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