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: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $ 40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.60 2006/12/23 00:35:04 swildner Exp $ 41 */ 42 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/filedesc.h> 49 #include <sys/kernel.h> 50 #include <sys/sysctl.h> 51 #include <sys/malloc.h> 52 #include <sys/proc.h> 53 #include <sys/resourcevar.h> 54 #include <sys/vnode.h> 55 #include <sys/acct.h> 56 #include <sys/ktrace.h> 57 #include <sys/unistd.h> 58 #include <sys/jail.h> 59 #include <sys/caps.h> 60 61 #include <vm/vm.h> 62 #include <sys/lock.h> 63 #include <vm/pmap.h> 64 #include <vm/vm_map.h> 65 #include <vm/vm_extern.h> 66 #include <vm/vm_zone.h> 67 68 #include <sys/vmmeter.h> 69 #include <sys/user.h> 70 #include <sys/thread2.h> 71 72 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 73 74 /* 75 * These are the stuctures used to create a callout list for things to do 76 * when forking a process 77 */ 78 struct forklist { 79 forklist_fn function; 80 TAILQ_ENTRY(forklist) next; 81 }; 82 83 TAILQ_HEAD(forklist_head, forklist); 84 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 85 86 int forksleep; /* Place for fork1() to sleep on. */ 87 88 /* ARGSUSED */ 89 int 90 sys_fork(struct fork_args *uap) 91 { 92 struct lwp *lp = curthread->td_lwp; 93 struct proc *p2; 94 int error; 95 96 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2); 97 if (error == 0) { 98 start_forked_proc(lp, p2); 99 uap->sysmsg_fds[0] = p2->p_pid; 100 uap->sysmsg_fds[1] = 0; 101 } 102 return error; 103 } 104 105 /* ARGSUSED */ 106 int 107 sys_vfork(struct vfork_args *uap) 108 { 109 struct lwp *lp = curthread->td_lwp; 110 struct proc *p2; 111 int error; 112 113 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2); 114 if (error == 0) { 115 start_forked_proc(lp, p2); 116 uap->sysmsg_fds[0] = p2->p_pid; 117 uap->sysmsg_fds[1] = 0; 118 } 119 return error; 120 } 121 122 /* 123 * Handle rforks. An rfork may (1) operate on the current process without 124 * creating a new, (2) create a new process that shared the current process's 125 * vmspace, signals, and/or descriptors, or (3) create a new process that does 126 * not share these things (normal fork). 127 * 128 * Note that we only call start_forked_proc() if a new process is actually 129 * created. 130 * 131 * rfork { int flags } 132 */ 133 int 134 sys_rfork(struct rfork_args *uap) 135 { 136 struct lwp *lp = curthread->td_lwp; 137 struct proc *p2; 138 int error; 139 140 if ((uap->flags & RFKERNELONLY) != 0) 141 return (EINVAL); 142 143 error = fork1(lp, uap->flags | RFPGLOCK, &p2); 144 if (error == 0) { 145 if (p2) 146 start_forked_proc(lp, p2); 147 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0; 148 uap->sysmsg_fds[1] = 0; 149 } 150 return error; 151 } 152 153 154 int nprocs = 1; /* process 0 */ 155 156 int 157 fork1(struct lwp *lp1, int flags, struct proc **procp) 158 { 159 struct proc *p1 = lp1->lwp_proc; 160 struct proc *p2, *pptr; 161 struct pgrp *pgrp; 162 struct lwp *lp2; 163 uid_t uid; 164 int ok, error; 165 static int curfail = 0; 166 static struct timeval lastfail; 167 struct forklist *ep; 168 struct filedesc_to_leader *fdtol; 169 170 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 171 return (EINVAL); 172 173 /* 174 * Here we don't create a new process, but we divorce 175 * certain parts of a process from itself. 176 */ 177 if ((flags & RFPROC) == 0) { 178 179 vm_fork(lp1, 0, flags); 180 181 /* 182 * Close all file descriptors. 183 */ 184 if (flags & RFCFDG) { 185 struct filedesc *fdtmp; 186 fdtmp = fdinit(p1); 187 fdfree(p1); 188 p1->p_fd = fdtmp; 189 } 190 191 /* 192 * Unshare file descriptors (from parent.) 193 */ 194 if (flags & RFFDG) { 195 if (p1->p_fd->fd_refcnt > 1) { 196 struct filedesc *newfd; 197 newfd = fdcopy(p1); 198 fdfree(p1); 199 p1->p_fd = newfd; 200 } 201 } 202 *procp = NULL; 203 return (0); 204 } 205 206 /* 207 * Interlock against process group signal delivery. If signals 208 * are pending after the interlock is obtained we have to restart 209 * the system call to process the signals. If we don't the child 210 * can miss a pgsignal (such as ^C) sent during the fork. 211 * 212 * We can't use CURSIG() here because it will process any STOPs 213 * and cause the process group lock to be held indefinitely. If 214 * a STOP occurs, the fork will be restarted after the CONT. 215 */ 216 error = 0; 217 pgrp = NULL; 218 if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) { 219 lockmgr(&pgrp->pg_lock, LK_SHARED); 220 if (CURSIGNB(p1)) { 221 error = ERESTART; 222 goto done; 223 } 224 } 225 226 /* 227 * Although process entries are dynamically created, we still keep 228 * a global limit on the maximum number we will create. Don't allow 229 * a nonprivileged user to use the last ten processes; don't let root 230 * exceed the limit. The variable nprocs is the current number of 231 * processes, maxproc is the limit. 232 */ 233 uid = p1->p_ucred->cr_ruid; 234 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 235 if (ppsratecheck(&lastfail, &curfail, 1)) 236 kprintf("maxproc limit exceeded by uid %d, please " 237 "see tuning(7) and login.conf(5).\n", uid); 238 tsleep(&forksleep, 0, "fork", hz / 2); 239 error = EAGAIN; 240 goto done; 241 } 242 /* 243 * Increment the nprocs resource before blocking can occur. There 244 * are hard-limits as to the number of processes that can run. 245 */ 246 nprocs++; 247 248 /* 249 * Increment the count of procs running with this uid. Don't allow 250 * a nonprivileged user to exceed their current limit. 251 */ 252 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1, 253 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 254 if (!ok) { 255 /* 256 * Back out the process count 257 */ 258 nprocs--; 259 if (ppsratecheck(&lastfail, &curfail, 1)) 260 kprintf("maxproc limit exceeded by uid %d, please " 261 "see tuning(7) and login.conf(5).\n", uid); 262 tsleep(&forksleep, 0, "fork", hz / 2); 263 error = EAGAIN; 264 goto done; 265 } 266 267 /* Allocate new proc. */ 268 p2 = zalloc(proc_zone); 269 270 /* 271 * Setup linkage for kernel based threading XXX lwp 272 */ 273 if (flags & RFTHREAD) { 274 p2->p_peers = p1->p_peers; 275 p1->p_peers = p2; 276 p2->p_leader = p1->p_leader; 277 } else { 278 p2->p_peers = NULL; 279 p2->p_leader = p2; 280 } 281 282 p2->p_wakeup = 0; 283 p2->p_vmspace = NULL; 284 p2->p_numposixlocks = 0; 285 p2->p_emuldata = NULL; 286 LIST_INIT(&p2->p_lwps); 287 288 /* XXX lwp */ 289 lp2 = &p2->p_lwp; 290 lp2->lwp_proc = p2; 291 lp2->lwp_tid = 0; 292 LIST_INSERT_HEAD(&p2->p_lwps, lp2, lwp_list); 293 p2->p_nthreads = 1; 294 p2->p_nstopped = 0; 295 p2->p_lasttid = 0; 296 297 /* 298 * Setting the state to SIDL protects the partially initialized 299 * process once it starts getting hooked into the rest of the system. 300 */ 301 p2->p_stat = SIDL; 302 proc_add_allproc(p2); 303 304 /* 305 * Make a proc table entry for the new process. 306 * Start by zeroing the section of proc that is zero-initialized, 307 * then copy the section that is copied directly from the parent. 308 */ 309 bzero(&p2->p_startzero, 310 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 311 bzero(&lp2->lwp_startzero, 312 (unsigned) ((caddr_t)&lp2->lwp_endzero - 313 (caddr_t)&lp2->lwp_startzero)); 314 bcopy(&p1->p_startcopy, &p2->p_startcopy, 315 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 316 bcopy(&p1->p_lwp.lwp_startcopy, &lp2->lwp_startcopy, 317 (unsigned) ((caddr_t)&lp2->lwp_endcopy - 318 (caddr_t)&lp2->lwp_startcopy)); 319 320 p2->p_aioinfo = NULL; 321 322 /* 323 * Duplicate sub-structures as needed. 324 * Increase reference counts on shared objects. 325 * The p_stats and p_sigacts substructs are set in vm_fork. 326 * p_lock is in the copy area and must be cleared. 327 */ 328 p2->p_flag = 0; 329 p2->p_lock = 0; 330 if (p1->p_flag & P_PROFIL) 331 startprofclock(p2); 332 p2->p_ucred = crhold(p1->p_ucred); 333 334 if (jailed(p2->p_ucred)) 335 p2->p_flag |= P_JAILED; 336 337 if (p2->p_args) 338 p2->p_args->ar_ref++; 339 340 if (flags & RFSIGSHARE) { 341 p2->p_procsig = p1->p_procsig; 342 p2->p_procsig->ps_refcnt++; 343 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 344 struct sigacts *newsigacts; 345 346 /* Create the shared sigacts structure */ 347 MALLOC(newsigacts, struct sigacts *, 348 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 349 crit_enter(); 350 /* 351 * Set p_sigacts to the new shared structure. 352 * Note that this is updating p1->p_sigacts at the 353 * same time, since p_sigacts is just a pointer to 354 * the shared p_procsig->ps_sigacts. 355 */ 356 p2->p_sigacts = newsigacts; 357 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 358 sizeof(*p2->p_sigacts)); 359 *p2->p_sigacts = p1->p_addr->u_sigacts; 360 crit_exit(); 361 } 362 } else { 363 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 364 M_SUBPROC, M_WAITOK); 365 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 366 p2->p_procsig->ps_refcnt = 1; 367 p2->p_sigacts = NULL; /* finished in vm_fork() */ 368 } 369 if (flags & RFLINUXTHPN) 370 p2->p_sigparent = SIGUSR1; 371 else 372 p2->p_sigparent = SIGCHLD; 373 374 /* bump references to the text vnode (for procfs) */ 375 p2->p_textvp = p1->p_textvp; 376 if (p2->p_textvp) 377 vref(p2->p_textvp); 378 379 /* 380 * Handle file descriptors 381 */ 382 if (flags & RFCFDG) { 383 p2->p_fd = fdinit(p1); 384 fdtol = NULL; 385 } else if (flags & RFFDG) { 386 p2->p_fd = fdcopy(p1); 387 fdtol = NULL; 388 } else { 389 p2->p_fd = fdshare(p1); 390 if (p1->p_fdtol == NULL) 391 p1->p_fdtol = 392 filedesc_to_leader_alloc(NULL, 393 p1->p_leader); 394 if ((flags & RFTHREAD) != 0) { 395 /* 396 * Shared file descriptor table and 397 * shared process leaders. 398 */ 399 fdtol = p1->p_fdtol; 400 fdtol->fdl_refcount++; 401 } else { 402 /* 403 * Shared file descriptor table, and 404 * different process leaders 405 */ 406 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 407 } 408 } 409 p2->p_fdtol = fdtol; 410 p2->p_limit = plimit_fork(p1->p_limit); 411 412 /* 413 * Preserve some more flags in subprocess. P_PROFIL has already 414 * been preserved. 415 */ 416 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 417 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 418 p2->p_flag |= P_CONTROLT; 419 if (flags & RFPPWAIT) 420 p2->p_flag |= P_PPWAIT; 421 422 /* 423 * Inherit the virtual kernel structure (allows a virtual kernel 424 * to fork to simulate multiple cpus). 425 */ 426 p2->p_vkernel = NULL; 427 if (p1->p_vkernel) 428 vkernel_inherit(p1, p2); 429 430 /* 431 * Once we are on a pglist we may receive signals. XXX we might 432 * race a ^C being sent to the process group by not receiving it 433 * at all prior to this line. 434 */ 435 LIST_INSERT_AFTER(p1, p2, p_pglist); 436 437 /* 438 * Attach the new process to its parent. 439 * 440 * If RFNOWAIT is set, the newly created process becomes a child 441 * of init. This effectively disassociates the child from the 442 * parent. 443 */ 444 if (flags & RFNOWAIT) 445 pptr = initproc; 446 else 447 pptr = p1; 448 p2->p_pptr = pptr; 449 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 450 LIST_INIT(&p2->p_children); 451 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 452 callout_init(&p2->p_ithandle); 453 454 #ifdef KTRACE 455 /* 456 * Copy traceflag and tracefile if enabled. If not inherited, 457 * these were zeroed above but we still could have a trace race 458 * so make sure p2's p_tracenode is NULL. 459 */ 460 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) { 461 p2->p_traceflag = p1->p_traceflag; 462 p2->p_tracenode = ktrinherit(p1->p_tracenode); 463 } 464 #endif 465 466 /* 467 * Inherit the scheduler and initialize scheduler-related fields. 468 * Set cpbase to the last timeout that occured (not the upcoming 469 * timeout). 470 * 471 * A critical section is required since a timer IPI can update 472 * scheduler specific data. 473 */ 474 crit_enter(); 475 p2->p_usched = p1->p_usched; 476 lp2->lwp_cpbase = mycpu->gd_schedclock.time - 477 mycpu->gd_schedclock.periodic; 478 p2->p_usched->heuristic_forking(&p1->p_lwp, lp2); 479 crit_exit(); 480 481 /* 482 * This begins the section where we must prevent the parent 483 * from being swapped. 484 */ 485 PHOLD(p1); 486 487 /* 488 * Finish creating the child process. It will return via a different 489 * execution path later. (ie: directly into user mode) 490 */ 491 vm_fork(lp1, p2, flags); 492 caps_fork(p1, p2, flags); 493 494 if (flags == (RFFDG | RFPROC)) { 495 mycpu->gd_cnt.v_forks++; 496 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 497 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 498 mycpu->gd_cnt.v_vforks++; 499 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 500 } else if (p1 == &proc0) { 501 mycpu->gd_cnt.v_kthreads++; 502 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 503 } else { 504 mycpu->gd_cnt.v_rforks++; 505 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 506 } 507 508 /* 509 * Both processes are set up, now check if any loadable modules want 510 * to adjust anything. 511 * What if they have an error? XXX 512 */ 513 TAILQ_FOREACH(ep, &fork_list, next) { 514 (*ep->function)(p1, p2, flags); 515 } 516 517 /* 518 * Set the start time. Note that the process is not runnable. The 519 * caller is responsible for making it runnable. 520 */ 521 microtime(&p2->p_start); 522 p2->p_acflag = AFORK; 523 524 /* 525 * tell any interested parties about the new process 526 */ 527 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 528 529 /* 530 * Return child proc pointer to parent. 531 */ 532 *procp = p2; 533 done: 534 if (pgrp) 535 lockmgr(&pgrp->pg_lock, LK_RELEASE); 536 return (error); 537 } 538 539 /* 540 * The next two functionms are general routines to handle adding/deleting 541 * items on the fork callout list. 542 * 543 * at_fork(): 544 * Take the arguments given and put them onto the fork callout list, 545 * However first make sure that it's not already there. 546 * Returns 0 on success or a standard error number. 547 */ 548 int 549 at_fork(forklist_fn function) 550 { 551 struct forklist *ep; 552 553 #ifdef INVARIANTS 554 /* let the programmer know if he's been stupid */ 555 if (rm_at_fork(function)) { 556 kprintf("WARNING: fork callout entry (%p) already present\n", 557 function); 558 } 559 #endif 560 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 561 ep->function = function; 562 TAILQ_INSERT_TAIL(&fork_list, ep, next); 563 return (0); 564 } 565 566 /* 567 * Scan the exit callout list for the given item and remove it.. 568 * Returns the number of items removed (0 or 1) 569 */ 570 int 571 rm_at_fork(forklist_fn function) 572 { 573 struct forklist *ep; 574 575 TAILQ_FOREACH(ep, &fork_list, next) { 576 if (ep->function == function) { 577 TAILQ_REMOVE(&fork_list, ep, next); 578 kfree(ep, M_ATFORK); 579 return(1); 580 } 581 } 582 return (0); 583 } 584 585 /* 586 * Add a forked process to the run queue after any remaining setup, such 587 * as setting the fork handler, has been completed. 588 */ 589 void 590 start_forked_proc(struct lwp *lp1, struct proc *p2) 591 { 592 struct lwp *lp2; 593 594 KKASSERT(p2 != NULL && p2->p_nthreads == 1); 595 596 lp2 = LIST_FIRST(&p2->p_lwps); 597 598 /* 599 * Move from SIDL to RUN queue, and activate the process's thread. 600 * Activation of the thread effectively makes the process "a" 601 * current process, so we do not setrunqueue(). 602 * 603 * YYY setrunqueue works here but we should clean up the trampoline 604 * code so we just schedule the LWKT thread and let the trampoline 605 * deal with the userland scheduler on return to userland. 606 */ 607 KASSERT(p2->p_stat == SIDL, 608 ("cannot start forked process, bad status: %p", p2)); 609 p2->p_usched->resetpriority(lp2); 610 crit_enter(); 611 p2->p_stat = SRUN; 612 p2->p_usched->setrunqueue(lp2); 613 crit_exit(); 614 615 /* 616 * Now can be swapped. 617 */ 618 PRELE(lp1->lwp_proc); 619 620 /* 621 * Preserve synchronization semantics of vfork. If waiting for 622 * child to exec or exit, set P_PPWAIT on child, and sleep on our 623 * proc (in case of exit). 624 */ 625 while (p2->p_flag & P_PPWAIT) 626 tsleep(lp1->lwp_proc, 0, "ppwait", 0); 627 } 628