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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 39 * $FreeBSD$ 40 */ 41 42 #include "opt_ktrace.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/filedesc.h> 48 #include <sys/kernel.h> 49 #include <sys/sysctl.h> 50 #include <sys/lock.h> 51 #include <sys/malloc.h> 52 #include <sys/mutex.h> 53 #include <sys/proc.h> 54 #include <sys/pioctl.h> 55 #include <sys/resourcevar.h> 56 #include <sys/sched.h> 57 #include <sys/syscall.h> 58 #include <sys/vnode.h> 59 #include <sys/acct.h> 60 #include <sys/ktr.h> 61 #include <sys/ktrace.h> 62 #include <sys/kthread.h> 63 #include <sys/unistd.h> 64 #include <sys/jail.h> 65 #include <sys/sx.h> 66 67 #include <vm/vm.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_map.h> 70 #include <vm/vm_extern.h> 71 #include <vm/uma.h> 72 73 #include <sys/vmmeter.h> 74 #include <sys/user.h> 75 #include <machine/critical.h> 76 77 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 78 79 /* 80 * These are the stuctures used to create a callout list for things to do 81 * when forking a process 82 */ 83 struct forklist { 84 forklist_fn function; 85 TAILQ_ENTRY(forklist) next; 86 }; 87 88 static struct sx fork_list_lock; 89 90 TAILQ_HEAD(forklist_head, forklist); 91 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 92 93 #ifndef _SYS_SYSPROTO_H_ 94 struct fork_args { 95 int dummy; 96 }; 97 #endif 98 99 int forksleep; /* Place for fork1() to sleep on. */ 100 101 static void 102 init_fork_list(void *data __unused) 103 { 104 105 sx_init(&fork_list_lock, "fork list"); 106 } 107 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL); 108 109 /* 110 * MPSAFE 111 */ 112 /* ARGSUSED */ 113 int 114 fork(td, uap) 115 struct thread *td; 116 struct fork_args *uap; 117 { 118 int error; 119 struct proc *p2; 120 121 mtx_lock(&Giant); 122 error = fork1(td, RFFDG | RFPROC, 0, &p2); 123 if (error == 0) { 124 td->td_retval[0] = p2->p_pid; 125 td->td_retval[1] = 0; 126 } 127 mtx_unlock(&Giant); 128 return error; 129 } 130 131 /* 132 * MPSAFE 133 */ 134 /* ARGSUSED */ 135 int 136 vfork(td, uap) 137 struct thread *td; 138 struct vfork_args *uap; 139 { 140 int error; 141 struct proc *p2; 142 143 mtx_lock(&Giant); 144 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 145 if (error == 0) { 146 td->td_retval[0] = p2->p_pid; 147 td->td_retval[1] = 0; 148 } 149 mtx_unlock(&Giant); 150 return error; 151 } 152 153 /* 154 * MPSAFE 155 */ 156 int 157 rfork(td, uap) 158 struct thread *td; 159 struct rfork_args *uap; 160 { 161 int error; 162 struct proc *p2; 163 164 /* Don't allow kernel only flags. */ 165 if ((uap->flags & RFKERNELONLY) != 0) 166 return (EINVAL); 167 mtx_lock(&Giant); 168 error = fork1(td, uap->flags, 0, &p2); 169 if (error == 0) { 170 td->td_retval[0] = p2 ? p2->p_pid : 0; 171 td->td_retval[1] = 0; 172 } 173 mtx_unlock(&Giant); 174 return error; 175 } 176 177 178 int nprocs = 1; /* process 0 */ 179 int lastpid = 0; 180 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 181 "Last used PID"); 182 183 /* 184 * Random component to lastpid generation. We mix in a random factor to make 185 * it a little harder to predict. We sanity check the modulus value to avoid 186 * doing it in critical paths. Don't let it be too small or we pointlessly 187 * waste randomness entropy, and don't let it be impossibly large. Using a 188 * modulus that is too big causes a LOT more process table scans and slows 189 * down fork processing as the pidchecked caching is defeated. 190 */ 191 static int randompid = 0; 192 193 static int 194 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 195 { 196 int error, pid; 197 198 sysctl_wire_old_buffer(req, sizeof(int)); 199 sx_xlock(&allproc_lock); 200 pid = randompid; 201 error = sysctl_handle_int(oidp, &pid, 0, req); 202 if (error == 0 && req->newptr != NULL) { 203 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 204 pid = PID_MAX - 100; 205 else if (pid < 2) /* NOP */ 206 pid = 0; 207 else if (pid < 100) /* Make it reasonable */ 208 pid = 100; 209 randompid = pid; 210 } 211 sx_xunlock(&allproc_lock); 212 return (error); 213 } 214 215 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 216 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 217 218 int 219 fork1(td, flags, pages, procp) 220 struct thread *td; /* parent proc */ 221 int flags; 222 int pages; 223 struct proc **procp; /* child proc */ 224 { 225 struct proc *p2, *pptr; 226 uid_t uid; 227 struct proc *newproc; 228 int trypid; 229 int ok; 230 static int pidchecked = 0; 231 struct forklist *ep; 232 struct filedesc *fd; 233 struct proc *p1 = td->td_proc; 234 struct thread *td2; 235 struct kse *ke2; 236 struct ksegrp *kg2; 237 struct sigacts *newsigacts; 238 struct procsig *newprocsig; 239 240 GIANT_REQUIRED; 241 242 /* Can't copy and clear */ 243 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 244 return (EINVAL); 245 246 /* 247 * Here we don't create a new process, but we divorce 248 * certain parts of a process from itself. 249 */ 250 if ((flags & RFPROC) == 0) { 251 vm_forkproc(td, NULL, NULL, flags); 252 253 /* 254 * Close all file descriptors. 255 */ 256 if (flags & RFCFDG) { 257 struct filedesc *fdtmp; 258 fdtmp = fdinit(td); /* XXXKSE */ 259 PROC_LOCK(p1); 260 fdfree(td); /* XXXKSE */ 261 p1->p_fd = fdtmp; 262 PROC_UNLOCK(p1); 263 } 264 265 /* 266 * Unshare file descriptors (from parent.) 267 */ 268 if (flags & RFFDG) { 269 FILEDESC_LOCK(p1->p_fd); 270 if (p1->p_fd->fd_refcnt > 1) { 271 struct filedesc *newfd; 272 273 newfd = fdcopy(td); 274 FILEDESC_UNLOCK(p1->p_fd); 275 PROC_LOCK(p1); 276 fdfree(td); 277 p1->p_fd = newfd; 278 PROC_UNLOCK(p1); 279 } else 280 FILEDESC_UNLOCK(p1->p_fd); 281 } 282 *procp = NULL; 283 return (0); 284 } 285 286 if (p1->p_flag & P_KSES) { 287 /* 288 * Idle the other threads for a second. 289 * Since the user space is copied, it must remain stable. 290 * In addition, all threads (from the user perspective) 291 * need to either be suspended or in the kernel, 292 * where they will try restart in the parent and will 293 * be aborted in the child. 294 */ 295 PROC_LOCK(p1); 296 if (thread_single(SINGLE_NO_EXIT)) { 297 /* Abort.. someone else is single threading before us */ 298 PROC_UNLOCK(p1); 299 return (ERESTART); 300 } 301 PROC_UNLOCK(p1); 302 /* 303 * All other activity in this process 304 * is now suspended at the user boundary, 305 * (or other safe places if we think of any). 306 */ 307 } 308 309 /* Allocate new proc. */ 310 newproc = uma_zalloc(proc_zone, M_WAITOK); 311 312 /* 313 * Although process entries are dynamically created, we still keep 314 * a global limit on the maximum number we will create. Don't allow 315 * a nonprivileged user to use the last ten processes; don't let root 316 * exceed the limit. The variable nprocs is the current number of 317 * processes, maxproc is the limit. 318 */ 319 sx_xlock(&allproc_lock); 320 uid = td->td_ucred->cr_ruid; 321 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 322 sx_xunlock(&allproc_lock); 323 uma_zfree(proc_zone, newproc); 324 if (p1->p_flag & P_KSES) { 325 PROC_LOCK(p1); 326 thread_single_end(); 327 PROC_UNLOCK(p1); 328 } 329 tsleep(&forksleep, PUSER, "fork", hz / 2); 330 return (EAGAIN); 331 } 332 /* 333 * Increment the count of procs running with this uid. Don't allow 334 * a nonprivileged user to exceed their current limit. 335 */ 336 PROC_LOCK(p1); 337 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 338 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 339 PROC_UNLOCK(p1); 340 if (!ok) { 341 sx_xunlock(&allproc_lock); 342 uma_zfree(proc_zone, newproc); 343 if (p1->p_flag & P_KSES) { 344 PROC_LOCK(p1); 345 thread_single_end(); 346 PROC_UNLOCK(p1); 347 } 348 tsleep(&forksleep, PUSER, "fork", hz / 2); 349 return (EAGAIN); 350 } 351 352 /* 353 * Increment the nprocs resource before blocking can occur. There 354 * are hard-limits as to the number of processes that can run. 355 */ 356 nprocs++; 357 358 /* 359 * Find an unused process ID. We remember a range of unused IDs 360 * ready to use (from lastpid+1 through pidchecked-1). 361 * 362 * If RFHIGHPID is set (used during system boot), do not allocate 363 * low-numbered pids. 364 */ 365 trypid = lastpid + 1; 366 if (flags & RFHIGHPID) { 367 if (trypid < 10) { 368 trypid = 10; 369 } 370 } else { 371 if (randompid) 372 trypid += arc4random() % randompid; 373 } 374 retry: 375 /* 376 * If the process ID prototype has wrapped around, 377 * restart somewhat above 0, as the low-numbered procs 378 * tend to include daemons that don't exit. 379 */ 380 if (trypid >= PID_MAX) { 381 trypid = trypid % PID_MAX; 382 if (trypid < 100) 383 trypid += 100; 384 pidchecked = 0; 385 } 386 if (trypid >= pidchecked) { 387 int doingzomb = 0; 388 389 pidchecked = PID_MAX; 390 /* 391 * Scan the active and zombie procs to check whether this pid 392 * is in use. Remember the lowest pid that's greater 393 * than trypid, so we can avoid checking for a while. 394 */ 395 p2 = LIST_FIRST(&allproc); 396 again: 397 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 398 PROC_LOCK(p2); 399 while (p2->p_pid == trypid || 400 p2->p_pgrp->pg_id == trypid || 401 p2->p_session->s_sid == trypid) { 402 trypid++; 403 if (trypid >= pidchecked) { 404 PROC_UNLOCK(p2); 405 goto retry; 406 } 407 } 408 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 409 pidchecked = p2->p_pid; 410 if (p2->p_pgrp->pg_id > trypid && 411 pidchecked > p2->p_pgrp->pg_id) 412 pidchecked = p2->p_pgrp->pg_id; 413 if (p2->p_session->s_sid > trypid && 414 pidchecked > p2->p_session->s_sid) 415 pidchecked = p2->p_session->s_sid; 416 PROC_UNLOCK(p2); 417 } 418 if (!doingzomb) { 419 doingzomb = 1; 420 p2 = LIST_FIRST(&zombproc); 421 goto again; 422 } 423 } 424 425 /* 426 * RFHIGHPID does not mess with the lastpid counter during boot. 427 */ 428 if (flags & RFHIGHPID) 429 pidchecked = 0; 430 else 431 lastpid = trypid; 432 433 p2 = newproc; 434 p2->p_state = PRS_NEW; /* protect against others */ 435 p2->p_pid = trypid; 436 LIST_INSERT_HEAD(&allproc, p2, p_list); 437 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 438 sx_xunlock(&allproc_lock); 439 440 /* 441 * Malloc things while we don't hold any locks. 442 */ 443 if (flags & RFSIGSHARE) { 444 MALLOC(newsigacts, struct sigacts *, 445 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 446 newprocsig = NULL; 447 } else { 448 newsigacts = NULL; 449 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 450 M_SUBPROC, M_WAITOK); 451 } 452 453 /* 454 * Copy filedesc. 455 * XXX: This is busted. fd*() need to not take proc 456 * arguments or something. 457 */ 458 if (flags & RFCFDG) 459 fd = fdinit(td); 460 else if (flags & RFFDG) { 461 FILEDESC_LOCK(p1->p_fd); 462 fd = fdcopy(td); 463 FILEDESC_UNLOCK(p1->p_fd); 464 } else 465 fd = fdshare(p1); 466 467 /* 468 * Make a proc table entry for the new process. 469 * Start by zeroing the section of proc that is zero-initialized, 470 * then copy the section that is copied directly from the parent. 471 */ 472 td2 = FIRST_THREAD_IN_PROC(p2); 473 kg2 = FIRST_KSEGRP_IN_PROC(p2); 474 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 475 476 /* Allocate and switch to an alternate kstack if specified */ 477 if (pages != 0) 478 pmap_new_altkstack(td2, pages); 479 480 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 481 482 bzero(&p2->p_startzero, 483 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 484 bzero(&ke2->ke_startzero, 485 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 486 bzero(&td2->td_startzero, 487 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 488 bzero(&kg2->kg_startzero, 489 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 490 491 mtx_init(&p2->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK); 492 PROC_LOCK(p2); 493 PROC_LOCK(p1); 494 495 bcopy(&p1->p_startcopy, &p2->p_startcopy, 496 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 497 bcopy(&td->td_startcopy, &td2->td_startcopy, 498 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 499 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 500 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 501 #undef RANGEOF 502 503 /* Set up the thread as an active thread (as if runnable). */ 504 ke2->ke_state = KES_THREAD; 505 ke2->ke_thread = td2; 506 td2->td_kse = ke2; 507 td2->td_flags &= ~TDF_UNBOUND; /* For the rest of this syscall. */ 508 509 /* 510 * Duplicate sub-structures as needed. 511 * Increase reference counts on shared objects. 512 * The p_stats and p_sigacts substructs are set in vm_forkproc. 513 */ 514 p2->p_flag = 0; 515 mtx_lock_spin(&sched_lock); 516 p2->p_sflag = PS_INMEM; 517 if (p1->p_sflag & PS_PROFIL) 518 startprofclock(p2); 519 /* 520 * Allow the scheduler to adjust the priority of the child and 521 * parent while we hold the sched_lock. 522 */ 523 sched_fork(td->td_ksegrp, kg2); 524 525 mtx_unlock_spin(&sched_lock); 526 p2->p_ucred = crhold(td->td_ucred); 527 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 528 529 /* 530 * Setup linkage for kernel based threading 531 */ 532 if((flags & RFTHREAD) != 0) { 533 /* 534 * XXX: This assumes a leader is a parent or grandparent of 535 * all processes in a task. 536 */ 537 if (p1->p_leader != p1) 538 PROC_LOCK(p1->p_leader); 539 p2->p_peers = p1->p_peers; 540 p1->p_peers = p2; 541 p2->p_leader = p1->p_leader; 542 if (p1->p_leader != p1) 543 PROC_UNLOCK(p1->p_leader); 544 } else { 545 p2->p_peers = NULL; 546 p2->p_leader = p2; 547 } 548 549 pargs_hold(p2->p_args); 550 551 if (flags & RFSIGSHARE) { 552 p2->p_procsig = p1->p_procsig; 553 p2->p_procsig->ps_refcnt++; 554 if (p1->p_sigacts == &p1->p_uarea->u_sigacts) { 555 /* 556 * Set p_sigacts to the new shared structure. 557 * Note that this is updating p1->p_sigacts at the 558 * same time, since p_sigacts is just a pointer to 559 * the shared p_procsig->ps_sigacts. 560 */ 561 p2->p_sigacts = newsigacts; 562 newsigacts = NULL; 563 *p2->p_sigacts = p1->p_uarea->u_sigacts; 564 } 565 } else { 566 p2->p_procsig = newprocsig; 567 newprocsig = NULL; 568 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 569 p2->p_procsig->ps_refcnt = 1; 570 p2->p_sigacts = NULL; /* finished in vm_forkproc() */ 571 } 572 if (flags & RFLINUXTHPN) 573 p2->p_sigparent = SIGUSR1; 574 else 575 p2->p_sigparent = SIGCHLD; 576 577 /* Bump references to the text vnode (for procfs) */ 578 p2->p_textvp = p1->p_textvp; 579 if (p2->p_textvp) 580 VREF(p2->p_textvp); 581 p2->p_fd = fd; 582 PROC_UNLOCK(p1); 583 PROC_UNLOCK(p2); 584 585 /* 586 * If p_limit is still copy-on-write, bump refcnt, 587 * otherwise get a copy that won't be modified. 588 * (If PL_SHAREMOD is clear, the structure is shared 589 * copy-on-write.) 590 */ 591 if (p1->p_limit->p_lflags & PL_SHAREMOD) 592 p2->p_limit = limcopy(p1->p_limit); 593 else { 594 p2->p_limit = p1->p_limit; 595 p2->p_limit->p_refcnt++; 596 } 597 598 sx_xlock(&proctree_lock); 599 PGRP_LOCK(p1->p_pgrp); 600 PROC_LOCK(p2); 601 PROC_LOCK(p1); 602 603 /* 604 * Preserve some more flags in subprocess. PS_PROFIL has already 605 * been preserved. 606 */ 607 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 608 SESS_LOCK(p1->p_session); 609 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 610 p2->p_flag |= P_CONTROLT; 611 SESS_UNLOCK(p1->p_session); 612 if (flags & RFPPWAIT) 613 p2->p_flag |= P_PPWAIT; 614 615 LIST_INSERT_AFTER(p1, p2, p_pglist); 616 PGRP_UNLOCK(p1->p_pgrp); 617 LIST_INIT(&p2->p_children); 618 619 callout_init(&p2->p_itcallout, 0); 620 621 #ifdef KTRACE 622 /* 623 * Copy traceflag and tracefile if enabled. 624 */ 625 mtx_lock(&ktrace_mtx); 626 KASSERT(p2->p_tracep == NULL, ("new process has a ktrace vnode")); 627 if (p1->p_traceflag & KTRFAC_INHERIT) { 628 p2->p_traceflag = p1->p_traceflag; 629 if ((p2->p_tracep = p1->p_tracep) != NULL) 630 VREF(p2->p_tracep); 631 } 632 mtx_unlock(&ktrace_mtx); 633 #endif 634 635 /* 636 * If PF_FORK is set, the child process inherits the 637 * procfs ioctl flags from its parent. 638 */ 639 if (p1->p_pfsflags & PF_FORK) { 640 p2->p_stops = p1->p_stops; 641 p2->p_pfsflags = p1->p_pfsflags; 642 } 643 644 /* 645 * This begins the section where we must prevent the parent 646 * from being swapped. 647 */ 648 _PHOLD(p1); 649 PROC_UNLOCK(p1); 650 651 /* 652 * Attach the new process to its parent. 653 * 654 * If RFNOWAIT is set, the newly created process becomes a child 655 * of init. This effectively disassociates the child from the 656 * parent. 657 */ 658 if (flags & RFNOWAIT) 659 pptr = initproc; 660 else 661 pptr = p1; 662 p2->p_pptr = pptr; 663 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 664 PROC_UNLOCK(p2); 665 sx_xunlock(&proctree_lock); 666 667 KASSERT(newprocsig == NULL, ("unused newprocsig")); 668 if (newsigacts != NULL) 669 FREE(newsigacts, M_SUBPROC); 670 /* 671 * Finish creating the child process. It will return via a different 672 * execution path later. (ie: directly into user mode) 673 */ 674 vm_forkproc(td, p2, td2, flags); 675 676 if (flags == (RFFDG | RFPROC)) { 677 cnt.v_forks++; 678 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 679 p2->p_vmspace->vm_ssize; 680 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 681 cnt.v_vforks++; 682 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 683 p2->p_vmspace->vm_ssize; 684 } else if (p1 == &proc0) { 685 cnt.v_kthreads++; 686 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 687 p2->p_vmspace->vm_ssize; 688 } else { 689 cnt.v_rforks++; 690 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 691 p2->p_vmspace->vm_ssize; 692 } 693 694 /* 695 * Both processes are set up, now check if any loadable modules want 696 * to adjust anything. 697 * What if they have an error? XXX 698 */ 699 sx_slock(&fork_list_lock); 700 TAILQ_FOREACH(ep, &fork_list, next) { 701 (*ep->function)(p1, p2, flags); 702 } 703 sx_sunlock(&fork_list_lock); 704 705 /* 706 * If RFSTOPPED not requested, make child runnable and add to 707 * run queue. 708 */ 709 microtime(&(p2->p_stats->p_start)); 710 p2->p_acflag = AFORK; 711 if ((flags & RFSTOPPED) == 0) { 712 mtx_lock_spin(&sched_lock); 713 p2->p_state = PRS_NORMAL; 714 TD_SET_CAN_RUN(td2); 715 setrunqueue(td2); 716 mtx_unlock_spin(&sched_lock); 717 } 718 719 /* 720 * Now can be swapped. 721 */ 722 PROC_LOCK(p1); 723 _PRELE(p1); 724 725 /* 726 * tell any interested parties about the new process 727 */ 728 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 729 PROC_UNLOCK(p1); 730 731 /* 732 * Preserve synchronization semantics of vfork. If waiting for 733 * child to exec or exit, set P_PPWAIT on child, and sleep on our 734 * proc (in case of exit). 735 */ 736 PROC_LOCK(p2); 737 while (p2->p_flag & P_PPWAIT) 738 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 739 PROC_UNLOCK(p2); 740 741 /* 742 * If other threads are waiting, let them continue now 743 */ 744 if (p1->p_flag & P_KSES) { 745 PROC_LOCK(p1); 746 thread_single_end(); 747 PROC_UNLOCK(p1); 748 } 749 750 /* 751 * Return child proc pointer to parent. 752 */ 753 *procp = p2; 754 return (0); 755 } 756 757 /* 758 * The next two functionms are general routines to handle adding/deleting 759 * items on the fork callout list. 760 * 761 * at_fork(): 762 * Take the arguments given and put them onto the fork callout list, 763 * However first make sure that it's not already there. 764 * Returns 0 on success or a standard error number. 765 */ 766 767 int 768 at_fork(function) 769 forklist_fn function; 770 { 771 struct forklist *ep; 772 773 #ifdef INVARIANTS 774 /* let the programmer know if he's been stupid */ 775 if (rm_at_fork(function)) 776 printf("WARNING: fork callout entry (%p) already present\n", 777 function); 778 #endif 779 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT); 780 if (ep == NULL) 781 return (ENOMEM); 782 ep->function = function; 783 sx_xlock(&fork_list_lock); 784 TAILQ_INSERT_TAIL(&fork_list, ep, next); 785 sx_xunlock(&fork_list_lock); 786 return (0); 787 } 788 789 /* 790 * Scan the exit callout list for the given item and remove it.. 791 * Returns the number of items removed (0 or 1) 792 */ 793 794 int 795 rm_at_fork(function) 796 forklist_fn function; 797 { 798 struct forklist *ep; 799 800 sx_xlock(&fork_list_lock); 801 TAILQ_FOREACH(ep, &fork_list, next) { 802 if (ep->function == function) { 803 TAILQ_REMOVE(&fork_list, ep, next); 804 sx_xunlock(&fork_list_lock); 805 free(ep, M_ATFORK); 806 return(1); 807 } 808 } 809 sx_xunlock(&fork_list_lock); 810 return (0); 811 } 812 813 /* 814 * Handle the return of a child process from fork1(). This function 815 * is called from the MD fork_trampoline() entry point. 816 */ 817 void 818 fork_exit(callout, arg, frame) 819 void (*callout)(void *, struct trapframe *); 820 void *arg; 821 struct trapframe *frame; 822 { 823 struct thread *td = curthread; 824 struct proc *p = td->td_proc; 825 826 td->td_kse->ke_oncpu = PCPU_GET(cpuid); 827 p->p_state = PRS_NORMAL; 828 /* 829 * Finish setting up thread glue. We need to initialize 830 * the thread into a td_critnest=1 state. Some platforms 831 * may have already partially or fully initialized td_critnest 832 * and/or td_md.md_savecrit (when applciable). 833 * 834 * see <arch>/<arch>/critical.c 835 */ 836 sched_lock.mtx_lock = (uintptr_t)td; 837 sched_lock.mtx_recurse = 0; 838 cpu_critical_fork_exit(); 839 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 840 p->p_comm); 841 if (PCPU_GET(switchtime.sec) == 0) 842 binuptime(PCPU_PTR(switchtime)); 843 PCPU_SET(switchticks, ticks); 844 mtx_unlock_spin(&sched_lock); 845 846 /* 847 * cpu_set_fork_handler intercepts this function call to 848 * have this call a non-return function to stay in kernel mode. 849 * initproc has its own fork handler, but it does return. 850 */ 851 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 852 callout(arg, frame); 853 854 /* 855 * Check if a kernel thread misbehaved and returned from its main 856 * function. 857 */ 858 PROC_LOCK(p); 859 if (p->p_flag & P_KTHREAD) { 860 PROC_UNLOCK(p); 861 mtx_lock(&Giant); 862 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 863 p->p_comm, p->p_pid); 864 kthread_exit(0); 865 } 866 PROC_UNLOCK(p); 867 #ifdef DIAGNOSTIC 868 cred_free_thread(td); 869 #endif 870 mtx_assert(&Giant, MA_NOTOWNED); 871 } 872 873 /* 874 * Simplified back end of syscall(), used when returning from fork() 875 * directly into user mode. Giant is not held on entry, and must not 876 * be held on return. This function is passed in to fork_exit() as the 877 * first parameter and is called when returning to a new userland process. 878 */ 879 void 880 fork_return(td, frame) 881 struct thread *td; 882 struct trapframe *frame; 883 { 884 885 userret(td, frame, 0); 886 #ifdef KTRACE 887 if (KTRPOINT(td, KTR_SYSRET)) 888 ktrsysret(SYS_fork, 0, 0); 889 #endif 890 mtx_assert(&Giant, MA_NOTOWNED); 891 } 892