1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ktrace.h" 43 #include "opt_kstack_pages.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/bitstring.h> 48 #include <sys/sysproto.h> 49 #include <sys/eventhandler.h> 50 #include <sys/fcntl.h> 51 #include <sys/filedesc.h> 52 #include <sys/jail.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/sysctl.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mutex.h> 59 #include <sys/priv.h> 60 #include <sys/proc.h> 61 #include <sys/procdesc.h> 62 #include <sys/pioctl.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 /* 133 * It is necessary to return fd by reference because 0 is a valid file 134 * descriptor number, and the child needs to be able to distinguish 135 * itself from the parent using the return value. 136 */ 137 error = fork1(td, &fr); 138 if (error == 0) { 139 td->td_retval[0] = pid; 140 td->td_retval[1] = 0; 141 error = copyout(&fd, uap->fdp, sizeof(fd)); 142 } 143 return (error); 144 } 145 146 /* ARGSUSED */ 147 int 148 sys_vfork(struct thread *td, struct vfork_args *uap) 149 { 150 struct fork_req fr; 151 int error, pid; 152 153 bzero(&fr, sizeof(fr)); 154 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 155 fr.fr_pidp = &pid; 156 error = fork1(td, &fr); 157 if (error == 0) { 158 td->td_retval[0] = pid; 159 td->td_retval[1] = 0; 160 } 161 return (error); 162 } 163 164 int 165 sys_rfork(struct thread *td, struct rfork_args *uap) 166 { 167 struct fork_req fr; 168 int error, pid; 169 170 /* Don't allow kernel-only flags. */ 171 if ((uap->flags & RFKERNELONLY) != 0) 172 return (EINVAL); 173 174 AUDIT_ARG_FFLAGS(uap->flags); 175 bzero(&fr, sizeof(fr)); 176 fr.fr_flags = uap->flags; 177 fr.fr_pidp = &pid; 178 error = fork1(td, &fr); 179 if (error == 0) { 180 td->td_retval[0] = pid; 181 td->td_retval[1] = 0; 182 } 183 return (error); 184 } 185 186 int __exclusive_cache_line nprocs = 1; /* process 0 */ 187 int lastpid = 0; 188 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 189 "Last used PID"); 190 191 /* 192 * Random component to lastpid generation. We mix in a random factor to make 193 * it a little harder to predict. We sanity check the modulus value to avoid 194 * doing it in critical paths. Don't let it be too small or we pointlessly 195 * waste randomness entropy, and don't let it be impossibly large. Using a 196 * modulus that is too big causes a LOT more process table scans and slows 197 * down fork processing as the pidchecked caching is defeated. 198 */ 199 static int randompid = 0; 200 201 static int 202 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 203 { 204 int error, pid; 205 206 error = sysctl_wire_old_buffer(req, sizeof(int)); 207 if (error != 0) 208 return(error); 209 sx_xlock(&allproc_lock); 210 pid = randompid; 211 error = sysctl_handle_int(oidp, &pid, 0, req); 212 if (error == 0 && req->newptr != NULL) { 213 if (pid == 0) 214 randompid = 0; 215 else if (pid == 1) 216 /* generate a random PID modulus between 100 and 1123 */ 217 randompid = 100 + arc4random() % 1024; 218 else if (pid < 0 || pid > pid_max - 100) 219 /* out of range */ 220 randompid = pid_max - 100; 221 else if (pid < 100) 222 /* Make it reasonable */ 223 randompid = 100; 224 else 225 randompid = pid; 226 } 227 sx_xunlock(&allproc_lock); 228 return (error); 229 } 230 231 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 232 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value"); 233 234 extern bitstr_t proc_id_pidmap; 235 extern bitstr_t proc_id_grpidmap; 236 extern bitstr_t proc_id_sessidmap; 237 extern bitstr_t proc_id_reapmap; 238 239 /* 240 * Find an unused process ID 241 * 242 * If RFHIGHPID is set (used during system boot), do not allocate 243 * low-numbered pids. 244 */ 245 static int 246 fork_findpid(int flags) 247 { 248 pid_t result; 249 int trypid, random; 250 251 /* 252 * Avoid calling arc4random with procid_lock held. 253 */ 254 random = 0; 255 if (__predict_false(randompid)) 256 random = arc4random() % randompid; 257 258 mtx_lock(&procid_lock); 259 260 trypid = lastpid + 1; 261 if (flags & RFHIGHPID) { 262 if (trypid < 10) 263 trypid = 10; 264 } else { 265 trypid += random; 266 } 267 retry: 268 if (trypid >= pid_max) 269 trypid = 2; 270 271 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result); 272 if (result == -1) { 273 KASSERT(trypid != 2, ("unexpectedly ran out of IDs")); 274 trypid = 2; 275 goto retry; 276 } 277 if (bit_test(&proc_id_grpidmap, result) || 278 bit_test(&proc_id_sessidmap, result) || 279 bit_test(&proc_id_reapmap, result)) { 280 trypid = result + 1; 281 goto retry; 282 } 283 284 /* 285 * RFHIGHPID does not mess with the lastpid counter during boot. 286 */ 287 if ((flags & RFHIGHPID) == 0) 288 lastpid = result; 289 290 bit_set(&proc_id_pidmap, result); 291 mtx_unlock(&procid_lock); 292 293 return (result); 294 } 295 296 static int 297 fork_norfproc(struct thread *td, int flags) 298 { 299 int error; 300 struct proc *p1; 301 302 KASSERT((flags & RFPROC) == 0, 303 ("fork_norfproc called with RFPROC set")); 304 p1 = td->td_proc; 305 306 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 307 (flags & (RFCFDG | RFFDG))) { 308 PROC_LOCK(p1); 309 if (thread_single(p1, SINGLE_BOUNDARY)) { 310 PROC_UNLOCK(p1); 311 return (ERESTART); 312 } 313 PROC_UNLOCK(p1); 314 } 315 316 error = vm_forkproc(td, NULL, NULL, NULL, flags); 317 if (error) 318 goto fail; 319 320 /* 321 * Close all file descriptors. 322 */ 323 if (flags & RFCFDG) { 324 struct filedesc *fdtmp; 325 fdtmp = fdinit(td->td_proc->p_fd, false); 326 fdescfree(td); 327 p1->p_fd = fdtmp; 328 } 329 330 /* 331 * Unshare file descriptors (from parent). 332 */ 333 if (flags & RFFDG) 334 fdunshare(td); 335 336 fail: 337 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 338 (flags & (RFCFDG | RFFDG))) { 339 PROC_LOCK(p1); 340 thread_single_end(p1, SINGLE_BOUNDARY); 341 PROC_UNLOCK(p1); 342 } 343 return (error); 344 } 345 346 static void 347 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 348 struct vmspace *vm2, struct file *fp_procdesc) 349 { 350 struct proc *p1, *pptr; 351 struct filedesc *fd; 352 struct filedesc_to_leader *fdtol; 353 struct sigacts *newsigacts; 354 355 p1 = td->td_proc; 356 357 PROC_LOCK(p1); 358 bcopy(&p1->p_startcopy, &p2->p_startcopy, 359 __rangeof(struct proc, p_startcopy, p_endcopy)); 360 pargs_hold(p2->p_args); 361 PROC_UNLOCK(p1); 362 363 bzero(&p2->p_startzero, 364 __rangeof(struct proc, p_startzero, p_endzero)); 365 366 /* Tell the prison that we exist. */ 367 prison_proc_hold(p2->p_ucred->cr_prison); 368 369 p2->p_state = PRS_NEW; /* protect against others */ 370 p2->p_pid = fork_findpid(fr->fr_flags); 371 AUDIT_ARG_PID(p2->p_pid); 372 373 sx_xlock(&allproc_lock); 374 LIST_INSERT_HEAD(&allproc, p2, p_list); 375 allproc_gen++; 376 sx_xunlock(&allproc_lock); 377 378 sx_xlock(PIDHASHLOCK(p2->p_pid)); 379 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 380 sx_xunlock(PIDHASHLOCK(p2->p_pid)); 381 382 tidhash_add(td2); 383 384 /* 385 * Malloc things while we don't hold any locks. 386 */ 387 if (fr->fr_flags & RFSIGSHARE) 388 newsigacts = NULL; 389 else 390 newsigacts = sigacts_alloc(); 391 392 /* 393 * Copy filedesc. 394 */ 395 if (fr->fr_flags & RFCFDG) { 396 fd = fdinit(p1->p_fd, false); 397 fdtol = NULL; 398 } else if (fr->fr_flags & RFFDG) { 399 fd = fdcopy(p1->p_fd); 400 fdtol = NULL; 401 } else { 402 fd = fdshare(p1->p_fd); 403 if (p1->p_fdtol == NULL) 404 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 405 p1->p_leader); 406 if ((fr->fr_flags & RFTHREAD) != 0) { 407 /* 408 * Shared file descriptor table, and shared 409 * process leaders. 410 */ 411 fdtol = p1->p_fdtol; 412 FILEDESC_XLOCK(p1->p_fd); 413 fdtol->fdl_refcount++; 414 FILEDESC_XUNLOCK(p1->p_fd); 415 } else { 416 /* 417 * Shared file descriptor table, and different 418 * process leaders. 419 */ 420 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 421 p1->p_fd, p2); 422 } 423 } 424 /* 425 * Make a proc table entry for the new process. 426 * Start by zeroing the section of proc that is zero-initialized, 427 * then copy the section that is copied directly from the parent. 428 */ 429 430 PROC_LOCK(p2); 431 PROC_LOCK(p1); 432 433 bzero(&td2->td_startzero, 434 __rangeof(struct thread, td_startzero, td_endzero)); 435 436 bcopy(&td->td_startcopy, &td2->td_startcopy, 437 __rangeof(struct thread, td_startcopy, td_endcopy)); 438 439 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 440 td2->td_sigstk = td->td_sigstk; 441 td2->td_flags = TDF_INMEM; 442 td2->td_lend_user_pri = PRI_MAX; 443 444 #ifdef VIMAGE 445 td2->td_vnet = NULL; 446 td2->td_vnet_lpush = NULL; 447 #endif 448 449 /* 450 * Allow the scheduler to initialize the child. 451 */ 452 thread_lock(td); 453 sched_fork(td, td2); 454 thread_unlock(td); 455 456 /* 457 * Duplicate sub-structures as needed. 458 * Increase reference counts on shared objects. 459 */ 460 p2->p_flag = P_INMEM; 461 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE | 462 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | 463 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP); 464 p2->p_swtick = ticks; 465 if (p1->p_flag & P_PROFIL) 466 startprofclock(p2); 467 468 if (fr->fr_flags & RFSIGSHARE) { 469 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 470 } else { 471 sigacts_copy(newsigacts, p1->p_sigacts); 472 p2->p_sigacts = newsigacts; 473 } 474 475 if (fr->fr_flags & RFTSIGZMB) 476 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 477 else if (fr->fr_flags & RFLINUXTHPN) 478 p2->p_sigparent = SIGUSR1; 479 else 480 p2->p_sigparent = SIGCHLD; 481 482 p2->p_textvp = p1->p_textvp; 483 p2->p_fd = fd; 484 p2->p_fdtol = fdtol; 485 486 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 487 p2->p_flag |= P_PROTECTED; 488 p2->p_flag2 |= P2_INHERIT_PROTECTED; 489 } 490 491 /* 492 * p_limit is copy-on-write. Bump its refcount. 493 */ 494 lim_fork(p1, p2); 495 496 thread_cow_get_proc(td2, p2); 497 498 pstats_fork(p1->p_stats, p2->p_stats); 499 500 PROC_UNLOCK(p1); 501 PROC_UNLOCK(p2); 502 503 /* Bump references to the text vnode (for procfs). */ 504 if (p2->p_textvp) 505 vrefact(p2->p_textvp); 506 507 /* 508 * Set up linkage for kernel based threading. 509 */ 510 if ((fr->fr_flags & RFTHREAD) != 0) { 511 mtx_lock(&ppeers_lock); 512 p2->p_peers = p1->p_peers; 513 p1->p_peers = p2; 514 p2->p_leader = p1->p_leader; 515 mtx_unlock(&ppeers_lock); 516 PROC_LOCK(p1->p_leader); 517 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 518 PROC_UNLOCK(p1->p_leader); 519 /* 520 * The task leader is exiting, so process p1 is 521 * going to be killed shortly. Since p1 obviously 522 * isn't dead yet, we know that the leader is either 523 * sending SIGKILL's to all the processes in this 524 * task or is sleeping waiting for all the peers to 525 * exit. We let p1 complete the fork, but we need 526 * to go ahead and kill the new process p2 since 527 * the task leader may not get a chance to send 528 * SIGKILL to it. We leave it on the list so that 529 * the task leader will wait for this new process 530 * to commit suicide. 531 */ 532 PROC_LOCK(p2); 533 kern_psignal(p2, SIGKILL); 534 PROC_UNLOCK(p2); 535 } else 536 PROC_UNLOCK(p1->p_leader); 537 } else { 538 p2->p_peers = NULL; 539 p2->p_leader = p2; 540 } 541 542 sx_xlock(&proctree_lock); 543 PGRP_LOCK(p1->p_pgrp); 544 PROC_LOCK(p2); 545 PROC_LOCK(p1); 546 547 /* 548 * Preserve some more flags in subprocess. P_PROFIL has already 549 * been preserved. 550 */ 551 p2->p_flag |= p1->p_flag & P_SUGID; 552 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 553 SESS_LOCK(p1->p_session); 554 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 555 p2->p_flag |= P_CONTROLT; 556 SESS_UNLOCK(p1->p_session); 557 if (fr->fr_flags & RFPPWAIT) 558 p2->p_flag |= P_PPWAIT; 559 560 p2->p_pgrp = p1->p_pgrp; 561 LIST_INSERT_AFTER(p1, p2, p_pglist); 562 PGRP_UNLOCK(p1->p_pgrp); 563 LIST_INIT(&p2->p_children); 564 LIST_INIT(&p2->p_orphans); 565 566 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 567 568 /* 569 * If PF_FORK is set, the child process inherits the 570 * procfs ioctl flags from its parent. 571 */ 572 if (p1->p_pfsflags & PF_FORK) { 573 p2->p_stops = p1->p_stops; 574 p2->p_pfsflags = p1->p_pfsflags; 575 } 576 577 /* 578 * This begins the section where we must prevent the parent 579 * from being swapped. 580 */ 581 _PHOLD(p1); 582 PROC_UNLOCK(p1); 583 584 /* 585 * Attach the new process to its parent. 586 * 587 * If RFNOWAIT is set, the newly created process becomes a child 588 * of init. This effectively disassociates the child from the 589 * parent. 590 */ 591 if ((fr->fr_flags & RFNOWAIT) != 0) { 592 pptr = p1->p_reaper; 593 p2->p_reaper = pptr; 594 } else { 595 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 596 p1 : p1->p_reaper; 597 pptr = p1; 598 } 599 p2->p_pptr = pptr; 600 p2->p_oppid = pptr->p_pid; 601 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 602 LIST_INIT(&p2->p_reaplist); 603 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 604 if (p2->p_reaper == p1 && p1 != initproc) { 605 p2->p_reapsubtree = p2->p_pid; 606 proc_id_set_cond(PROC_ID_REAP, p2->p_pid); 607 } 608 sx_xunlock(&proctree_lock); 609 610 /* Inform accounting that we have forked. */ 611 p2->p_acflag = AFORK; 612 PROC_UNLOCK(p2); 613 614 #ifdef KTRACE 615 ktrprocfork(p1, p2); 616 #endif 617 618 /* 619 * Finish creating the child process. It will return via a different 620 * execution path later. (ie: directly into user mode) 621 */ 622 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 623 624 if (fr->fr_flags == (RFFDG | RFPROC)) { 625 VM_CNT_INC(v_forks); 626 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize + 627 p2->p_vmspace->vm_ssize); 628 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 629 VM_CNT_INC(v_vforks); 630 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize + 631 p2->p_vmspace->vm_ssize); 632 } else if (p1 == &proc0) { 633 VM_CNT_INC(v_kthreads); 634 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize + 635 p2->p_vmspace->vm_ssize); 636 } else { 637 VM_CNT_INC(v_rforks); 638 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize + 639 p2->p_vmspace->vm_ssize); 640 } 641 642 /* 643 * Associate the process descriptor with the process before anything 644 * can happen that might cause that process to need the descriptor. 645 * However, don't do this until after fork(2) can no longer fail. 646 */ 647 if (fr->fr_flags & RFPROCDESC) 648 procdesc_new(p2, fr->fr_pd_flags); 649 650 /* 651 * Both processes are set up, now check if any loadable modules want 652 * to adjust anything. 653 */ 654 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags); 655 656 /* 657 * Set the child start time and mark the process as being complete. 658 */ 659 PROC_LOCK(p2); 660 PROC_LOCK(p1); 661 microuptime(&p2->p_stats->p_start); 662 PROC_SLOCK(p2); 663 p2->p_state = PRS_NORMAL; 664 PROC_SUNLOCK(p2); 665 666 #ifdef KDTRACE_HOOKS 667 /* 668 * Tell the DTrace fasttrap provider about the new process so that any 669 * tracepoints inherited from the parent can be removed. We have to do 670 * this only after p_state is PRS_NORMAL since the fasttrap module will 671 * use pfind() later on. 672 */ 673 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 674 dtrace_fasttrap_fork(p1, p2); 675 #endif 676 if (fr->fr_flags & RFPPWAIT) { 677 td->td_pflags |= TDP_RFPPWAIT; 678 td->td_rfppwait_p = p2; 679 td->td_dbgflags |= TDB_VFORK; 680 } 681 PROC_UNLOCK(p2); 682 683 /* 684 * Tell any interested parties about the new process. 685 */ 686 knote_fork(p1->p_klist, p2->p_pid); 687 688 /* 689 * Now can be swapped. 690 */ 691 _PRELE(p1); 692 PROC_UNLOCK(p1); 693 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 694 695 if (fr->fr_flags & RFPROCDESC) { 696 procdesc_finit(p2->p_procdesc, fp_procdesc); 697 fdrop(fp_procdesc, td); 698 } 699 700 /* 701 * Speculative check for PTRACE_FORK. PTRACE_FORK is not 702 * synced with forks in progress so it is OK if we miss it 703 * if being set atm. 704 */ 705 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 706 sx_xlock(&proctree_lock); 707 PROC_LOCK(p2); 708 709 /* 710 * p1->p_ptevents & p1->p_pptr are protected by both 711 * process and proctree locks for modifications, 712 * so owning proctree_lock allows the race-free read. 713 */ 714 if ((p1->p_ptevents & PTRACE_FORK) != 0) { 715 /* 716 * Arrange for debugger to receive the fork event. 717 * 718 * We can report PL_FLAG_FORKED regardless of 719 * P_FOLLOWFORK settings, but it does not make a sense 720 * for runaway child. 721 */ 722 td->td_dbgflags |= TDB_FORK; 723 td->td_dbg_forked = p2->p_pid; 724 td2->td_dbgflags |= TDB_STOPATFORK; 725 proc_set_traced(p2, true); 726 CTR2(KTR_PTRACE, 727 "do_fork: attaching to new child pid %d: oppid %d", 728 p2->p_pid, p2->p_oppid); 729 proc_reparent(p2, p1->p_pptr, false); 730 } 731 PROC_UNLOCK(p2); 732 sx_xunlock(&proctree_lock); 733 } 734 735 racct_proc_fork_done(p2); 736 737 if ((fr->fr_flags & RFSTOPPED) == 0) { 738 if (fr->fr_pidp != NULL) 739 *fr->fr_pidp = p2->p_pid; 740 /* 741 * If RFSTOPPED not requested, make child runnable and 742 * add to run queue. 743 */ 744 thread_lock(td2); 745 TD_SET_CAN_RUN(td2); 746 sched_add(td2, SRQ_BORING); 747 thread_unlock(td2); 748 } else { 749 *fr->fr_procp = p2; 750 } 751 } 752 753 void 754 fork_rfppwait(struct thread *td) 755 { 756 struct proc *p, *p2; 757 758 MPASS(td->td_pflags & TDP_RFPPWAIT); 759 760 p = td->td_proc; 761 /* 762 * Preserve synchronization semantics of vfork. If 763 * waiting for child to exec or exit, fork set 764 * P_PPWAIT on child, and there we sleep on our proc 765 * (in case of exit). 766 * 767 * Do it after the ptracestop() above is finished, to 768 * not block our debugger until child execs or exits 769 * to finish vfork wait. 770 */ 771 td->td_pflags &= ~TDP_RFPPWAIT; 772 p2 = td->td_rfppwait_p; 773 again: 774 PROC_LOCK(p2); 775 while (p2->p_flag & P_PPWAIT) { 776 PROC_LOCK(p); 777 if (thread_suspend_check_needed()) { 778 PROC_UNLOCK(p2); 779 thread_suspend_check(0); 780 PROC_UNLOCK(p); 781 goto again; 782 } else { 783 PROC_UNLOCK(p); 784 } 785 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz); 786 } 787 PROC_UNLOCK(p2); 788 789 if (td->td_dbgflags & TDB_VFORK) { 790 PROC_LOCK(p); 791 if (p->p_ptevents & PTRACE_VFORK) 792 ptracestop(td, SIGTRAP, NULL); 793 td->td_dbgflags &= ~TDB_VFORK; 794 PROC_UNLOCK(p); 795 } 796 } 797 798 int 799 fork1(struct thread *td, struct fork_req *fr) 800 { 801 struct proc *p1, *newproc; 802 struct thread *td2; 803 struct vmspace *vm2; 804 struct ucred *cred; 805 struct file *fp_procdesc; 806 vm_ooffset_t mem_charged; 807 int error, nprocs_new; 808 static int curfail; 809 static struct timeval lastfail; 810 int flags, pages; 811 812 flags = fr->fr_flags; 813 pages = fr->fr_pages; 814 815 if ((flags & RFSTOPPED) != 0) 816 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 817 else 818 MPASS(fr->fr_procp == NULL); 819 820 /* Check for the undefined or unimplemented flags. */ 821 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 822 return (EINVAL); 823 824 /* Signal value requires RFTSIGZMB. */ 825 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 826 return (EINVAL); 827 828 /* Can't copy and clear. */ 829 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 830 return (EINVAL); 831 832 /* Check the validity of the signal number. */ 833 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 834 return (EINVAL); 835 836 if ((flags & RFPROCDESC) != 0) { 837 /* Can't not create a process yet get a process descriptor. */ 838 if ((flags & RFPROC) == 0) 839 return (EINVAL); 840 841 /* Must provide a place to put a procdesc if creating one. */ 842 if (fr->fr_pd_fd == NULL) 843 return (EINVAL); 844 845 /* Check if we are using supported flags. */ 846 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) 847 return (EINVAL); 848 } 849 850 p1 = td->td_proc; 851 852 /* 853 * Here we don't create a new process, but we divorce 854 * certain parts of a process from itself. 855 */ 856 if ((flags & RFPROC) == 0) { 857 if (fr->fr_procp != NULL) 858 *fr->fr_procp = NULL; 859 else if (fr->fr_pidp != NULL) 860 *fr->fr_pidp = 0; 861 return (fork_norfproc(td, flags)); 862 } 863 864 fp_procdesc = NULL; 865 newproc = NULL; 866 vm2 = NULL; 867 868 /* 869 * Increment the nprocs resource before allocations occur. 870 * Although process entries are dynamically created, we still 871 * keep a global limit on the maximum number we will 872 * create. There are hard-limits as to the number of processes 873 * that can run, established by the KVA and memory usage for 874 * the process data. 875 * 876 * Don't allow a nonprivileged user to use the last ten 877 * processes; don't let root exceed the limit. 878 */ 879 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 880 if (nprocs_new >= maxproc - 10) { 881 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 || 882 nprocs_new >= maxproc) { 883 error = EAGAIN; 884 sx_xlock(&allproc_lock); 885 if (ppsratecheck(&lastfail, &curfail, 1)) { 886 printf("maxproc limit exceeded by uid %u " 887 "(pid %d); see tuning(7) and " 888 "login.conf(5)\n", 889 td->td_ucred->cr_ruid, p1->p_pid); 890 } 891 sx_xunlock(&allproc_lock); 892 goto fail2; 893 } 894 } 895 896 /* 897 * If required, create a process descriptor in the parent first; we 898 * will abandon it if something goes wrong. We don't finit() until 899 * later. 900 */ 901 if (flags & RFPROCDESC) { 902 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, 903 fr->fr_pd_flags, fr->fr_pd_fcaps); 904 if (error != 0) 905 goto fail2; 906 } 907 908 mem_charged = 0; 909 if (pages == 0) 910 pages = kstack_pages; 911 /* Allocate new proc. */ 912 newproc = uma_zalloc(proc_zone, M_WAITOK); 913 td2 = FIRST_THREAD_IN_PROC(newproc); 914 if (td2 == NULL) { 915 td2 = thread_alloc(pages); 916 if (td2 == NULL) { 917 error = ENOMEM; 918 goto fail2; 919 } 920 proc_linkup(newproc, td2); 921 } else { 922 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 923 if (td2->td_kstack != 0) 924 vm_thread_dispose(td2); 925 if (!thread_alloc_stack(td2, pages)) { 926 error = ENOMEM; 927 goto fail2; 928 } 929 } 930 } 931 932 if ((flags & RFMEM) == 0) { 933 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 934 if (vm2 == NULL) { 935 error = ENOMEM; 936 goto fail2; 937 } 938 if (!swap_reserve(mem_charged)) { 939 /* 940 * The swap reservation failed. The accounting 941 * from the entries of the copied vm2 will be 942 * subtracted in vmspace_free(), so force the 943 * reservation there. 944 */ 945 swap_reserve_force(mem_charged); 946 error = ENOMEM; 947 goto fail2; 948 } 949 } else 950 vm2 = NULL; 951 952 /* 953 * XXX: This is ugly; when we copy resource usage, we need to bump 954 * per-cred resource counters. 955 */ 956 proc_set_cred_init(newproc, crhold(td->td_ucred)); 957 958 /* 959 * Initialize resource accounting for the child process. 960 */ 961 error = racct_proc_fork(p1, newproc); 962 if (error != 0) { 963 error = EAGAIN; 964 goto fail1; 965 } 966 967 #ifdef MAC 968 mac_proc_init(newproc); 969 #endif 970 newproc->p_klist = knlist_alloc(&newproc->p_mtx); 971 STAILQ_INIT(&newproc->p_ktr); 972 973 /* 974 * Increment the count of procs running with this uid. Don't allow 975 * a nonprivileged user to exceed their current limit. 976 */ 977 cred = td->td_ucred; 978 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) { 979 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0) 980 goto fail0; 981 chgproccnt(cred->cr_ruidinfo, 1, 0); 982 } 983 984 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 985 return (0); 986 fail0: 987 error = EAGAIN; 988 #ifdef MAC 989 mac_proc_destroy(newproc); 990 #endif 991 racct_proc_exit(newproc); 992 fail1: 993 crfree(newproc->p_ucred); 994 newproc->p_ucred = NULL; 995 fail2: 996 if (vm2 != NULL) 997 vmspace_free(vm2); 998 uma_zfree(proc_zone, newproc); 999 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 1000 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 1001 fdrop(fp_procdesc, td); 1002 } 1003 atomic_add_int(&nprocs, -1); 1004 pause("fork", hz / 2); 1005 return (error); 1006 } 1007 1008 /* 1009 * Handle the return of a child process from fork1(). This function 1010 * is called from the MD fork_trampoline() entry point. 1011 */ 1012 void 1013 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1014 struct trapframe *frame) 1015 { 1016 struct proc *p; 1017 struct thread *td; 1018 struct thread *dtd; 1019 1020 td = curthread; 1021 p = td->td_proc; 1022 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1023 1024 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1025 td, td_get_sched(td), p->p_pid, td->td_name); 1026 1027 sched_fork_exit(td); 1028 /* 1029 * Processes normally resume in mi_switch() after being 1030 * cpu_switch()'ed to, but when children start up they arrive here 1031 * instead, so we must do much the same things as mi_switch() would. 1032 */ 1033 if ((dtd = PCPU_GET(deadthread))) { 1034 PCPU_SET(deadthread, NULL); 1035 thread_stash(dtd); 1036 } 1037 thread_unlock(td); 1038 1039 /* 1040 * cpu_fork_kthread_handler intercepts this function call to 1041 * have this call a non-return function to stay in kernel mode. 1042 * initproc has its own fork handler, but it does return. 1043 */ 1044 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1045 callout(arg, frame); 1046 1047 /* 1048 * Check if a kernel thread misbehaved and returned from its main 1049 * function. 1050 */ 1051 if (p->p_flag & P_KPROC) { 1052 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1053 td->td_name, p->p_pid); 1054 kthread_exit(); 1055 } 1056 mtx_assert(&Giant, MA_NOTOWNED); 1057 1058 if (p->p_sysent->sv_schedtail != NULL) 1059 (p->p_sysent->sv_schedtail)(td); 1060 td->td_pflags &= ~TDP_FORKING; 1061 } 1062 1063 /* 1064 * Simplified back end of syscall(), used when returning from fork() 1065 * directly into user mode. This function is passed in to fork_exit() 1066 * as the first parameter and is called when returning to a new 1067 * userland process. 1068 */ 1069 void 1070 fork_return(struct thread *td, struct trapframe *frame) 1071 { 1072 struct proc *p; 1073 1074 p = td->td_proc; 1075 if (td->td_dbgflags & TDB_STOPATFORK) { 1076 PROC_LOCK(p); 1077 if ((p->p_flag & P_TRACED) != 0) { 1078 /* 1079 * Inform the debugger if one is still present. 1080 */ 1081 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; 1082 ptracestop(td, SIGSTOP, NULL); 1083 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1084 } else { 1085 /* 1086 * ... otherwise clear the request. 1087 */ 1088 td->td_dbgflags &= ~TDB_STOPATFORK; 1089 } 1090 PROC_UNLOCK(p); 1091 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1092 /* 1093 * This is the start of a new thread in a traced 1094 * process. Report a system call exit event. 1095 */ 1096 PROC_LOCK(p); 1097 td->td_dbgflags |= TDB_SCX; 1098 _STOPEVENT(p, S_SCX, td->td_sa.code); 1099 if ((p->p_ptevents & PTRACE_SCX) != 0 || 1100 (td->td_dbgflags & TDB_BORN) != 0) 1101 ptracestop(td, SIGTRAP, NULL); 1102 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1103 PROC_UNLOCK(p); 1104 } 1105 1106 userret(td, frame); 1107 1108 #ifdef KTRACE 1109 if (KTRPOINT(td, KTR_SYSRET)) 1110 ktrsysret(SYS_fork, 0, 0); 1111 #endif 1112 } 1113