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 1.36 2005/06/27 18:37:57 dillon 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 fork(struct fork_args *uap) 91 { 92 struct proc *p = curproc; 93 struct proc *p2; 94 int error; 95 96 error = fork1(p, RFFDG | RFPROC, &p2); 97 if (error == 0) { 98 start_forked_proc(p, 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 vfork(struct vfork_args *uap) 108 { 109 struct proc *p = curproc; 110 struct proc *p2; 111 int error; 112 113 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2); 114 if (error == 0) { 115 start_forked_proc(p, 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 rfork(struct rfork_args *uap) 135 { 136 struct proc *p = curproc; 137 struct proc *p2; 138 int error; 139 140 if ((uap->flags & RFKERNELONLY) != 0) 141 return (EINVAL); 142 143 error = fork1(p, uap->flags, &p2); 144 if (error == 0) { 145 if (p2) 146 start_forked_proc(p, 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 static int nextpid = 0; 156 157 /* 158 * Random component to nextpid generation. We mix in a random factor to make 159 * it a little harder to predict. We sanity check the modulus value to avoid 160 * doing it in critical paths. Don't let it be too small or we pointlessly 161 * waste randomness entropy, and don't let it be impossibly large. Using a 162 * modulus that is too big causes a LOT more process table scans and slows 163 * down fork processing as the pidchecked caching is defeated. 164 */ 165 static int randompid = 0; 166 167 static int 168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 169 { 170 int error, pid; 171 172 pid = randompid; 173 error = sysctl_handle_int(oidp, &pid, 0, req); 174 if (error || !req->newptr) 175 return (error); 176 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 177 pid = PID_MAX - 100; 178 else if (pid < 2) /* NOP */ 179 pid = 0; 180 else if (pid < 100) /* Make it reasonable */ 181 pid = 100; 182 randompid = pid; 183 return (error); 184 } 185 186 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 187 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 188 189 int 190 fork1(struct proc *p1, int flags, struct proc **procp) 191 { 192 struct proc *p2, *pptr; 193 uid_t uid; 194 struct proc *newproc; 195 int ok; 196 static int curfail = 0, pidchecked = 0; 197 static struct timeval lastfail; 198 struct forklist *ep; 199 struct filedesc_to_leader *fdtol; 200 201 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 202 return (EINVAL); 203 204 /* 205 * Here we don't create a new process, but we divorce 206 * certain parts of a process from itself. 207 */ 208 if ((flags & RFPROC) == 0) { 209 210 vm_fork(p1, 0, flags); 211 212 /* 213 * Close all file descriptors. 214 */ 215 if (flags & RFCFDG) { 216 struct filedesc *fdtmp; 217 fdtmp = fdinit(p1); 218 fdfree(p1); 219 p1->p_fd = fdtmp; 220 } 221 222 /* 223 * Unshare file descriptors (from parent.) 224 */ 225 if (flags & RFFDG) { 226 if (p1->p_fd->fd_refcnt > 1) { 227 struct filedesc *newfd; 228 newfd = fdcopy(p1); 229 fdfree(p1); 230 p1->p_fd = newfd; 231 } 232 } 233 *procp = NULL; 234 return (0); 235 } 236 237 /* 238 * Although process entries are dynamically created, we still keep 239 * a global limit on the maximum number we will create. Don't allow 240 * a nonprivileged user to use the last ten processes; don't let root 241 * exceed the limit. The variable nprocs is the current number of 242 * processes, maxproc is the limit. 243 */ 244 uid = p1->p_ucred->cr_ruid; 245 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 246 if (ppsratecheck(&lastfail, &curfail, 1)) 247 printf("maxproc limit exceeded by uid %d, please " 248 "see tuning(7) and login.conf(5).\n", uid); 249 tsleep(&forksleep, 0, "fork", hz / 2); 250 return (EAGAIN); 251 } 252 /* 253 * Increment the nprocs resource before blocking can occur. There 254 * are hard-limits as to the number of processes that can run. 255 */ 256 nprocs++; 257 258 /* 259 * Increment the count of procs running with this uid. Don't allow 260 * a nonprivileged user to exceed their current limit. 261 */ 262 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1, 263 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 264 if (!ok) { 265 /* 266 * Back out the process count 267 */ 268 nprocs--; 269 if (ppsratecheck(&lastfail, &curfail, 1)) 270 printf("maxproc limit exceeded by uid %d, please " 271 "see tuning(7) and login.conf(5).\n", uid); 272 tsleep(&forksleep, 0, "fork", hz / 2); 273 return (EAGAIN); 274 } 275 276 /* Allocate new proc. */ 277 newproc = zalloc(proc_zone); 278 279 /* 280 * Setup linkage for kernel based threading 281 */ 282 if ((flags & RFTHREAD) != 0) { 283 newproc->p_peers = p1->p_peers; 284 p1->p_peers = newproc; 285 newproc->p_leader = p1->p_leader; 286 } else { 287 newproc->p_peers = 0; 288 newproc->p_leader = newproc; 289 } 290 291 newproc->p_wakeup = 0; 292 newproc->p_vmspace = NULL; 293 TAILQ_INIT(&newproc->p_sysmsgq); 294 295 /* 296 * Find an unused process ID. We remember a range of unused IDs 297 * ready to use (from nextpid+1 through pidchecked-1). 298 */ 299 nextpid++; 300 if (randompid) 301 nextpid += arc4random() % randompid; 302 retry: 303 /* 304 * If the process ID prototype has wrapped around, 305 * restart somewhat above 0, as the low-numbered procs 306 * tend to include daemons that don't exit. 307 */ 308 if (nextpid >= PID_MAX) { 309 nextpid = nextpid % PID_MAX; 310 if (nextpid < 100) 311 nextpid += 100; 312 pidchecked = 0; 313 } 314 if (nextpid >= pidchecked) { 315 int doingzomb = 0; 316 317 pidchecked = PID_MAX; 318 /* 319 * Scan the active and zombie procs to check whether this pid 320 * is in use. Remember the lowest pid that's greater 321 * than nextpid, so we can avoid checking for a while. 322 */ 323 p2 = LIST_FIRST(&allproc); 324 again: 325 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 326 while (p2->p_pid == nextpid || 327 p2->p_pgrp->pg_id == nextpid || 328 p2->p_session->s_sid == nextpid) { 329 nextpid++; 330 if (nextpid >= pidchecked) 331 goto retry; 332 } 333 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 334 pidchecked = p2->p_pid; 335 if (p2->p_pgrp->pg_id > nextpid && 336 pidchecked > p2->p_pgrp->pg_id) 337 pidchecked = p2->p_pgrp->pg_id; 338 if (p2->p_session->s_sid > nextpid && 339 pidchecked > p2->p_session->s_sid) 340 pidchecked = p2->p_session->s_sid; 341 } 342 if (!doingzomb) { 343 doingzomb = 1; 344 p2 = LIST_FIRST(&zombproc); 345 goto again; 346 } 347 } 348 349 p2 = newproc; 350 p2->p_stat = SIDL; /* protect against others */ 351 p2->p_pid = nextpid; 352 LIST_INSERT_HEAD(&allproc, p2, p_list); 353 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 354 355 /* 356 * Make a proc table entry for the new process. 357 * Start by zeroing the section of proc that is zero-initialized, 358 * then copy the section that is copied directly from the parent. 359 */ 360 bzero(&p2->p_startzero, 361 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 362 bcopy(&p1->p_startcopy, &p2->p_startcopy, 363 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 364 365 p2->p_aioinfo = NULL; 366 367 /* 368 * Duplicate sub-structures as needed. 369 * Increase reference counts on shared objects. 370 * The p_stats and p_sigacts substructs are set in vm_fork. 371 */ 372 p2->p_flag = P_INMEM; 373 if (p1->p_flag & P_PROFIL) 374 startprofclock(p2); 375 p2->p_ucred = crhold(p1->p_ucred); 376 377 if (jailed(p2->p_ucred)) 378 p2->p_flag |= P_JAILED; 379 380 if (p2->p_args) 381 p2->p_args->ar_ref++; 382 383 if (flags & RFSIGSHARE) { 384 p2->p_procsig = p1->p_procsig; 385 p2->p_procsig->ps_refcnt++; 386 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 387 struct sigacts *newsigacts; 388 389 /* Create the shared sigacts structure */ 390 MALLOC(newsigacts, struct sigacts *, 391 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 392 crit_enter(); 393 /* 394 * Set p_sigacts to the new shared structure. 395 * Note that this is updating p1->p_sigacts at the 396 * same time, since p_sigacts is just a pointer to 397 * the shared p_procsig->ps_sigacts. 398 */ 399 p2->p_sigacts = newsigacts; 400 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 401 sizeof(*p2->p_sigacts)); 402 *p2->p_sigacts = p1->p_addr->u_sigacts; 403 crit_exit(); 404 } 405 } else { 406 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 407 M_SUBPROC, M_WAITOK); 408 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 409 p2->p_procsig->ps_refcnt = 1; 410 p2->p_sigacts = NULL; /* finished in vm_fork() */ 411 } 412 if (flags & RFLINUXTHPN) 413 p2->p_sigparent = SIGUSR1; 414 else 415 p2->p_sigparent = SIGCHLD; 416 417 /* bump references to the text vnode (for procfs) */ 418 p2->p_textvp = p1->p_textvp; 419 if (p2->p_textvp) 420 vref(p2->p_textvp); 421 422 if (flags & RFCFDG) { 423 p2->p_fd = fdinit(p1); 424 fdtol = NULL; 425 } else if (flags & RFFDG) { 426 p2->p_fd = fdcopy(p1); 427 fdtol = NULL; 428 } else { 429 p2->p_fd = fdshare(p1); 430 if (p1->p_fdtol == NULL) 431 p1->p_fdtol = 432 filedesc_to_leader_alloc(NULL, 433 p1->p_leader); 434 if ((flags & RFTHREAD) != 0) { 435 /* 436 * Shared file descriptor table and 437 * shared process leaders. 438 */ 439 fdtol = p1->p_fdtol; 440 fdtol->fdl_refcount++; 441 } else { 442 /* 443 * Shared file descriptor table, and 444 * different process leaders 445 */ 446 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 447 } 448 } 449 p2->p_fdtol = fdtol; 450 451 /* 452 * If p_limit is still copy-on-write, bump refcnt, 453 * otherwise get a copy that won't be modified. 454 * (If PL_SHAREMOD is clear, the structure is shared 455 * copy-on-write.) 456 */ 457 if (p1->p_limit->p_lflags & PL_SHAREMOD) { 458 p2->p_limit = limcopy(p1->p_limit); 459 } else { 460 p2->p_limit = p1->p_limit; 461 p2->p_limit->p_refcnt++; 462 } 463 464 /* 465 * Preserve some more flags in subprocess. P_PROFIL has already 466 * been preserved. 467 */ 468 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 469 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 470 p2->p_flag |= P_CONTROLT; 471 if (flags & RFPPWAIT) 472 p2->p_flag |= P_PPWAIT; 473 474 /* 475 * Once we are on a pglist we may receive signals. XXX we might 476 * race a ^C being sent to the process group by not receiving it 477 * at all prior to this line. 478 */ 479 LIST_INSERT_AFTER(p1, p2, p_pglist); 480 481 /* 482 * Attach the new process to its parent. 483 * 484 * If RFNOWAIT is set, the newly created process becomes a child 485 * of init. This effectively disassociates the child from the 486 * parent. 487 */ 488 if (flags & RFNOWAIT) 489 pptr = initproc; 490 else 491 pptr = p1; 492 p2->p_pptr = pptr; 493 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 494 LIST_INIT(&p2->p_children); 495 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 496 callout_init(&p2->p_ithandle); 497 498 #ifdef KTRACE 499 /* 500 * Copy traceflag and tracefile if enabled. If not inherited, 501 * these were zeroed above but we still could have a trace race 502 * so make sure p2's p_tracep is NULL. 503 */ 504 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 505 p2->p_traceflag = p1->p_traceflag; 506 if ((p2->p_tracep = p1->p_tracep) != NULL) 507 vref(p2->p_tracep); 508 } 509 #endif 510 511 /* 512 * Inherit the scheduler and initialize scheduler-related fields. 513 * Set cpbase to the last timeout that occured (not the upcoming 514 * timeout). 515 */ 516 p2->p_usched = p1->p_usched; 517 p2->p_cpbase = mycpu->gd_schedclock.time - 518 mycpu->gd_schedclock.periodic; 519 p2->p_usched->heuristic_forking(p1, p2); 520 521 /* 522 * This begins the section where we must prevent the parent 523 * from being swapped. 524 */ 525 PHOLD(p1); 526 527 /* 528 * Finish creating the child process. It will return via a different 529 * execution path later. (ie: directly into user mode) 530 */ 531 vm_fork(p1, p2, flags); 532 caps_fork(p1, p2, flags); 533 534 if (flags == (RFFDG | RFPROC)) { 535 mycpu->gd_cnt.v_forks++; 536 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 537 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 538 mycpu->gd_cnt.v_vforks++; 539 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 540 } else if (p1 == &proc0) { 541 mycpu->gd_cnt.v_kthreads++; 542 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 543 } else { 544 mycpu->gd_cnt.v_rforks++; 545 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 546 } 547 548 /* 549 * Both processes are set up, now check if any loadable modules want 550 * to adjust anything. 551 * What if they have an error? XXX 552 */ 553 TAILQ_FOREACH(ep, &fork_list, next) { 554 (*ep->function)(p1, p2, flags); 555 } 556 557 /* 558 * Set the start time. Note that the process is not runnable. The 559 * caller is responsible for making it runnable. 560 */ 561 microtime(&p2->p_thread->td_start); 562 p2->p_acflag = AFORK; 563 564 /* 565 * tell any interested parties about the new process 566 */ 567 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 568 569 /* 570 * Return child proc pointer to parent. 571 */ 572 *procp = p2; 573 return (0); 574 } 575 576 /* 577 * The next two functionms are general routines to handle adding/deleting 578 * items on the fork callout list. 579 * 580 * at_fork(): 581 * Take the arguments given and put them onto the fork callout list, 582 * However first make sure that it's not already there. 583 * Returns 0 on success or a standard error number. 584 */ 585 int 586 at_fork(forklist_fn function) 587 { 588 struct forklist *ep; 589 590 #ifdef INVARIANTS 591 /* let the programmer know if he's been stupid */ 592 if (rm_at_fork(function)) { 593 printf("WARNING: fork callout entry (%p) already present\n", 594 function); 595 } 596 #endif 597 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 598 ep->function = function; 599 TAILQ_INSERT_TAIL(&fork_list, ep, next); 600 return (0); 601 } 602 603 /* 604 * Scan the exit callout list for the given item and remove it.. 605 * Returns the number of items removed (0 or 1) 606 */ 607 int 608 rm_at_fork(forklist_fn function) 609 { 610 struct forklist *ep; 611 612 TAILQ_FOREACH(ep, &fork_list, next) { 613 if (ep->function == function) { 614 TAILQ_REMOVE(&fork_list, ep, next); 615 free(ep, M_ATFORK); 616 return(1); 617 } 618 } 619 return (0); 620 } 621 622 /* 623 * Add a forked process to the run queue after any remaining setup, such 624 * as setting the fork handler, has been completed. 625 */ 626 void 627 start_forked_proc(struct proc *p1, struct proc *p2) 628 { 629 /* 630 * Move from SIDL to RUN queue, and activate the process's thread. 631 * Activation of the thread effectively makes the process "a" 632 * current process, so we do not setrunqueue(). 633 * 634 * YYY setrunqueue works here but we should clean up the trampoline 635 * code so we just schedule the LWKT thread and let the trampoline 636 * deal with the userland scheduler on return to userland. 637 */ 638 KASSERT(p2 && p2->p_stat == SIDL, 639 ("cannot start forked process, bad status: %p", p2)); 640 p2->p_usched->resetpriority(p2); 641 crit_enter(); 642 p2->p_stat = SRUN; 643 p2->p_usched->setrunqueue(p2); 644 crit_exit(); 645 646 /* 647 * Now can be swapped. 648 */ 649 PRELE(p1); 650 651 /* 652 * Preserve synchronization semantics of vfork. If waiting for 653 * child to exec or exit, set P_PPWAIT on child, and sleep on our 654 * proc (in case of exit). 655 */ 656 while (p2->p_flag & P_PPWAIT) 657 tsleep(p1, 0, "ppwait", 0); 658 } 659 660