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_kdtrace.h" 41 #include "opt_ktrace.h" 42 #include "opt_kstack_pages.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/eventhandler.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/pioctl.h> 59 #include <sys/racct.h> 60 #include <sys/resourcevar.h> 61 #include <sys/sched.h> 62 #include <sys/syscall.h> 63 #include <sys/vmmeter.h> 64 #include <sys/vnode.h> 65 #include <sys/acct.h> 66 #include <sys/ktr.h> 67 #include <sys/ktrace.h> 68 #include <sys/unistd.h> 69 #include <sys/sdt.h> 70 #include <sys/sx.h> 71 #include <sys/sysent.h> 72 #include <sys/signalvar.h> 73 74 #include <security/audit/audit.h> 75 #include <security/mac/mac_framework.h> 76 77 #include <vm/vm.h> 78 #include <vm/pmap.h> 79 #include <vm/vm_map.h> 80 #include <vm/vm_extern.h> 81 #include <vm/uma.h> 82 83 #ifdef KDTRACE_HOOKS 84 #include <sys/dtrace_bsd.h> 85 dtrace_fork_func_t dtrace_fasttrap_fork; 86 #endif 87 88 SDT_PROVIDER_DECLARE(proc); 89 SDT_PROBE_DEFINE(proc, kernel, , create, create); 90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *"); 91 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *"); 92 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int"); 93 94 #ifndef _SYS_SYSPROTO_H_ 95 struct fork_args { 96 int dummy; 97 }; 98 #endif 99 100 /* ARGSUSED */ 101 int 102 fork(struct thread *td, struct fork_args *uap) 103 { 104 int error; 105 struct proc *p2; 106 107 error = fork1(td, RFFDG | RFPROC, 0, &p2); 108 if (error == 0) { 109 td->td_retval[0] = p2->p_pid; 110 td->td_retval[1] = 0; 111 } 112 return (error); 113 } 114 115 /* ARGSUSED */ 116 int 117 vfork(struct thread *td, struct vfork_args *uap) 118 { 119 int error, flags; 120 struct proc *p2; 121 122 #ifdef XEN 123 flags = RFFDG | RFPROC; /* validate that this is still an issue */ 124 #else 125 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 126 #endif 127 error = fork1(td, flags, 0, &p2); 128 if (error == 0) { 129 td->td_retval[0] = p2->p_pid; 130 td->td_retval[1] = 0; 131 } 132 return (error); 133 } 134 135 int 136 rfork(struct thread *td, struct rfork_args *uap) 137 { 138 struct proc *p2; 139 int error; 140 141 /* Don't allow kernel-only flags. */ 142 if ((uap->flags & RFKERNELONLY) != 0) 143 return (EINVAL); 144 145 AUDIT_ARG_FFLAGS(uap->flags); 146 error = fork1(td, uap->flags, 0, &p2); 147 if (error == 0) { 148 td->td_retval[0] = p2 ? p2->p_pid : 0; 149 td->td_retval[1] = 0; 150 } 151 return (error); 152 } 153 154 int nprocs = 1; /* process 0 */ 155 int lastpid = 0; 156 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 157 "Last used PID"); 158 159 /* 160 * Random component to lastpid generation. We mix in a random factor to make 161 * it a little harder to predict. We sanity check the modulus value to avoid 162 * doing it in critical paths. Don't let it be too small or we pointlessly 163 * waste randomness entropy, and don't let it be impossibly large. Using a 164 * modulus that is too big causes a LOT more process table scans and slows 165 * down fork processing as the pidchecked caching is defeated. 166 */ 167 static int randompid = 0; 168 169 static int 170 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 171 { 172 int error, pid; 173 174 error = sysctl_wire_old_buffer(req, sizeof(int)); 175 if (error != 0) 176 return(error); 177 sx_xlock(&allproc_lock); 178 pid = randompid; 179 error = sysctl_handle_int(oidp, &pid, 0, req); 180 if (error == 0 && req->newptr != NULL) { 181 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 182 pid = PID_MAX - 100; 183 else if (pid < 2) /* NOP */ 184 pid = 0; 185 else if (pid < 100) /* Make it reasonable */ 186 pid = 100; 187 randompid = pid; 188 } 189 sx_xunlock(&allproc_lock); 190 return (error); 191 } 192 193 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 194 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 195 196 static int 197 fork_findpid(int flags) 198 { 199 struct proc *p; 200 int trypid; 201 static int pidchecked = 0; 202 203 /* 204 * Requires allproc_lock in order to iterate over the list 205 * of processes, and proctree_lock to access p_pgrp. 206 */ 207 sx_assert(&allproc_lock, SX_LOCKED); 208 sx_assert(&proctree_lock, SX_LOCKED); 209 210 /* 211 * Find an unused process ID. We remember a range of unused IDs 212 * ready to use (from lastpid+1 through pidchecked-1). 213 * 214 * If RFHIGHPID is set (used during system boot), do not allocate 215 * low-numbered pids. 216 */ 217 trypid = lastpid + 1; 218 if (flags & RFHIGHPID) { 219 if (trypid < 10) 220 trypid = 10; 221 } else { 222 if (randompid) 223 trypid += arc4random() % randompid; 224 } 225 retry: 226 /* 227 * If the process ID prototype has wrapped around, 228 * restart somewhat above 0, as the low-numbered procs 229 * tend to include daemons that don't exit. 230 */ 231 if (trypid >= PID_MAX) { 232 trypid = trypid % PID_MAX; 233 if (trypid < 100) 234 trypid += 100; 235 pidchecked = 0; 236 } 237 if (trypid >= pidchecked) { 238 int doingzomb = 0; 239 240 pidchecked = PID_MAX; 241 /* 242 * Scan the active and zombie procs to check whether this pid 243 * is in use. Remember the lowest pid that's greater 244 * than trypid, so we can avoid checking for a while. 245 */ 246 p = LIST_FIRST(&allproc); 247 again: 248 for (; p != NULL; p = LIST_NEXT(p, p_list)) { 249 while (p->p_pid == trypid || 250 (p->p_pgrp != NULL && 251 (p->p_pgrp->pg_id == trypid || 252 (p->p_session != NULL && 253 p->p_session->s_sid == trypid)))) { 254 trypid++; 255 if (trypid >= pidchecked) 256 goto retry; 257 } 258 if (p->p_pid > trypid && pidchecked > p->p_pid) 259 pidchecked = p->p_pid; 260 if (p->p_pgrp != NULL) { 261 if (p->p_pgrp->pg_id > trypid && 262 pidchecked > p->p_pgrp->pg_id) 263 pidchecked = p->p_pgrp->pg_id; 264 if (p->p_session != NULL && 265 p->p_session->s_sid > trypid && 266 pidchecked > p->p_session->s_sid) 267 pidchecked = p->p_session->s_sid; 268 } 269 } 270 if (!doingzomb) { 271 doingzomb = 1; 272 p = LIST_FIRST(&zombproc); 273 goto again; 274 } 275 } 276 277 /* 278 * RFHIGHPID does not mess with the lastpid counter during boot. 279 */ 280 if (flags & RFHIGHPID) 281 pidchecked = 0; 282 else 283 lastpid = trypid; 284 285 return (trypid); 286 } 287 288 static int 289 fork_norfproc(struct thread *td, int flags) 290 { 291 int error; 292 struct proc *p1; 293 294 KASSERT((flags & RFPROC) == 0, 295 ("fork_norfproc called with RFPROC set")); 296 p1 = td->td_proc; 297 298 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 299 (flags & (RFCFDG | RFFDG))) { 300 PROC_LOCK(p1); 301 if (thread_single(SINGLE_BOUNDARY)) { 302 PROC_UNLOCK(p1); 303 return (ERESTART); 304 } 305 PROC_UNLOCK(p1); 306 } 307 308 error = vm_forkproc(td, NULL, NULL, NULL, flags); 309 if (error) 310 goto fail; 311 312 /* 313 * Close all file descriptors. 314 */ 315 if (flags & RFCFDG) { 316 struct filedesc *fdtmp; 317 fdtmp = fdinit(td->td_proc->p_fd); 318 fdfree(td); 319 p1->p_fd = fdtmp; 320 } 321 322 /* 323 * Unshare file descriptors (from parent). 324 */ 325 if (flags & RFFDG) 326 fdunshare(p1, td); 327 328 fail: 329 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 330 (flags & (RFCFDG | RFFDG))) { 331 PROC_LOCK(p1); 332 thread_single_end(); 333 PROC_UNLOCK(p1); 334 } 335 return (error); 336 } 337 338 static void 339 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2, 340 struct vmspace *vm2) 341 { 342 struct proc *p1, *pptr; 343 int p2_held, trypid; 344 struct filedesc *fd; 345 struct filedesc_to_leader *fdtol; 346 struct sigacts *newsigacts; 347 348 sx_assert(&proctree_lock, SX_SLOCKED); 349 sx_assert(&allproc_lock, SX_XLOCKED); 350 351 p2_held = 0; 352 p1 = td->td_proc; 353 354 /* 355 * Increment the nprocs resource before blocking can occur. There 356 * are hard-limits as to the number of processes that can run. 357 */ 358 nprocs++; 359 360 trypid = fork_findpid(flags); 361 362 sx_sunlock(&proctree_lock); 363 364 p2->p_state = PRS_NEW; /* protect against others */ 365 p2->p_pid = trypid; 366 AUDIT_ARG_PID(p2->p_pid); 367 LIST_INSERT_HEAD(&allproc, p2, p_list); 368 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 369 tidhash_add(td2); 370 PROC_LOCK(p2); 371 PROC_LOCK(p1); 372 373 sx_xunlock(&allproc_lock); 374 375 bcopy(&p1->p_startcopy, &p2->p_startcopy, 376 __rangeof(struct proc, p_startcopy, p_endcopy)); 377 pargs_hold(p2->p_args); 378 PROC_UNLOCK(p1); 379 380 bzero(&p2->p_startzero, 381 __rangeof(struct proc, p_startzero, p_endzero)); 382 383 p2->p_ucred = crhold(td->td_ucred); 384 385 /* Tell the prison that we exist. */ 386 prison_proc_hold(p2->p_ucred->cr_prison); 387 388 PROC_UNLOCK(p2); 389 390 /* 391 * Malloc things while we don't hold any locks. 392 */ 393 if (flags & RFSIGSHARE) 394 newsigacts = NULL; 395 else 396 newsigacts = sigacts_alloc(); 397 398 /* 399 * Copy filedesc. 400 */ 401 if (flags & RFCFDG) { 402 fd = fdinit(p1->p_fd); 403 fdtol = NULL; 404 } else if (flags & RFFDG) { 405 fd = fdcopy(p1->p_fd); 406 fdtol = NULL; 407 } else { 408 fd = fdshare(p1->p_fd); 409 if (p1->p_fdtol == NULL) 410 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, 411 p1->p_leader); 412 if ((flags & RFTHREAD) != 0) { 413 /* 414 * Shared file descriptor table, and shared 415 * process leaders. 416 */ 417 fdtol = p1->p_fdtol; 418 FILEDESC_XLOCK(p1->p_fd); 419 fdtol->fdl_refcount++; 420 FILEDESC_XUNLOCK(p1->p_fd); 421 } else { 422 /* 423 * Shared file descriptor table, and different 424 * process leaders. 425 */ 426 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 427 p1->p_fd, p2); 428 } 429 } 430 /* 431 * Make a proc table entry for the new process. 432 * Start by zeroing the section of proc that is zero-initialized, 433 * then copy the section that is copied directly from the parent. 434 */ 435 436 PROC_LOCK(p2); 437 PROC_LOCK(p1); 438 439 bzero(&td2->td_startzero, 440 __rangeof(struct thread, td_startzero, td_endzero)); 441 442 bcopy(&td->td_startcopy, &td2->td_startcopy, 443 __rangeof(struct thread, td_startcopy, td_endcopy)); 444 445 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 446 td2->td_sigstk = td->td_sigstk; 447 td2->td_sigmask = td->td_sigmask; 448 td2->td_flags = TDF_INMEM; 449 td2->td_lend_user_pri = PRI_MAX; 450 451 #ifdef VIMAGE 452 td2->td_vnet = NULL; 453 td2->td_vnet_lpush = NULL; 454 #endif 455 456 /* 457 * Allow the scheduler to initialize the child. 458 */ 459 thread_lock(td); 460 sched_fork(td, td2); 461 thread_unlock(td); 462 463 /* 464 * Duplicate sub-structures as needed. 465 * Increase reference counts on shared objects. 466 */ 467 p2->p_flag = P_INMEM; 468 p2->p_swtick = ticks; 469 if (p1->p_flag & P_PROFIL) 470 startprofclock(p2); 471 td2->td_ucred = crhold(p2->p_ucred); 472 473 if (flags & RFSIGSHARE) { 474 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 475 } else { 476 sigacts_copy(newsigacts, p1->p_sigacts); 477 p2->p_sigacts = newsigacts; 478 } 479 480 if (flags & RFTSIGZMB) 481 p2->p_sigparent = RFTSIGNUM(flags); 482 else if (flags & RFLINUXTHPN) 483 p2->p_sigparent = SIGUSR1; 484 else 485 p2->p_sigparent = SIGCHLD; 486 487 p2->p_textvp = p1->p_textvp; 488 p2->p_fd = fd; 489 p2->p_fdtol = fdtol; 490 491 /* 492 * p_limit is copy-on-write. Bump its refcount. 493 */ 494 lim_fork(p1, p2); 495 496 pstats_fork(p1->p_stats, p2->p_stats); 497 498 PROC_UNLOCK(p1); 499 PROC_UNLOCK(p2); 500 501 /* Bump references to the text vnode (for procfs). */ 502 if (p2->p_textvp) 503 vref(p2->p_textvp); 504 505 /* 506 * Set up linkage for kernel based threading. 507 */ 508 if ((flags & RFTHREAD) != 0) { 509 mtx_lock(&ppeers_lock); 510 p2->p_peers = p1->p_peers; 511 p1->p_peers = p2; 512 p2->p_leader = p1->p_leader; 513 mtx_unlock(&ppeers_lock); 514 PROC_LOCK(p1->p_leader); 515 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 516 PROC_UNLOCK(p1->p_leader); 517 /* 518 * The task leader is exiting, so process p1 is 519 * going to be killed shortly. Since p1 obviously 520 * isn't dead yet, we know that the leader is either 521 * sending SIGKILL's to all the processes in this 522 * task or is sleeping waiting for all the peers to 523 * exit. We let p1 complete the fork, but we need 524 * to go ahead and kill the new process p2 since 525 * the task leader may not get a chance to send 526 * SIGKILL to it. We leave it on the list so that 527 * the task leader will wait for this new process 528 * to commit suicide. 529 */ 530 PROC_LOCK(p2); 531 psignal(p2, SIGKILL); 532 PROC_UNLOCK(p2); 533 } else 534 PROC_UNLOCK(p1->p_leader); 535 } else { 536 p2->p_peers = NULL; 537 p2->p_leader = p2; 538 } 539 540 sx_xlock(&proctree_lock); 541 PGRP_LOCK(p1->p_pgrp); 542 PROC_LOCK(p2); 543 PROC_LOCK(p1); 544 545 /* 546 * Preserve some more flags in subprocess. P_PROFIL has already 547 * been preserved. 548 */ 549 p2->p_flag |= p1->p_flag & P_SUGID; 550 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 551 SESS_LOCK(p1->p_session); 552 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 553 p2->p_flag |= P_CONTROLT; 554 SESS_UNLOCK(p1->p_session); 555 if (flags & RFPPWAIT) 556 p2->p_flag |= P_PPWAIT; 557 558 p2->p_pgrp = p1->p_pgrp; 559 LIST_INSERT_AFTER(p1, p2, p_pglist); 560 PGRP_UNLOCK(p1->p_pgrp); 561 LIST_INIT(&p2->p_children); 562 563 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 564 565 /* 566 * If PF_FORK is set, the child process inherits the 567 * procfs ioctl flags from its parent. 568 */ 569 if (p1->p_pfsflags & PF_FORK) { 570 p2->p_stops = p1->p_stops; 571 p2->p_pfsflags = p1->p_pfsflags; 572 } 573 574 /* 575 * This begins the section where we must prevent the parent 576 * from being swapped. 577 */ 578 _PHOLD(p1); 579 PROC_UNLOCK(p1); 580 581 /* 582 * Attach the new process to its parent. 583 * 584 * If RFNOWAIT is set, the newly created process becomes a child 585 * of init. This effectively disassociates the child from the 586 * parent. 587 */ 588 if (flags & RFNOWAIT) 589 pptr = initproc; 590 else 591 pptr = p1; 592 p2->p_pptr = pptr; 593 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 594 sx_xunlock(&proctree_lock); 595 596 /* Inform accounting that we have forked. */ 597 p2->p_acflag = AFORK; 598 PROC_UNLOCK(p2); 599 600 #ifdef KTRACE 601 ktrprocfork(p1, p2); 602 #endif 603 604 /* 605 * Finish creating the child process. It will return via a different 606 * execution path later. (ie: directly into user mode) 607 */ 608 vm_forkproc(td, p2, td2, vm2, flags); 609 610 if (flags == (RFFDG | RFPROC)) { 611 PCPU_INC(cnt.v_forks); 612 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 613 p2->p_vmspace->vm_ssize); 614 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 615 PCPU_INC(cnt.v_vforks); 616 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 617 p2->p_vmspace->vm_ssize); 618 } else if (p1 == &proc0) { 619 PCPU_INC(cnt.v_kthreads); 620 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 621 p2->p_vmspace->vm_ssize); 622 } else { 623 PCPU_INC(cnt.v_rforks); 624 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 625 p2->p_vmspace->vm_ssize); 626 } 627 628 /* 629 * Both processes are set up, now check if any loadable modules want 630 * to adjust anything. 631 */ 632 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 633 634 /* 635 * Set the child start time and mark the process as being complete. 636 */ 637 PROC_LOCK(p2); 638 PROC_LOCK(p1); 639 microuptime(&p2->p_stats->p_start); 640 PROC_SLOCK(p2); 641 p2->p_state = PRS_NORMAL; 642 PROC_SUNLOCK(p2); 643 644 #ifdef KDTRACE_HOOKS 645 /* 646 * Tell the DTrace fasttrap provider about the new process 647 * if it has registered an interest. We have to do this only after 648 * p_state is PRS_NORMAL since the fasttrap module will use pfind() 649 * later on. 650 */ 651 if (dtrace_fasttrap_fork) 652 dtrace_fasttrap_fork(p1, p2); 653 #endif 654 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED | 655 P_FOLLOWFORK)) { 656 /* 657 * Arrange for debugger to receive the fork event. 658 * 659 * We can report PL_FLAG_FORKED regardless of 660 * P_FOLLOWFORK settings, but it does not make a sense 661 * for runaway child. 662 */ 663 td->td_dbgflags |= TDB_FORK; 664 td->td_dbg_forked = p2->p_pid; 665 td2->td_dbgflags |= TDB_STOPATFORK; 666 _PHOLD(p2); 667 p2_held = 1; 668 } 669 PROC_UNLOCK(p2); 670 if ((flags & RFSTOPPED) == 0) { 671 /* 672 * If RFSTOPPED not requested, make child runnable and 673 * add to run queue. 674 */ 675 thread_lock(td2); 676 TD_SET_CAN_RUN(td2); 677 sched_add(td2, SRQ_BORING); 678 thread_unlock(td2); 679 } 680 681 /* 682 * Now can be swapped. 683 */ 684 _PRELE(p1); 685 PROC_UNLOCK(p1); 686 687 /* 688 * Tell any interested parties about the new process. 689 */ 690 knote_fork(&p1->p_klist, p2->p_pid); 691 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); 692 693 /* 694 * Wait until debugger is attached to child. 695 */ 696 PROC_LOCK(p2); 697 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0) 698 cv_wait(&p2->p_dbgwait, &p2->p_mtx); 699 if (p2_held) 700 _PRELE(p2); 701 702 /* 703 * Preserve synchronization semantics of vfork. If waiting for 704 * child to exec or exit, set P_PPWAIT on child, and sleep on our 705 * proc (in case of exit). 706 */ 707 while (p2->p_flag & P_PPWAIT) 708 cv_wait(&p2->p_pwait, &p2->p_mtx); 709 PROC_UNLOCK(p2); 710 } 711 712 int 713 fork1(struct thread *td, int flags, int pages, struct proc **procp) 714 { 715 struct proc *p1; 716 struct proc *newproc; 717 int ok; 718 struct thread *td2; 719 struct vmspace *vm2; 720 vm_ooffset_t mem_charged; 721 int error; 722 static int curfail; 723 static struct timeval lastfail; 724 725 /* Check for the undefined or unimplemented flags. */ 726 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) 727 return (EINVAL); 728 729 /* Signal value requires RFTSIGZMB. */ 730 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) 731 return (EINVAL); 732 733 /* Can't copy and clear. */ 734 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 735 return (EINVAL); 736 737 /* Check the validity of the signal number. */ 738 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) 739 return (EINVAL); 740 741 p1 = td->td_proc; 742 743 /* 744 * Here we don't create a new process, but we divorce 745 * certain parts of a process from itself. 746 */ 747 if ((flags & RFPROC) == 0) { 748 *procp = NULL; 749 return (fork_norfproc(td, flags)); 750 } 751 752 #ifdef RACCT 753 PROC_LOCK(p1); 754 error = racct_add(p1, RACCT_NPROC, 1); 755 PROC_UNLOCK(p1); 756 if (error != 0) 757 return (EAGAIN); 758 #endif 759 760 mem_charged = 0; 761 vm2 = NULL; 762 if (pages == 0) 763 pages = KSTACK_PAGES; 764 /* Allocate new proc. */ 765 newproc = uma_zalloc(proc_zone, M_WAITOK); 766 td2 = FIRST_THREAD_IN_PROC(newproc); 767 if (td2 == NULL) { 768 td2 = thread_alloc(pages); 769 if (td2 == NULL) { 770 error = ENOMEM; 771 goto fail1; 772 } 773 proc_linkup(newproc, td2); 774 } else { 775 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { 776 if (td2->td_kstack != 0) 777 vm_thread_dispose(td2); 778 if (!thread_alloc_stack(td2, pages)) { 779 error = ENOMEM; 780 goto fail1; 781 } 782 } 783 } 784 785 if ((flags & RFMEM) == 0) { 786 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 787 if (vm2 == NULL) { 788 error = ENOMEM; 789 goto fail1; 790 } 791 if (!swap_reserve(mem_charged)) { 792 /* 793 * The swap reservation failed. The accounting 794 * from the entries of the copied vm2 will be 795 * substracted in vmspace_free(), so force the 796 * reservation there. 797 */ 798 swap_reserve_force(mem_charged); 799 error = ENOMEM; 800 goto fail1; 801 } 802 } else 803 vm2 = NULL; 804 #ifdef MAC 805 mac_proc_init(newproc); 806 #endif 807 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); 808 STAILQ_INIT(&newproc->p_ktr); 809 810 /* 811 * XXX: This is ugly; when we copy resource usage, we need to bump 812 * per-cred resource counters. 813 */ 814 newproc->p_ucred = p1->p_ucred; 815 816 /* 817 * Initialize resource accounting for the child process. 818 */ 819 error = racct_proc_fork(p1, newproc); 820 if (error != 0) { 821 error = EAGAIN; 822 goto fail1; 823 } 824 825 /* We have to lock the process tree while we look for a pid. */ 826 sx_slock(&proctree_lock); 827 828 /* 829 * Although process entries are dynamically created, we still keep 830 * a global limit on the maximum number we will create. Don't allow 831 * a nonprivileged user to use the last ten processes; don't let root 832 * exceed the limit. The variable nprocs is the current number of 833 * processes, maxproc is the limit. 834 */ 835 sx_xlock(&allproc_lock); 836 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 837 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 838 error = EAGAIN; 839 goto fail; 840 } 841 842 #ifdef RACCT 843 /* 844 * After fork, there is exactly one thread running. 845 */ 846 PROC_LOCK(newproc); 847 error = racct_set(newproc, RACCT_NTHR, 1); 848 PROC_UNLOCK(newproc); 849 if (error != 0) { 850 error = EAGAIN; 851 goto fail; 852 } 853 #endif 854 855 /* 856 * Increment the count of procs running with this uid. Don't allow 857 * a nonprivileged user to exceed their current limit. 858 * 859 * XXXRW: Can we avoid privilege here if it's not needed? 860 */ 861 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 862 if (error == 0) 863 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 864 else { 865 PROC_LOCK(p1); 866 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 867 lim_cur(p1, RLIMIT_NPROC)); 868 PROC_UNLOCK(p1); 869 } 870 if (ok) { 871 do_fork(td, flags, newproc, td2, vm2); 872 873 /* 874 * Return child proc pointer to parent. 875 */ 876 *procp = newproc; 877 return (0); 878 } 879 880 error = EAGAIN; 881 fail: 882 racct_proc_exit(newproc); 883 sx_sunlock(&proctree_lock); 884 if (ppsratecheck(&lastfail, &curfail, 1)) 885 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 886 td->td_ucred->cr_ruid); 887 sx_xunlock(&allproc_lock); 888 #ifdef MAC 889 mac_proc_destroy(newproc); 890 #endif 891 fail1: 892 if (vm2 != NULL) 893 vmspace_free(vm2); 894 uma_zfree(proc_zone, newproc); 895 pause("fork", hz / 2); 896 #ifdef RACCT 897 PROC_LOCK(p1); 898 racct_sub(p1, RACCT_NPROC, 1); 899 PROC_UNLOCK(p1); 900 #endif 901 return (error); 902 } 903 904 /* 905 * Handle the return of a child process from fork1(). This function 906 * is called from the MD fork_trampoline() entry point. 907 */ 908 void 909 fork_exit(void (*callout)(void *, struct trapframe *), void *arg, 910 struct trapframe *frame) 911 { 912 struct proc *p; 913 struct thread *td; 914 struct thread *dtd; 915 916 td = curthread; 917 p = td->td_proc; 918 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 919 920 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 921 td, td->td_sched, p->p_pid, td->td_name); 922 923 sched_fork_exit(td); 924 /* 925 * Processes normally resume in mi_switch() after being 926 * cpu_switch()'ed to, but when children start up they arrive here 927 * instead, so we must do much the same things as mi_switch() would. 928 */ 929 if ((dtd = PCPU_GET(deadthread))) { 930 PCPU_SET(deadthread, NULL); 931 thread_stash(dtd); 932 } 933 thread_unlock(td); 934 935 /* 936 * cpu_set_fork_handler intercepts this function call to 937 * have this call a non-return function to stay in kernel mode. 938 * initproc has its own fork handler, but it does return. 939 */ 940 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 941 callout(arg, frame); 942 943 /* 944 * Check if a kernel thread misbehaved and returned from its main 945 * function. 946 */ 947 if (p->p_flag & P_KTHREAD) { 948 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 949 td->td_name, p->p_pid); 950 kproc_exit(0); 951 } 952 mtx_assert(&Giant, MA_NOTOWNED); 953 954 if (p->p_sysent->sv_schedtail != NULL) 955 (p->p_sysent->sv_schedtail)(td); 956 } 957 958 /* 959 * Simplified back end of syscall(), used when returning from fork() 960 * directly into user mode. Giant is not held on entry, and must not 961 * be held on return. This function is passed in to fork_exit() as the 962 * first parameter and is called when returning to a new userland process. 963 */ 964 void 965 fork_return(struct thread *td, struct trapframe *frame) 966 { 967 struct proc *p, *dbg; 968 969 if (td->td_dbgflags & TDB_STOPATFORK) { 970 p = td->td_proc; 971 sx_xlock(&proctree_lock); 972 PROC_LOCK(p); 973 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) == 974 (P_TRACED | P_FOLLOWFORK)) { 975 /* 976 * If debugger still wants auto-attach for the 977 * parent's children, do it now. 978 */ 979 dbg = p->p_pptr->p_pptr; 980 p->p_flag |= P_TRACED; 981 p->p_oppid = p->p_pptr->p_pid; 982 proc_reparent(p, dbg); 983 sx_xunlock(&proctree_lock); 984 ptracestop(td, SIGSTOP); 985 } else { 986 /* 987 * ... otherwise clear the request. 988 */ 989 sx_xunlock(&proctree_lock); 990 td->td_dbgflags &= ~TDB_STOPATFORK; 991 cv_broadcast(&p->p_dbgwait); 992 } 993 PROC_UNLOCK(p); 994 } 995 996 userret(td, frame); 997 998 #ifdef KTRACE 999 if (KTRPOINT(td, KTR_SYSRET)) 1000 ktrsysret(SYS_fork, 0, 0); 1001 #endif 1002 mtx_assert(&Giant, MA_NOTOWNED); 1003 } 1004