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(td, uap) 121 struct thread *td; 122 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 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 || pid > pid_max - 100) /* out of range */ 214 pid = pid_max - 100; 215 else if (pid < 2) /* NOP */ 216 pid = 0; 217 else if (pid < 100) /* Make it reasonable */ 218 pid = 100; 219 randompid = pid; 220 } 221 sx_xunlock(&allproc_lock); 222 return (error); 223 } 224 225 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 226 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 227 228 static int 229 fork_findpid(int flags) 230 { 231 struct proc *p; 232 int trypid; 233 static int pidchecked = 0; 234 235 /* 236 * Requires allproc_lock in order to iterate over the list 237 * of processes, and proctree_lock to access p_pgrp. 238 */ 239 sx_assert(&allproc_lock, SX_LOCKED); 240 sx_assert(&proctree_lock, SX_LOCKED); 241 242 /* 243 * Find an unused process ID. We remember a range of unused IDs 244 * ready to use (from lastpid+1 through pidchecked-1). 245 * 246 * If RFHIGHPID is set (used during system boot), do not allocate 247 * low-numbered pids. 248 */ 249 trypid = lastpid + 1; 250 if (flags & RFHIGHPID) { 251 if (trypid < 10) 252 trypid = 10; 253 } else { 254 if (randompid) 255 trypid += arc4random() % randompid; 256 } 257 retry: 258 /* 259 * If the process ID prototype has wrapped around, 260 * restart somewhat above 0, as the low-numbered procs 261 * tend to include daemons that don't exit. 262 */ 263 if (trypid >= pid_max) { 264 trypid = trypid % pid_max; 265 if (trypid < 100) 266 trypid += 100; 267 pidchecked = 0; 268 } 269 if (trypid >= pidchecked) { 270 int doingzomb = 0; 271 272 pidchecked = PID_MAX; 273 /* 274 * Scan the active and zombie procs to check whether this pid 275 * is in use. Remember the lowest pid that's greater 276 * than trypid, so we can avoid checking for a while. 277 * 278 * Avoid reuse of the process group id, session id or 279 * the reaper subtree id. Note that for process group 280 * and sessions, the amount of reserved pids is 281 * limited by process limit. For the subtree ids, the 282 * id is kept reserved only while there is a 283 * non-reaped process in the subtree, so amount of 284 * reserved pids is limited by process limit times 285 * two. 286 */ 287 p = LIST_FIRST(&allproc); 288 again: 289 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 290 while (p->p_pid == trypid || 291 p->p_reapsubtree == trypid || 292 (p->p_pgrp != NULL && 293 (p->p_pgrp->pg_id == trypid || 294 (p->p_session != NULL && 295 p->p_session->s_sid == trypid)))) { 296 trypid++; 297 if (trypid >= pidchecked) 298 goto retry; 299 } 300 if (p->p_pid > trypid && pidchecked > p->p_pid) 301 pidchecked = p->p_pid; 302 if (p->p_pgrp != NULL) { 303 if (p->p_pgrp->pg_id > trypid && 304 pidchecked > p->p_pgrp->pg_id) 305 pidchecked = p->p_pgrp->pg_id; 306 if (p->p_session != NULL && 307 p->p_session->s_sid > trypid && 308 pidchecked > p->p_session->s_sid) 309 pidchecked = p->p_session->s_sid; 310 } 311 } 312 if (!doingzomb) { 313 doingzomb = 1; 314 p = LIST_FIRST(&zombproc); 315 goto again; 316 } 317 } 318 319 /* 320 * RFHIGHPID does not mess with the lastpid counter during boot. 321 */ 322 if (flags & RFHIGHPID) 323 pidchecked = 0; 324 else 325 lastpid = trypid; 326 327 return (trypid); 328 } 329 330 static int 331 fork_norfproc(struct thread *td, int flags) 332 { 333 int error; 334 struct proc *p1; 335 336 KASSERT((flags & RFPROC) == 0, 337 ("fork_norfproc called with RFPROC set")); 338 p1 = td->td_proc; 339 340 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 341 (flags & (RFCFDG | RFFDG))) { 342 PROC_LOCK(p1); 343 if (thread_single(p1, SINGLE_BOUNDARY)) { 344 PROC_UNLOCK(p1); 345 return (ERESTART); 346 } 347 PROC_UNLOCK(p1); 348 } 349 350 error = vm_forkproc(td, NULL, NULL, NULL, flags); 351 if (error) 352 goto fail; 353 354 /* 355 * Close all file descriptors. 356 */ 357 if (flags & RFCFDG) { 358 struct filedesc *fdtmp; 359 fdtmp = fdinit(td->td_proc->p_fd, false); 360 fdescfree(td); 361 p1->p_fd = fdtmp; 362 } 363 364 /* 365 * Unshare file descriptors (from parent). 366 */ 367 if (flags & RFFDG) 368 fdunshare(td); 369 370 fail: 371 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 372 (flags & (RFCFDG | RFFDG))) { 373 PROC_LOCK(p1); 374 thread_single_end(p1, SINGLE_BOUNDARY); 375 PROC_UNLOCK(p1); 376 } 377 return (error); 378 } 379 380 static void 381 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, 382 struct vmspace *vm2, struct file *fp_procdesc) 383 { 384 struct proc *p1, *pptr; 385 int trypid; 386 struct filedesc *fd; 387 struct filedesc_to_leader *fdtol; 388 struct sigacts *newsigacts; 389 390 sx_assert(&proctree_lock, SX_SLOCKED); 391 sx_assert(&allproc_lock, SX_XLOCKED); 392 393 p1 = td->td_proc; 394 395 trypid = fork_findpid(fr->fr_flags); 396 397 sx_sunlock(&proctree_lock); 398 399 p2->p_state = PRS_NEW; /* protect against others */ 400 p2->p_pid = trypid; 401 AUDIT_ARG_PID(p2->p_pid); 402 LIST_INSERT_HEAD(&allproc, p2, p_list); 403 allproc_gen++; 404 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 405 tidhash_add(td2); 406 PROC_LOCK(p2); 407 PROC_LOCK(p1); 408 409 sx_xunlock(&allproc_lock); 410 411 bcopy(&p1->p_startcopy, &p2->p_startcopy, 412 __rangeof(struct proc, p_startcopy, p_endcopy)); 413 pargs_hold(p2->p_args); 414 415 PROC_UNLOCK(p1); 416 417 bzero(&p2->p_startzero, 418 __rangeof(struct proc, p_startzero, p_endzero)); 419 420 /* Tell the prison that we exist. */ 421 prison_proc_hold(p2->p_ucred->cr_prison); 422 423 PROC_UNLOCK(p2); 424 425 /* 426 * Malloc things while we don't hold any locks. 427 */ 428 if (fr->fr_flags & RFSIGSHARE) 429 newsigacts = NULL; 430 else 431 newsigacts = sigacts_alloc(); 432 433 /* 434 * Copy filedesc. 435 */ 436 if (fr->fr_flags & RFCFDG) { 437 fd = fdinit(p1->p_fd, false); 438 fdtol = NULL; 439 } else if (fr->fr_flags & RFFDG) { 440 fd = fdcopy(p1->p_fd); 441 fdtol = NULL; 442 } else { 443 fd = fdshare(p1->p_fd); 444 if (p1->p_fdtol == NULL) 445 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 446 p1->p_leader); 447 if ((fr->fr_flags & RFTHREAD) != 0) { 448 /* 449 * Shared file descriptor table, and shared 450 * process leaders. 451 */ 452 fdtol = p1->p_fdtol; 453 FILEDESC_XLOCK(p1->p_fd); 454 fdtol->fdl_refcount++; 455 FILEDESC_XUNLOCK(p1->p_fd); 456 } else { 457 /* 458 * Shared file descriptor table, and different 459 * process leaders. 460 */ 461 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 462 p1->p_fd, p2); 463 } 464 } 465 /* 466 * Make a proc table entry for the new process. 467 * Start by zeroing the section of proc that is zero-initialized, 468 * then copy the section that is copied directly from the parent. 469 */ 470 471 PROC_LOCK(p2); 472 PROC_LOCK(p1); 473 474 bzero(&td2->td_startzero, 475 __rangeof(struct thread, td_startzero, td_endzero)); 476 477 bcopy(&td->td_startcopy, &td2->td_startcopy, 478 __rangeof(struct thread, td_startcopy, td_endcopy)); 479 480 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 481 td2->td_sigstk = td->td_sigstk; 482 td2->td_flags = TDF_INMEM; 483 td2->td_lend_user_pri = PRI_MAX; 484 485 #ifdef VIMAGE 486 td2->td_vnet = NULL; 487 td2->td_vnet_lpush = NULL; 488 #endif 489 490 /* 491 * Allow the scheduler to initialize the child. 492 */ 493 thread_lock(td); 494 sched_fork(td, td2); 495 thread_unlock(td); 496 497 /* 498 * Duplicate sub-structures as needed. 499 * Increase reference counts on shared objects. 500 */ 501 p2->p_flag = P_INMEM; 502 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC); 503 p2->p_swtick = ticks; 504 if (p1->p_flag & P_PROFIL) 505 startprofclock(p2); 506 507 /* 508 * Whilst the proc lock is held, copy the VM domain data out 509 * using the VM domain method. 510 */ 511 vm_domain_policy_init(&p2->p_vm_dom_policy); 512 vm_domain_policy_localcopy(&p2->p_vm_dom_policy, 513 &p1->p_vm_dom_policy); 514 515 if (fr->fr_flags & RFSIGSHARE) { 516 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 517 } else { 518 sigacts_copy(newsigacts, p1->p_sigacts); 519 p2->p_sigacts = newsigacts; 520 } 521 522 if (fr->fr_flags & RFTSIGZMB) 523 p2->p_sigparent = RFTSIGNUM(fr->fr_flags); 524 else if (fr->fr_flags & RFLINUXTHPN) 525 p2->p_sigparent = SIGUSR1; 526 else 527 p2->p_sigparent = SIGCHLD; 528 529 p2->p_textvp = p1->p_textvp; 530 p2->p_fd = fd; 531 p2->p_fdtol = fdtol; 532 533 if (p1->p_flag2 & P2_INHERIT_PROTECTED) { 534 p2->p_flag |= P_PROTECTED; 535 p2->p_flag2 |= P2_INHERIT_PROTECTED; 536 } 537 538 /* 539 * p_limit is copy-on-write. Bump its refcount. 540 */ 541 lim_fork(p1, p2); 542 543 thread_cow_get_proc(td2, p2); 544 545 pstats_fork(p1->p_stats, p2->p_stats); 546 547 PROC_UNLOCK(p1); 548 PROC_UNLOCK(p2); 549 550 /* Bump references to the text vnode (for procfs). */ 551 if (p2->p_textvp) 552 vref(p2->p_textvp); 553 554 /* 555 * Set up linkage for kernel based threading. 556 */ 557 if ((fr->fr_flags & RFTHREAD) != 0) { 558 mtx_lock(&ppeers_lock); 559 p2->p_peers = p1->p_peers; 560 p1->p_peers = p2; 561 p2->p_leader = p1->p_leader; 562 mtx_unlock(&ppeers_lock); 563 PROC_LOCK(p1->p_leader); 564 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 565 PROC_UNLOCK(p1->p_leader); 566 /* 567 * The task leader is exiting, so process p1 is 568 * going to be killed shortly. Since p1 obviously 569 * isn't dead yet, we know that the leader is either 570 * sending SIGKILL's to all the processes in this 571 * task or is sleeping waiting for all the peers to 572 * exit. We let p1 complete the fork, but we need 573 * to go ahead and kill the new process p2 since 574 * the task leader may not get a chance to send 575 * SIGKILL to it. We leave it on the list so that 576 * the task leader will wait for this new process 577 * to commit suicide. 578 */ 579 PROC_LOCK(p2); 580 kern_psignal(p2, SIGKILL); 581 PROC_UNLOCK(p2); 582 } else 583 PROC_UNLOCK(p1->p_leader); 584 } else { 585 p2->p_peers = NULL; 586 p2->p_leader = p2; 587 } 588 589 sx_xlock(&proctree_lock); 590 PGRP_LOCK(p1->p_pgrp); 591 PROC_LOCK(p2); 592 PROC_LOCK(p1); 593 594 /* 595 * Preserve some more flags in subprocess. P_PROFIL has already 596 * been preserved. 597 */ 598 p2->p_flag |= p1->p_flag & P_SUGID; 599 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; 600 SESS_LOCK(p1->p_session); 601 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 602 p2->p_flag |= P_CONTROLT; 603 SESS_UNLOCK(p1->p_session); 604 if (fr->fr_flags & RFPPWAIT) 605 p2->p_flag |= P_PPWAIT; 606 607 p2->p_pgrp = p1->p_pgrp; 608 LIST_INSERT_AFTER(p1, p2, p_pglist); 609 PGRP_UNLOCK(p1->p_pgrp); 610 LIST_INIT(&p2->p_children); 611 LIST_INIT(&p2->p_orphans); 612 613 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); 614 615 /* 616 * If PF_FORK is set, the child process inherits the 617 * procfs ioctl flags from its parent. 618 */ 619 if (p1->p_pfsflags & PF_FORK) { 620 p2->p_stops = p1->p_stops; 621 p2->p_pfsflags = p1->p_pfsflags; 622 } 623 624 /* 625 * This begins the section where we must prevent the parent 626 * from being swapped. 627 */ 628 _PHOLD(p1); 629 PROC_UNLOCK(p1); 630 631 /* 632 * Attach the new process to its parent. 633 * 634 * If RFNOWAIT is set, the newly created process becomes a child 635 * of init. This effectively disassociates the child from the 636 * parent. 637 */ 638 if ((fr->fr_flags & RFNOWAIT) != 0) { 639 pptr = p1->p_reaper; 640 p2->p_reaper = pptr; 641 } else { 642 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? 643 p1 : p1->p_reaper; 644 pptr = p1; 645 } 646 p2->p_pptr = pptr; 647 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 648 LIST_INIT(&p2->p_reaplist); 649 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); 650 if (p2->p_reaper == p1) 651 p2->p_reapsubtree = p2->p_pid; 652 sx_xunlock(&proctree_lock); 653 654 /* Inform accounting that we have forked. */ 655 p2->p_acflag = AFORK; 656 PROC_UNLOCK(p2); 657 658 #ifdef KTRACE 659 ktrprocfork(p1, p2); 660 #endif 661 662 /* 663 * Finish creating the child process. It will return via a different 664 * execution path later. (ie: directly into user mode) 665 */ 666 vm_forkproc(td, p2, td2, vm2, fr->fr_flags); 667 668 if (fr->fr_flags == (RFFDG | RFPROC)) { 669 PCPU_INC(cnt.v_forks); 670 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 671 p2->p_vmspace->vm_ssize); 672 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 673 PCPU_INC(cnt.v_vforks); 674 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 675 p2->p_vmspace->vm_ssize); 676 } else if (p1 == &proc0) { 677 PCPU_INC(cnt.v_kthreads); 678 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 679 p2->p_vmspace->vm_ssize); 680 } else { 681 PCPU_INC(cnt.v_rforks); 682 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 683 p2->p_vmspace->vm_ssize); 684 } 685 686 /* 687 * Associate the process descriptor with the process before anything 688 * can happen that might cause that process to need the descriptor. 689 * However, don't do this until after fork(2) can no longer fail. 690 */ 691 if (fr->fr_flags & RFPROCDESC) 692 procdesc_new(p2, fr->fr_pd_flags); 693 694 /* 695 * Both processes are set up, now check if any loadable modules want 696 * to adjust anything. 697 */ 698 EVENTHANDLER_INVOKE(process_fork, p1, p2, fr->fr_flags); 699 700 /* 701 * Set the child start time and mark the process as being complete. 702 */ 703 PROC_LOCK(p2); 704 PROC_LOCK(p1); 705 microuptime(&p2->p_stats->p_start); 706 PROC_SLOCK(p2); 707 p2->p_state = PRS_NORMAL; 708 PROC_SUNLOCK(p2); 709 710 #ifdef KDTRACE_HOOKS 711 /* 712 * Tell the DTrace fasttrap provider about the new process so that any 713 * tracepoints inherited from the parent can be removed. We have to do 714 * this only after p_state is PRS_NORMAL since the fasttrap module will 715 * use pfind() later on. 716 */ 717 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) 718 dtrace_fasttrap_fork(p1, p2); 719 #endif 720 /* 721 * Hold the process so that it cannot exit after we make it runnable, 722 * but before we wait for the debugger. 723 */ 724 _PHOLD(p2); 725 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED | 726 P_FOLLOWFORK)) { 727 /* 728 * Arrange for debugger to receive the fork event. 729 * 730 * We can report PL_FLAG_FORKED regardless of 731 * P_FOLLOWFORK settings, but it does not make a sense 732 * for runaway child. 733 */ 734 td->td_dbgflags |= TDB_FORK; 735 td->td_dbg_forked = p2->p_pid; 736 td2->td_dbgflags |= TDB_STOPATFORK; 737 } 738 if (fr->fr_flags & RFPPWAIT) { 739 td->td_pflags |= TDP_RFPPWAIT; 740 td->td_rfppwait_p = p2; 741 } 742 PROC_UNLOCK(p2); 743 744 /* 745 * Now can be swapped. 746 */ 747 _PRELE(p1); 748 PROC_UNLOCK(p1); 749 750 /* 751 * Tell any interested parties about the new process. 752 */ 753 knote_fork(&p1->p_klist, p2->p_pid); 754 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); 755 756 if (fr->fr_flags & RFPROCDESC) { 757 procdesc_finit(p2->p_procdesc, fp_procdesc); 758 fdrop(fp_procdesc, td); 759 } 760 761 if ((fr->fr_flags & RFSTOPPED) == 0) { 762 /* 763 * If RFSTOPPED not requested, make child runnable and 764 * add to run queue. 765 */ 766 thread_lock(td2); 767 TD_SET_CAN_RUN(td2); 768 sched_add(td2, SRQ_BORING); 769 thread_unlock(td2); 770 if (fr->fr_pidp != NULL) 771 *fr->fr_pidp = p2->p_pid; 772 } else { 773 *fr->fr_procp = p2; 774 } 775 776 PROC_LOCK(p2); 777 /* 778 * Wait until debugger is attached to child. 779 */ 780 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0) 781 cv_wait(&p2->p_dbgwait, &p2->p_mtx); 782 _PRELE(p2); 783 racct_proc_fork_done(p2); 784 PROC_UNLOCK(p2); 785 } 786 787 int 788 fork1(struct thread *td, struct fork_req *fr) 789 { 790 struct proc *p1, *newproc; 791 struct thread *td2; 792 struct vmspace *vm2; 793 struct file *fp_procdesc; 794 vm_ooffset_t mem_charged; 795 int error, nprocs_new, ok; 796 static int curfail; 797 static struct timeval lastfail; 798 int flags, pages; 799 800 flags = fr->fr_flags; 801 pages = fr->fr_pages; 802 803 if ((flags & RFSTOPPED) != 0) 804 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); 805 else 806 MPASS(fr->fr_procp == NULL); 807 808 /* Check for the undefined or unimplemented flags. */ 809 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 810 return (EINVAL); 811 812 /* Signal value requires RFTSIGZMB. */ 813 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 814 return (EINVAL); 815 816 /* Can't copy and clear. */ 817 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 818 return (EINVAL); 819 820 /* Check the validity of the signal number. */ 821 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 822 return (EINVAL); 823 824 if ((flags & RFPROCDESC) != 0) { 825 /* Can't not create a process yet get a process descriptor. */ 826 if ((flags & RFPROC) == 0) 827 return (EINVAL); 828 829 /* Must provide a place to put a procdesc if creating one. */ 830 if (fr->fr_pd_fd == NULL) 831 return (EINVAL); 832 } 833 834 p1 = td->td_proc; 835 836 /* 837 * Here we don't create a new process, but we divorce 838 * certain parts of a process from itself. 839 */ 840 if ((flags & RFPROC) == 0) { 841 if (fr->fr_procp != NULL) 842 *fr->fr_procp = NULL; 843 else if (fr->fr_pidp != NULL) 844 *fr->fr_pidp = 0; 845 return (fork_norfproc(td, flags)); 846 } 847 848 fp_procdesc = NULL; 849 newproc = NULL; 850 vm2 = NULL; 851 852 /* 853 * Increment the nprocs resource before allocations occur. 854 * Although process entries are dynamically created, we still 855 * keep a global limit on the maximum number we will 856 * create. There are hard-limits as to the number of processes 857 * that can run, established by the KVA and memory usage for 858 * the process data. 859 * 860 * Don't allow a nonprivileged user to use the last ten 861 * processes; don't let root exceed the limit. 862 */ 863 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; 864 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, 865 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { 866 error = EAGAIN; 867 sx_xlock(&allproc_lock); 868 if (ppsratecheck(&lastfail, &curfail, 1)) { 869 printf("maxproc limit exceeded by uid %u (pid %d); " 870 "see tuning(7) and login.conf(5)\n", 871 td->td_ucred->cr_ruid, p1->p_pid); 872 } 873 sx_xunlock(&allproc_lock); 874 goto fail2; 875 } 876 877 /* 878 * If required, create a process descriptor in the parent first; we 879 * will abandon it if something goes wrong. We don't finit() until 880 * later. 881 */ 882 if (flags & RFPROCDESC) { 883 error = falloc_caps(td, &fp_procdesc, fr->fr_pd_fd, 0, 884 fr->fr_pd_fcaps); 885 if (error != 0) 886 goto fail2; 887 } 888 889 mem_charged = 0; 890 if (pages == 0) 891 pages = kstack_pages; 892 /* Allocate new proc. */ 893 newproc = uma_zalloc(proc_zone, M_WAITOK); 894 td2 = FIRST_THREAD_IN_PROC(newproc); 895 if (td2 == NULL) { 896 td2 = thread_alloc(pages); 897 if (td2 == NULL) { 898 error = ENOMEM; 899 goto fail2; 900 } 901 proc_linkup(newproc, td2); 902 } else { 903 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 904 if (td2->td_kstack != 0) 905 vm_thread_dispose(td2); 906 if (!thread_alloc_stack(td2, pages)) { 907 error = ENOMEM; 908 goto fail2; 909 } 910 } 911 } 912 913 if ((flags & RFMEM) == 0) { 914 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 915 if (vm2 == NULL) { 916 error = ENOMEM; 917 goto fail2; 918 } 919 if (!swap_reserve(mem_charged)) { 920 /* 921 * The swap reservation failed. The accounting 922 * from the entries of the copied vm2 will be 923 * substracted in vmspace_free(), so force the 924 * reservation there. 925 */ 926 swap_reserve_force(mem_charged); 927 error = ENOMEM; 928 goto fail2; 929 } 930 } else 931 vm2 = NULL; 932 933 /* 934 * XXX: This is ugly; when we copy resource usage, we need to bump 935 * per-cred resource counters. 936 */ 937 proc_set_cred_init(newproc, crhold(td->td_ucred)); 938 939 /* 940 * Initialize resource accounting for the child process. 941 */ 942 error = racct_proc_fork(p1, newproc); 943 if (error != 0) { 944 error = EAGAIN; 945 goto fail1; 946 } 947 948 #ifdef MAC 949 mac_proc_init(newproc); 950 #endif 951 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); 952 STAILQ_INIT(&newproc->p_ktr); 953 954 /* We have to lock the process tree while we look for a pid. */ 955 sx_slock(&proctree_lock); 956 sx_xlock(&allproc_lock); 957 958 /* 959 * Increment the count of procs running with this uid. Don't allow 960 * a nonprivileged user to exceed their current limit. 961 * 962 * XXXRW: Can we avoid privilege here if it's not needed? 963 */ 964 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 965 if (error == 0) 966 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 967 else { 968 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 969 lim_cur(td, RLIMIT_NPROC)); 970 } 971 if (ok) { 972 do_fork(td, fr, newproc, td2, vm2, fp_procdesc); 973 return (0); 974 } 975 976 error = EAGAIN; 977 sx_sunlock(&proctree_lock); 978 sx_xunlock(&allproc_lock); 979 #ifdef MAC 980 mac_proc_destroy(newproc); 981 #endif 982 racct_proc_exit(newproc); 983 fail1: 984 crfree(newproc->p_ucred); 985 newproc->p_ucred = NULL; 986 fail2: 987 if (vm2 != NULL) 988 vmspace_free(vm2); 989 uma_zfree(proc_zone, newproc); 990 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { 991 fdclose(td, fp_procdesc, *fr->fr_pd_fd); 992 fdrop(fp_procdesc, td); 993 } 994 atomic_add_int(&nprocs, -1); 995 pause("fork", hz / 2); 996 return (error); 997 } 998 999 /* 1000 * Handle the return of a child process from fork1(). This function 1001 * is called from the MD fork_trampoline() entry point. 1002 */ 1003 void 1004 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 1005 struct trapframe *frame) 1006 { 1007 struct proc *p; 1008 struct thread *td; 1009 struct thread *dtd; 1010 1011 td = curthread; 1012 p = td->td_proc; 1013 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 1014 1015 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 1016 td, td->td_sched, p->p_pid, td->td_name); 1017 1018 sched_fork_exit(td); 1019 /* 1020 * Processes normally resume in mi_switch() after being 1021 * cpu_switch()'ed to, but when children start up they arrive here 1022 * instead, so we must do much the same things as mi_switch() would. 1023 */ 1024 if ((dtd = PCPU_GET(deadthread))) { 1025 PCPU_SET(deadthread, NULL); 1026 thread_stash(dtd); 1027 } 1028 thread_unlock(td); 1029 1030 /* 1031 * cpu_set_fork_handler intercepts this function call to 1032 * have this call a non-return function to stay in kernel mode. 1033 * initproc has its own fork handler, but it does return. 1034 */ 1035 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 1036 callout(arg, frame); 1037 1038 /* 1039 * Check if a kernel thread misbehaved and returned from its main 1040 * function. 1041 */ 1042 if (p->p_flag & P_KTHREAD) { 1043 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 1044 td->td_name, p->p_pid); 1045 kproc_exit(0); 1046 } 1047 mtx_assert(&Giant, MA_NOTOWNED); 1048 1049 if (p->p_sysent->sv_schedtail != NULL) 1050 (p->p_sysent->sv_schedtail)(td); 1051 td->td_pflags &= ~TDP_FORKING; 1052 } 1053 1054 /* 1055 * Simplified back end of syscall(), used when returning from fork() 1056 * directly into user mode. Giant is not held on entry, and must not 1057 * be held on return. This function is passed in to fork_exit() as the 1058 * first parameter and is called when returning to a new userland process. 1059 */ 1060 void 1061 fork_return(struct thread *td, struct trapframe *frame) 1062 { 1063 struct proc *p, *dbg; 1064 1065 p = td->td_proc; 1066 if (td->td_dbgflags & TDB_STOPATFORK) { 1067 sx_xlock(&proctree_lock); 1068 PROC_LOCK(p); 1069 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) == 1070 (P_TRACED | P_FOLLOWFORK)) { 1071 /* 1072 * If debugger still wants auto-attach for the 1073 * parent's children, do it now. 1074 */ 1075 dbg = p->p_pptr->p_pptr; 1076 p->p_flag |= P_TRACED; 1077 p->p_oppid = p->p_pptr->p_pid; 1078 CTR2(KTR_PTRACE, 1079 "fork_return: attaching to new child pid %d: oppid %d", 1080 p->p_pid, p->p_oppid); 1081 proc_reparent(p, dbg); 1082 sx_xunlock(&proctree_lock); 1083 td->td_dbgflags |= TDB_CHILD | TDB_SCX; 1084 ptracestop(td, SIGSTOP); 1085 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); 1086 } else { 1087 /* 1088 * ... otherwise clear the request. 1089 */ 1090 sx_xunlock(&proctree_lock); 1091 td->td_dbgflags &= ~TDB_STOPATFORK; 1092 cv_broadcast(&p->p_dbgwait); 1093 } 1094 PROC_UNLOCK(p); 1095 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { 1096 /* 1097 * This is the start of a new thread in a traced 1098 * process. Report a system call exit event. 1099 */ 1100 PROC_LOCK(p); 1101 td->td_dbgflags |= TDB_SCX; 1102 _STOPEVENT(p, S_SCX, td->td_dbg_sc_code); 1103 if ((p->p_stops & S_PT_SCX) != 0 || 1104 (td->td_dbgflags & TDB_BORN) != 0) 1105 ptracestop(td, SIGTRAP); 1106 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); 1107 PROC_UNLOCK(p); 1108 } 1109 1110 userret(td, frame); 1111 1112 #ifdef KTRACE 1113 if (KTRPOINT(td, KTR_SYSRET)) 1114 ktrsysret(SYS_fork, 0, 0); 1115 #endif 1116 } 1117