC 1034 mac_proc_init(newproc); 1035 #endif 1036 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 1037 STAILQ_INIT(&newproc->p_ktr); 1038 1039 /* 1040 * Increment the count of procs running with this uid. Don't allow 1041 * a nonprivileged user to exceed their current limit. 1042 */ 1043 cred = td->td_ucred; 1044 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) { 1045 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0) 1046 goto fail0; 1047 chgproccnt(cred->cr_ruidinfo, 1, 0); 1048 } 1049 1050 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 1051 return (0); 1052 fail0: 1053 error = EAGAIN; 1054 #ifdef MAC 1055 mac_proc_destroy(newproc); 1056 #endif 1057 racct_proc_exit(newproc); 1058 fail1: 1059 proc_unset_cred(newproc); 1060 fail2: 1061 if (vm2 != NULL) 1062 vmspace_free(vm2); 1063 uma_zfree(proc_zone, newproc); 1064 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1065 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1066 fdrop(fp_procdesc, td); 1067 } 1068 atomic_add_int(&nprocs, -1); 1069 pause("fork", hz / 2); 1070 return (error); 1071 } 1072 1073 /* 1074 * Handle the return of a child process from fork1(). This function 1075 * is called from the MD fork_trampoline() entry point. 1076 */ 1077 void 1078 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1079 struct trapframe *frame) 1080 { 1081 struct proc *p; 1082 struct thread *td; 1083 struct thread *dtd; 1084 1085 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED); 1086 1087 td = curthread; 1088 p = td->td_proc; 1089 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1090 1091 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1092 td, td_get_sched(td), p->p_pid, td->td_name); 1093 1094 sched_fork_exit(td); 1095 /* 1096 * Processes normally resume in mi_switch() after being 1097 * cpu_switch()'ed to, but when children start up they arrive here 1098 * instead, so we must do much the same things as mi_switch() would. 1099 */ 1100 if ((dtd = PCPU_GET(deadthread))) { 1101 PCPU_SET(deadthread, NULL); 1102 thread_stash(dtd); 1103 } 1104 thread_unlock(td); 1105 1106 /* 1107 * cpu_fork_kthread_handler intercepts this function call to 1108 * have this call a non-return function to stay in kernel mode. 1109 * initproc has its own fork handler, but it does return. 1110 */ 1111 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1112 callout(arg, frame); 1113 1114 /* 1115 * Check if a kernel thread misbehaved and returned from its main 1116 * function. 1117 */ 1118 if (p->p_flag & P_KPROC) { 1119 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1120 td->td_name, p->p_pid); 1121 kthread_exit(); 1122 } 1123 mtx_assert(&Giant, MA_NOTOWNED); 1124 1125 if (p->p_sysent->sv_schedtail != NULL) 1126 (p->p_sysent->sv_schedtail)(td); 1127 } 1128 1129 /* 1130 * Simplified back end of syscall(), used when returning from fork() 1131 * directly into user mode. This function is passed in to fork_exit() 1132 * as the first parameter and is called when returning to a new 1133 * userland process. 1134 */ 1135 void 1136 fork_return(struct thread *td, struct trapframe *frame) 1137 { 1138 struct proc *p; 1139 1140 p = td->td_proc; 1141 if (td->td_dbgflags & TDB_STOPATFORK) { 1142 PROC_LOCK(p); 1143 if ((p->p_flag & P_TRACED) != 0) { 1144 /* 1145 * Inform the debugger if one is still present. 1146 */ 1147 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1148 ptracestop(td, SIGSTOP, NULL); 1149 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1150 } else { 1151 /* 1152 * ... otherwise clear the request. 1153 */ 1154 td->td_dbgflags &= ~TDB_STOPATFORK; 1155 } 1156 PROC_UNLOCK(p); 1157 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1158 /* 1159 * This is the start of a new thread in a traced 1160 * process. Report a system call exit event. 1161 */ 1162 PROC_LOCK(p); 1163 td->td_dbgflags |= TDB_SCX; 1164 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1165 (td->td_dbgflags & TDB_BORN) != 0) 1166 ptracestop(td, SIGTRAP, NULL); 1167 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1168 PROC_UNLOCK(p); 1169 } 1170 1171 /* 1172 * If the prison was killed mid-fork, die along with it. 1173 */ 1174 if (!prison_isalive(td->td_ucred->cr_prison)) 1175 exit1(td, 0, SIGKILL); 1176 1177 userret(td, frame); 1178 1179 #ifdef KTRACE 1180 if (KTRPOINT(td, KTR_SYSRET)) 1181 ktrsysret(SYS_fork, 0, 0); 1182 #endif 1183 } 1184