1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_ktrace.h" 41 #include "opt_kstack_pages.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/eventhandler.h> 47 #include <sys/fcntl.h> 48 #include <sys/filedesc.h> 49 #include <sys/jail.h> 50 #include <sys/kernel.h> 51 #include <sys/kthread.h> 52 #include <sys/sysctl.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mutex.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/procdesc.h> 59 #include <sys/pioctl.h> 60 #include <sys/ptrace.h> 61 #include <sys/racct.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sched.h> 64 #include <sys/syscall.h> 65 #include <sys/vmmeter.h> 66 #include <sys/vnode.h> 67 #include <sys/acct.h> 68 #include <sys/ktr.h> 69 #include <sys/ktrace.h> 70 #include <sys/unistd.h> 71 #include <sys/sdt.h> 72 #include <sys/sx.h> 73 #include <sys/sysent.h> 74 #include <sys/signalvar.h> 75 76 #include <security/audit/audit.h> 77 #include <security/mac/mac_framework.h> 78 79 #include <vm/vm.h> 80 #include <vm/pmap.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_extern.h> 83 #include <vm/uma.h> 84 #include <vm/vm_domain.h> 85 86 #ifdef KDTRACE_HOOKS 87 #include <sys/dtrace_bsd.h> 88 dtrace_fork_func_t dtrace_fasttrap_fork; 89 #endif 90 91 SDT_PROVIDER_DECLARE(proc); 92 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); 93 94 #ifndef _SYS_SYSPROTO_H_ 95 struct fork_args { 96 int dummy; 97 }; 98 #endif 99 100 /* ARGSUSED */ 101 int 102 sys_fork(struct thread *td, struct fork_args *uap) 103 { 104 struct fork_req fr; 105 int error, pid; 106 107 bzero(&fr, sizeof(fr)); 108 fr.fr_flags = RFFDG | RFPROC; 109 fr.fr_pidp = &pid; 110 error = fork1(td, &fr); 111 if (error == 0) { 112 td->td_retval[0] = pid; 113 td->td_retval[1] = 0; 114 } 115 return (error); 116 } 117 118 /* ARGUSED */ 119 int 120 sys_pdfork(struct thread *td, struct pdfork_args *uap) 121 { 122 struct fork_req fr; 123 int error, fd, pid; 124 125 bzero(&fr, sizeof(fr)); 126 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC; 127 fr.fr_pidp = &pid; 128 fr.fr_pd_fd = &fd; 129 fr.fr_pd_flags = uap->flags; 130 /* 131 * It is necessary to return fd by reference because 0 is a valid file 132 * descriptor number, and the child needs to be able to distinguish 133 * itself from the parent using the return value. 134 */ 135 error = fork1(td, &fr); 136 if (error == 0) { 137 td->td_retval[0] = pid; 138 td->td_retval[1] = 0; 139 error = copyout(&fd, uap->fdp, sizeof(fd)); 140 } 141 return (error); 142 } 143 144 /* ARGSUSED */ 145 int 146 sys_vfork(struct thread *td, struct vfork_args *uap) 147 { 148 struct fork_req fr; 149 int error, pid; 150 151 bzero(&fr, sizeof(fr)); 152 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 153 fr.fr_pidp = &pid; 154 error = fork1(td, &fr); 155 if (error == 0) { 156 td->td_retval[0] = pid; 157 td->td_retval[1] = 0; 158 } 159 return (error); 160 } 161 162 int 163 sys_rfork(struct thread *td, struct rfork_args *uap) 164 { 165 struct fork_req fr; 166 int error, pid; 167 168 /* Don't allow kernel-only flags. */ 169 if ((uap->flags & RFKERNELONLY) != 0) 170 return (EINVAL); 171 172 AUDIT_ARG_FFLAGS(uap->flags); 173 bzero(&fr, sizeof(fr)); 174 fr.fr_flags = uap->flags; 175 fr.fr_pidp = &pid; 176 error = fork1(td, &fr); 177 if (error == 0) { 178 td->td_retval[0] = pid; 179 td->td_retval[1] = 0; 180 } 181 return (error); 182 } 183 184 int nprocs = 1; /* process 0 */ 185 int lastpid = 0; 186 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 187 "Last used PID"); 188 189 /* 190 * Random component to lastpid generation. We mix in a random factor to make 191 * it a little harder to predict. We sanity check the modulus value to avoid 192 * doing it in critical paths. Don't let it be too small or we pointlessly 193 * waste randomness entropy, and don't let it be impossibly large. Using a 194 * modulus that is too big causes a LOT more process table scans and slows 195 * down fork processing as the pidchecked caching is defeated. 196 */ 197 static int randompid = 0; 198 199 static int 200 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 201 { 202 int error, pid; 203 204 error = sysctl_wire_old_buffer(req, sizeof(int)); 205 if (error != 0) 206 return(error); 207 sx_xlock(&allproc_lock); 208 pid = randompid; 209 error = sysctl_handle_int(oidp, &pid, 0, req); 210 if (error == 0 && req->newptr != NULL) { 211 if (pid < 0 || pid > pid_max - 100) /* out of range */ 212 pid = pid_max - 100; 213 else if (pid < 2) /* NOP */ 214 pid = 0; 215 else if (pid < 100) /* Make it reasonable */ 216 pid = 100; 217 randompid = pid; 218 } 219 sx_xunlock(&allproc_lock); 220 return (error); 221 } 222 223 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 224 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 225 226 static int 227 fork_findpid(int flags) 228 { 229 struct proc *p; 230 int trypid; 231 static int pidchecked = 0; 232 233 /* 234 * Requires allproc_lock in order to iterate over the list 235 * of processes, and proctree_lock to access p_pgrp. 236 */ 237 sx_assert(&allproc_lock, SX_LOCKED); 238 sx_assert(&proctree_lock, SX_LOCKED); 239 240 /* 241 * Find an unused process ID. We remember a range of unused IDs 242 * ready to use (from lastpid+1 through pidchecked-1). 243 * 244 * If RFHIGHPID is set (used during system boot), do not allocate 245 * low-numbered pids. 246 */ 247 trypid = lastpid + 1; 248 if (flags & RFHIGHPID) { 249 if (trypid < 10) 250 trypid = 10; 251 } else { 252 if (randompid) 253 trypid += arc4random() % randompid; 254 } 255 retry: 256 /* 257 * If the process ID prototype has wrapped around, 258 * restart somewhat above 0, as the low-numbered procs 259 * tend to include daemons that don't exit. 260 */ 261 if (trypid >= pid_max) { 262 trypid = trypid % pid_max; 263 if (trypid < 100) 264 trypid += 100; 265 pidchecked = 0; 266 } 267 if (trypid >= pidchecked) { 268 int doingzomb = 0; 269 270 pidchecked = PID_MAX; 271 /* 272 * Scan the active and zombie procs to check whether this pid 273 * is in use. Remember the lowest pid that's greater 274 * than trypid, so we can avoid checking for a while. 275 * 276 * Avoid reuse of the process group id, session id or 277 * the reaper subtree id. Note that for process group 278 * and sessions, the amount of reserved pids is 279 * limited by process limit. For the subtree ids, the 280 * id is kept reserved only while there is a 281 * non-reaped process in the subtree, so amount of 282 * reserved pids is limited by process limit times 283 * two. 284 */ 285 p = LIST_FIRST(&allproc); 286 again: 287 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 288 while (p->p_pid == trypid || 289 p->p_reapsubtree == trypid || 290 (p->p_pgrp != NULL && 291 (p->p_pgrp->pg_id == trypid || 292 (p->p_session != NULL && 293 p->p_session->s_sid == trypid)))) { 294 trypid++; 295 if (trypid >= pidchecked) 296 goto retry; 297 } 298 if (p->p_pid > trypid && pidchecked > p->p_pid) 299 pidchecked = p->p_pid; 300 if (p->p_pgrp != NULL) { 301 if (p->p_pgrp->pg_id > trypid && 302 pidchecked > p->p_pgrp->pg_id) 303 pidchecked = p->p_pgrp->pg_id; 304 if (p->p_session != NULL && 305 p->p_session->s_sid > trypid && 306 pidchecked > p->p_session->s_sid) 307 pidchecked = p->p_session->s_sid; 308 } 309 } 310 if (!doingzomb) { 311 doingzomb = 1; 312 p = LIST_FIRST(&zombproc); 313 goto again; 314 } 315 } 316 317 /* 318 * RFHIGHPID does not mess with the lastpid counter during boot. 319 */ 320 if (flags & RFHIGHPID) 321 pidchecked = 0; 322 else 323 lastpid = trypid; 324 325 return (trypid); 326 } 327 328 static int 329 fork_norfproc(struct thread *td, int flags) 330 { 331 int error; 332 struct proc *p1; 333 334 KASSERT((flags & RFPROC) == 0, 335 ("fork_norfproc called with RFPROC set")); 336 p1 = td->td_proc; 337 338 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 339 (flags & (RFCFDG | RFFDG))) { 340 PROC_LOCK(p1); 341 if (thread_single(p1, SINGLE_BOUNDARY)) { 342 PROC_UNLOCK(p1); 343 return (ERESTART); 344 } 345 PROC_UNLOCK(p1); 346 } 347 348 error = vm_forkproc(td, NULL, NULL, NULL, flags); 349 if (error) 350 goto fail; 351 352 /* 353 * Close all file descriptors. 354 */ 355 if (flags & RFCFDG) { 356 struct filedesc *fdtmp; 357 fdtmp = fdinit(td->td_proc->p_fd, false); 358 fdescfree(td); 359 p1->p_fd = fdtmp; 360 } 361 362 /* 363 * Unshare file descriptors (from parent). 364 */ 365 if (flags & RFFDG) 366 fdunshare(td); 367 368 fail: 369 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 370 (flags & (RFCFDG | RFFDG))) { 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 int trypid; 384 struct filedesc *fd; 385 struct filedesc_to_leader *fdtol; 386 struct sigacts *newsigacts; 387 388 sx_assert(&proctree_lock, SX_SLOCKED); 389 sx_assert(&allproc_lock, SX_XLOCKED); 390 391 p1 = td->td_proc; 392 393 trypid = fork_findpid(fr->fr_flags); 394 395 sx_sunlock(&proctree_lock); 396 397 p2->p_state = PRS_NEW; /* protect against others */ 398 p2->p_pid = trypid; 399 AUDIT_ARG_PID(p2->p_pid); 400 LIST_INSERT_HEAD(&allproc, p2, p_list); 401 allproc_gen++; 402 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 403 tidhash_add(td2); 404 PROC_LOCK(p2); 405 PROC_LOCK(p1); 406 407 sx_xunlock(&allproc_lock); 408 409 bcopy(&p1->p_startcopy, &p2->p_startcopy, 410 __rangeof(struct proc, p_startcopy, p_endcopy)); 411 pargs_hold(p2->p_args); 412 413 PROC_UNLOCK(p1); 414 415 bzero(&p2->p_startzero, 416 __rangeof(struct proc, p_startzero, p_endzero)); 417 418 /* Tell the prison that we exist. */ 419 prison_proc_hold(p2->p_ucred->cr_prison); 420 421 PROC_UNLOCK(p2); 422 423 /* 424 * Malloc things while we don't hold any locks. 425 */ 426 if (fr->fr_flags & RFSIGSHARE) 427 newsigacts = NULL; 428 else 429 newsigacts = sigacts_alloc(); 430 431 /* 432 * Copy filedesc. 433 */ 434 if (fr->fr_flags & RFCFDG) { 435 fd = fdinit(p1->p_fd, false); 436 fdtol = NULL; 437 } else if (fr->fr_flags & RFFDG) { 438 fd = fdcopy(p1->p_fd); 439 fdtol = NULL; 440 } else { 441 fd = fdshare(p1->p_fd); 442 if (p1->p_fdtol == NULL) 443 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 444 p1->p_leader); 445 if ((fr->fr_flags & RFTHREAD) != 0) { 446 /* 447 * Shared file descriptor table, and shared 448 * process leaders. 449 */ 450 fdtol = p1->p_fdtol; 451 FILEDESC_XLOCK(p1->p_fd); 452 fdtol->fdl_refcount++; 453 FILEDESC_XUNLOCK(p1->p_fd); 454 } else { 455 /* 456 * Shared file descriptor table, and different 457 * process leaders. 458 */ 459 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 460 p1->p_fd, p2); 461 } 462 } 463 /* 464 * Make a proc table entry for the new process. 465 * Start by zeroing the section of proc that is zero-initialized, 466 * then copy the section that is copied directly from the parent. 467 */ 468 469 PROC_LOCK(p2); 470 PROC_LOCK(p1); 471 472 bzero(&td2->td_startzero, 473 __rangeof(struct thread, td_startzero, td_endzero)); 474 475 bcopy(&td->td_startcopy, &td2->td_startcopy, 476 __rangeof(struct thread, td_startcopy, td_endcopy)); 477 478 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 479 td2->td_sigstk = td->td_sigstk; 480 td2->td_flags = TDF_INMEM; 481 td2->td_lend_user_pri = PRI_MAX; 482 483 #ifdef VIMAGE 484 td2->td_vnet = NULL; 485 td2->td_vnet_lpush = NULL; 486 #endif 487 488 /* 489 * Allow the scheduler to initialize the child. 490 */ 491 thread_lock(td); 492 sched_fork(td, td2); 493 thread_unlock(td); 494 495 /* 496 * Duplicate sub-structures as needed. 497 * Increase reference counts on shared objects. 498 */ 499 p2->p_flag = P_INMEM; 500 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC); 501 p2->p_swtick = ticks; 502 if (p1->p_flag & P_PROFIL) 503 startprofclock(p2); 504 505 /* 506 * Whilst the proc lock is held, copy the VM domain data out 507 * using the VM domain method. 508 */ 509 vm_domain_policy_init(&p2->p_vm_dom_policy); 510 vm_domain_policy_localcopy(&p2->p_vm_dom_policy, 511 &p1->p_vm_dom_policy); 512 513 if (fr->fr_flags & RFSIGSHARE) { 514 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 515 } else { 516 sigacts_copy(newsigacts, p1->p_sigacts); 517 p2->p_sigacts = newsigacts; 518 } 519 520 if (fr->fr_flags & RFTSIGZMB) 521 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 522 else if (fr->fr_flags & RFLINUXTHPN) 523 p2->p_sigparent = SIGUSR1; 524 else 525 p2->p_sigparent = SIGCHLD; 526 527 p2->p_textvp = p1->p_textvp; 528 p2->p_fd = fd; 529 p2->p_fdtol = fdtol; 530 531 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 532 p2->p_flag |= P_PROTECTED; 533 p2->p_flag2 |= P2_INHERIT_PROTECTED; 534 } 535 536 /* 537 * p_limit is copy-on-write. Bump its refcount. 538 */ 539 lim_fork(p1, p2); 540 541 thread_cow_get_proc(td2, p2); 542 543 pstats_fork(p1->p_stats, p2->p_stats); 544 545 PROC_UNLOCK(p1); 546 PROC_UNLOCK(p2); 547 548 /* Bump references to the text vnode (for procfs). */ 549 if (p2->p_textvp) 550 vref(p2->p_textvp); 551 552 /* 553 * Set up linkage for kernel based threading. 554 */ 555 if ((fr->fr_flags & RFTHREAD) != 0) { 556 mtx_lock(&ppeers_lock); 557 p2->p_peers = p1->p_peers; 558 p1->p_peers = p2; 559 p2->p_leader = p1->p_leader; 560 mtx_unlock(&ppeers_lock); 561 PROC_LOCK(p1->p_leader); 562 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 563 PROC_UNLOCK(p1->p_leader); 564 /* 565 * The task leader is exiting, so process p1 is 566 * going to be killed shortly. Since p1 obviously 567 * isn't dead yet, we know that the leader is either 568 * sending SIGKILL's to all the processes in this 569 * task or is sleeping waiting for all the peers to 570 * exit. We let p1 complete the fork, but we need 571 * to go ahead and kill the new process p2 since 572 * the task leader may not get a chance to send 573 * SIGKILL to it. We leave it on the list so that 574 * the task leader will wait for this new process 575 * to commit suicide. 576 */ 577 PROC_LOCK(p2); 578 kern_psignal(p2, SIGKILL); 579 PROC_UNLOCK(p2); 580 } else 581 PROC_UNLOCK(p1->p_leader); 582 } else { 583 p2->p_peers = NULL; 584 p2->p_leader = p2; 585 } 586 587 sx_xlock(&proctree_lock); 588 PGRP_LOCK(p1->p_pgrp); 589 PROC_LOCK(p2); 590 PROC_LOCK(p1); 591 592 /* 593 * Preserve some more flags in subprocess. P_PROFIL has already 594 * been preserved. 595 */ 596 p2->p_flag |= p1->p_flag & P_SUGID; 597 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 598 SESS_LOCK(p1->p_session); 599 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 600 p2->p_flag |= P_CONTROLT; 601 SESS_UNLOCK(p1->p_session); 602 if (fr->fr_flags & RFPPWAIT) 603 p2->p_flag |= P_PPWAIT; 604 605 p2->p_pgrp = p1->p_pgrp; 606 LIST_INSERT_AFTER(p1, p2, p_pglist); 607 PGRP_UNLOCK(p1->p_pgrp); 608 LIST_INIT(&p2->p_children); 609 LIST_INIT(&p2->p_orphans); 610 611 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 612 613 /* 614 * If PF_FORK is set, the child process inherits the 615 * procfs ioctl flags from its parent. 616 */ 617 if (p1->p_pfsflags & PF_FORK) { 618 p2->p_stops = p1->p_stops; 619 p2->p_pfsflags = p1->p_pfsflags; 620 } 621 622 /* 623 * This begins the section where we must prevent the parent 624 * from being swapped. 625 */ 626 _PHOLD(p1); 627 PROC_UNLOCK(p1); 628 629 /* 630 * Attach the new process to its parent. 631 * 632 * If RFNOWAIT is set, the newly created process becomes a child 633 * of init. This effectively disassociates the child from the 634 * parent. 635 */ 636 if ((fr->fr_flags & RFNOWAIT) != 0) { 637 pptr = p1->p_reaper; 638 p2->p_reaper = pptr; 639 } else { 640 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 641 p1 : p1->p_reaper; 642 pptr = p1; 643 } 644 p2->p_pptr = pptr; 645 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 646 LIST_INIT(&p2->p_reaplist); 647 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 648 if (p2->p_reaper == p1) 649 p2->p_reapsubtree = p2->p_pid; 650 sx_xunlock(&proctree_lock); 651 652 /* Inform accounting that we have forked. */ 653 p2->p_acflag = AFORK; 654 PROC_UNLOCK(p2); 655 656 #ifdef KTRACE 657 ktrprocfork(p1, p2); 658 #endif 659 660 /* 661 * Finish creating the child process. It will return via a different 662 * execution path later. (ie: directly into user mode) 663 */ 664 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 665 666 if (fr->fr_flags == (RFFDG | RFPROC)) { 667 PCPU_INC(cnt.v_forks); 668 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 669 p2->p_vmspace->vm_ssize); 670 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 671 PCPU_INC(cnt.v_vforks); 672 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 673 p2->p_vmspace->vm_ssize); 674 } else if (p1 == &proc0) { 675 PCPU_INC(cnt.v_kthreads); 676 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 677 p2->p_vmspace->vm_ssize); 678 } else { 679 PCPU_INC(cnt.v_rforks); 680 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 681 p2->p_vmspace->vm_ssize); 682 } 683 684 /* 685 * Associate the process descriptor with the process before anything 686 * can happen that might cause that process to need the descriptor. 687 * However, don't do this until after fork(2) can no longer fail. 688 */ 689 if (fr->fr_flags & RFPROCDESC) 690 procdesc_new(p2, fr->fr_pd_flags); 691 692 /* 693 * Both processes are set up, now check if any loadable modules want 694 * to adjust anything. 695 */ 696 EVENTHANDLER_INVOKE(process_fork, p1, p2, fr->fr_flags); 697 698 /* 699 * Set the child start time and mark the process as being complete. 700 */ 701 PROC_LOCK(p2); 702 PROC_LOCK(p1); 703 microuptime(&p2->p_stats->p_start); 704 PROC_SLOCK(p2); 705 p2->p_state = PRS_NORMAL; 706 PROC_SUNLOCK(p2); 707 708 #ifdef KDTRACE_HOOKS 709 /* 710 * Tell the DTrace fasttrap provider about the new process so that any 711 * tracepoints inherited from the parent can be removed. We have to do 712 * this only after p_state is PRS_NORMAL since the fasttrap module will 713 * use pfind() later on. 714 */ 715 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 716 dtrace_fasttrap_fork(p1, p2); 717 #endif 718 /* 719 * Hold the process so that it cannot exit after we make it runnable, 720 * but before we wait for the debugger. 721 */ 722 _PHOLD(p2); 723 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED | 724 P_FOLLOWFORK)) { 725 /* 726 * Arrange for debugger to receive the fork event. 727 * 728 * We can report PL_FLAG_FORKED regardless of 729 * P_FOLLOWFORK settings, but it does not make a sense 730 * for runaway child. 731 */ 732 td->td_dbgflags |= TDB_FORK; 733 td->td_dbg_forked = p2->p_pid; 734 td2->td_dbgflags |= TDB_STOPATFORK; 735 } 736 if (fr->fr_flags & RFPPWAIT) { 737 td->td_pflags |= TDP_RFPPWAIT; 738 td->td_rfppwait_p = p2; 739 } 740 PROC_UNLOCK(p2); 741 742 /* 743 * Now can be swapped. 744 */ 745 _PRELE(p1); 746 PROC_UNLOCK(p1); 747 748 /* 749 * Tell any interested parties about the new process. 750 */ 751 knote_fork(&p1->p_klist, p2->p_pid); 752 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 753 754 if (fr->fr_flags & RFPROCDESC) { 755 procdesc_finit(p2->p_procdesc, fp_procdesc); 756 fdrop(fp_procdesc, td); 757 } 758 759 if ((fr->fr_flags & RFSTOPPED) == 0) { 760 /* 761 * If RFSTOPPED not requested, make child runnable and 762 * add to run queue. 763 */ 764 thread_lock(td2); 765 TD_SET_CAN_RUN(td2); 766 sched_add(td2, SRQ_BORING); 767 thread_unlock(td2); 768 if (fr->fr_pidp != NULL) 769 *fr->fr_pidp = p2->p_pid; 770 } else { 771 *fr->fr_procp = p2; 772 } 773 774 PROC_LOCK(p2); 775 /* 776 * Wait until debugger is attached to child. 777 */ 778 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0) 779 cv_wait(&p2->p_dbgwait, &p2->p_mtx); 780 _PRELE(p2); 781 racct_proc_fork_done(p2); 782 PROC_UNLOCK(p2); 783 } 784 785 int 786 fork1(struct thread *td, struct fork_req *fr) 787 { 788 struct proc *p1, *newproc; 789 struct thread *td2; 790 struct vmspace *vm2; 791 struct file *fp_procdesc; 792 vm_ooffset_t mem_charged; 793 int error, nprocs_new, ok; 794 static int curfail; 795 static struct timeval lastfail; 796 int flags, pages; 797 798 flags = fr->fr_flags; 799 pages = fr->fr_pages; 800 801 if ((flags & RFSTOPPED) != 0) 802 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 803 else 804 MPASS(fr->fr_procp == NULL); 805 806 /* Check for the undefined or unimplemented flags. */ 807 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 808 return (EINVAL); 809 810 /* Signal value requires RFTSIGZMB. */ 811 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 812 return (EINVAL); 813 814 /* Can't copy and clear. */ 815 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 816 return (EINVAL); 817 818 /* Check the validity of the signal number. */ 819 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 820 return (EINVAL); 821 822 if ((flags & RFPROCDESC) != 0) { 823 /* Can't not create a process yet get a process descriptor. */ 824 if ((flags & RFPROC) == 0) 825 return (EINVAL); 826 827 /* Must provide a place to put a procdesc if creating one. */ 828 if (fr->fr_pd_fd == NULL) 829 return (EINVAL); 830 } 831 832 p1 = td->td_proc; 833 834 /* 835 * Here we don't create a new process, but we divorce 836 * certain parts of a process from itself. 837 */ 838 if ((flags & RFPROC) == 0) { 839 if (fr->fr_procp != NULL) 840 *fr->fr_procp = NULL; 841 else if (fr->fr_pidp != NULL) 842 *fr->fr_pidp = 0; 843 return (fork_norfproc(td, flags)); 844 } 845 846 fp_procdesc = NULL; 847 newproc = NULL; 848 vm2 = NULL; 849 850 /* 851 * Increment the nprocs resource before allocations occur. 852 * Although process entries are dynamically created, we still 853 * keep a global limit on the maximum number we will 854 * create. There are hard-limits as to the number of processes 855 * that can run, established by the KVA and memory usage for 856 * the process data. 857 * 858 * Don't allow a nonprivileged user to use the last ten 859 * processes; don't let root exceed the limit. 860 */ 861 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 862 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, 863 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { 864 error = EAGAIN; 865 sx_xlock(&allproc_lock); 866 if (ppsratecheck(&lastfail, &curfail, 1)) { 867 printf("maxproc limit exceeded by uid %u (pid %d); " 868 "see tuning(7) and login.conf(5)\n", 869 td->td_ucred->cr_ruid, p1->p_pid); 870 } 871 sx_xunlock(&allproc_lock); 872 goto fail2; 873 } 874 875 /* 876 * If required, create a process descriptor in the parent first; we 877 * will abandon it if something goes wrong. We don't finit() until 878 * later. 879 */ 880 if (flags & RFPROCDESC) { 881 error = falloc_caps(td, &fp_procdesc, fr->fr_pd_fd, 0, 882 fr->fr_pd_fcaps); 883 if (error != 0) 884 goto fail2; 885 } 886 887 mem_charged = 0; 888 if (pages == 0) 889 pages = kstack_pages; 890 /* Allocate new proc. */ 891 newproc = uma_zalloc(proc_zone, M_WAITOK); 892 td2 = FIRST_THREAD_IN_PROC(newproc); 893 if (td2 == NULL) { 894 td2 = thread_alloc(pages); 895 if (td2 == NULL) { 896 error = ENOMEM; 897 goto fail2; 898 } 899 proc_linkup(newproc, td2); 900 } else { 901 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 902 if (td2->td_kstack != 0) 903 vm_thread_dispose(td2); 904 if (!thread_alloc_stack(td2, pages)) { 905 error = ENOMEM; 906 goto fail2; 907 } 908 } 909 } 910 911 if ((flags & RFMEM) == 0) { 912 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 913 if (vm2 == NULL) { 914 error = ENOMEM; 915 goto fail2; 916 } 917 if (!swap_reserve(mem_charged)) { 918 /* 919 * The swap reservation failed. The accounting 920 * from the entries of the copied vm2 will be 921 * substracted in vmspace_free(), so force the 922 * reservation there. 923 */ 924 swap_reserve_force(mem_charged); 925 error = ENOMEM; 926 goto fail2; 927 } 928 } else 929 vm2 = NULL; 930 931 /* 932 * XXX: This is ugly; when we copy resource usage, we need to bump 933 * per-cred resource counters. 934 */ 935 proc_set_cred_init(newproc, crhold(td->td_ucred)); 936 937 /* 938 * Initialize resource accounting for the child process. 939 */ 940 error = racct_proc_fork(p1, newproc); 941 if (error != 0) { 942 error = EAGAIN; 943 goto fail1; 944 } 945 946 #ifdef MAC 947 mac_proc_init(newproc); 948 #endif 949 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); 950 STAILQ_INIT(&newproc->p_ktr); 951 952 /* We have to lock the process tree while we look for a pid. */ 953 sx_slock(&proctree_lock); 954 sx_xlock(&allproc_lock); 955 956 /* 957 * Increment the count of procs running with this uid. Don't allow 958 * a nonprivileged user to exceed their current limit. 959 * 960 * XXXRW: Can we avoid privilege here if it's not needed? 961 */ 962 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 963 if (error == 0) 964 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 965 else { 966 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 967 lim_cur(td, RLIMIT_NPROC)); 968 } 969 if (ok) { 970 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 971 return (0); 972 } 973 974 error = EAGAIN; 975 sx_sunlock(&proctree_lock); 976 sx_xunlock(&allproc_lock); 977 #ifdef MAC 978 mac_proc_destroy(newproc); 979 #endif 980 racct_proc_exit(newproc); 981 fail1: 982 crfree(newproc->p_ucred); 983 newproc->p_ucred = NULL; 984 fail2: 985 if (vm2 != NULL) 986 vmspace_free(vm2); 987 uma_zfree(proc_zone, newproc); 988 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 989 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 990 fdrop(fp_procdesc, td); 991 } 992 atomic_add_int(&nprocs, -1); 993 pause("fork", hz / 2); 994 return (error); 995 } 996 997 /* 998 * Handle the return of a child process from fork1(). This function 999 * is called from the MD fork_trampoline() entry point. 1000 */ 1001 void 1002 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1003 struct trapframe *frame) 1004 { 1005 struct proc *p; 1006 struct thread *td; 1007 struct thread *dtd; 1008 1009 td = curthread; 1010 p = td->td_proc; 1011 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1012 1013 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1014 td, td->td_sched, p->p_pid, td->td_name); 1015 1016 sched_fork_exit(td); 1017 /* 1018 * Processes normally resume in mi_switch() after being 1019 * cpu_switch()'ed to, but when children start up they arrive here 1020 * instead, so we must do much the same things as mi_switch() would. 1021 */ 1022 if ((dtd = PCPU_GET(deadthread))) { 1023 PCPU_SET(deadthread, NULL); 1024 thread_stash(dtd); 1025 } 1026 thread_unlock(td); 1027 1028 /* 1029 * cpu_set_fork_handler intercepts this function call to 1030 * have this call a non-return function to stay in kernel mode. 1031 * initproc has its own fork handler, but it does return. 1032 */ 1033 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1034 callout(arg, frame); 1035 1036 /* 1037 * Check if a kernel thread misbehaved and returned from its main 1038 * function. 1039 */ 1040 if (p->p_flag & P_KPROC) { 1041 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1042 td->td_name, p->p_pid); 1043 kthread_exit(); 1044 } 1045 mtx_assert(&Giant, MA_NOTOWNED); 1046 1047 if (p->p_sysent->sv_schedtail != NULL) 1048 (p->p_sysent->sv_schedtail)(td); 1049 td->td_pflags &= ~TDP_FORKING; 1050 } 1051 1052 /* 1053 * Simplified back end of syscall(), used when returning from fork() 1054 * directly into user mode. Giant is not held on entry, and must not 1055 * be held on return. This function is passed in to fork_exit() as the 1056 * first parameter and is called when returning to a new userland process. 1057 */ 1058 void 1059 fork_return(struct thread *td, struct trapframe *frame) 1060 { 1061 struct proc *p, *dbg; 1062 1063 p = td->td_proc; 1064 if (td->td_dbgflags & TDB_STOPATFORK) { 1065 sx_xlock(&proctree_lock); 1066 PROC_LOCK(p); 1067 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) == 1068 (P_TRACED | P_FOLLOWFORK)) { 1069 /* 1070 * If debugger still wants auto-attach for the 1071 * parent's children, do it now. 1072 */ 1073 dbg = p->p_pptr->p_pptr; 1074 p->p_flag |= P_TRACED; 1075 p->p_oppid = p->p_pptr->p_pid; 1076 CTR2(KTR_PTRACE, 1077 "fork_return: attaching to new child pid %d: oppid %d", 1078 p->p_pid, p->p_oppid); 1079 proc_reparent(p, dbg); 1080 sx_xunlock(&proctree_lock); 1081 td->td_dbgflags |= TDB_CHILD | TDB_SCX; 1082 ptracestop(td, SIGSTOP); 1083 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1084 } else { 1085 /* 1086 * ... otherwise clear the request. 1087 */ 1088 sx_xunlock(&proctree_lock); 1089 td->td_dbgflags &= ~TDB_STOPATFORK; 1090 cv_broadcast(&p->p_dbgwait); 1091 } 1092 PROC_UNLOCK(p); 1093 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1094 /* 1095 * This is the start of a new thread in a traced 1096 * process. Report a system call exit event. 1097 */ 1098 PROC_LOCK(p); 1099 td->td_dbgflags |= TDB_SCX; 1100 _STOPEVENT(p, S_SCX, td->td_dbg_sc_code); 1101 if ((p->p_stops & S_PT_SCX) != 0 || 1102 (td->td_dbgflags & TDB_BORN) != 0) 1103 ptracestop(td, SIGTRAP); 1104 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1105 PROC_UNLOCK(p); 1106 } 1107 1108 userret(td, frame); 1109 1110 #ifdef KTRACE 1111 if (KTRPOINT(td, KTR_SYSRET)) 1112 ktrsysret(SYS_fork, 0, 0); 1113 #endif 1114 } 1115