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 1.30 2004/10/12 19:20:46 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 71 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 72 73 /* 74 * These are the stuctures used to create a callout list for things to do 75 * when forking a process 76 */ 77 struct forklist { 78 forklist_fn function; 79 TAILQ_ENTRY(forklist) next; 80 }; 81 82 TAILQ_HEAD(forklist_head, forklist); 83 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 84 85 int forksleep; /* Place for fork1() to sleep on. */ 86 87 /* ARGSUSED */ 88 int 89 fork(struct fork_args *uap) 90 { 91 struct proc *p = curproc; 92 struct proc *p2; 93 int error; 94 95 error = fork1(p, RFFDG | RFPROC, &p2); 96 if (error == 0) { 97 start_forked_proc(p, p2); 98 uap->sysmsg_fds[0] = p2->p_pid; 99 uap->sysmsg_fds[1] = 0; 100 } 101 return error; 102 } 103 104 /* ARGSUSED */ 105 int 106 vfork(struct vfork_args *uap) 107 { 108 struct proc *p = curproc; 109 struct proc *p2; 110 int error; 111 112 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2); 113 if (error == 0) { 114 start_forked_proc(p, p2); 115 uap->sysmsg_fds[0] = p2->p_pid; 116 uap->sysmsg_fds[1] = 0; 117 } 118 return error; 119 } 120 121 /* 122 * Handle rforks. An rfork may (1) operate on the current process without 123 * creating a new, (2) create a new process that shared the current process's 124 * vmspace, signals, and/or descriptors, or (3) create a new process that does 125 * not share these things (normal fork). 126 * 127 * Note that we only call start_forked_proc() if a new process is actually 128 * created. 129 * 130 * rfork { int flags } 131 */ 132 int 133 rfork(struct rfork_args *uap) 134 { 135 struct proc *p = curproc; 136 struct proc *p2; 137 int error; 138 139 if ((uap->flags & RFKERNELONLY) != 0) 140 return (EINVAL); 141 142 error = fork1(p, uap->flags, &p2); 143 if (error == 0) { 144 if (p2) 145 start_forked_proc(p, p2); 146 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0; 147 uap->sysmsg_fds[1] = 0; 148 } 149 return error; 150 } 151 152 153 int nprocs = 1; /* process 0 */ 154 static int nextpid = 0; 155 156 /* 157 * Random component to nextpid generation. We mix in a random factor to make 158 * it a little harder to predict. We sanity check the modulus value to avoid 159 * doing it in critical paths. Don't let it be too small or we pointlessly 160 * waste randomness entropy, and don't let it be impossibly large. Using a 161 * modulus that is too big causes a LOT more process table scans and slows 162 * down fork processing as the pidchecked caching is defeated. 163 */ 164 static int randompid = 0; 165 166 static int 167 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 168 { 169 int error, pid; 170 171 pid = randompid; 172 error = sysctl_handle_int(oidp, &pid, 0, req); 173 if (error || !req->newptr) 174 return (error); 175 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 176 pid = PID_MAX - 100; 177 else if (pid < 2) /* NOP */ 178 pid = 0; 179 else if (pid < 100) /* Make it reasonable */ 180 pid = 100; 181 randompid = pid; 182 return (error); 183 } 184 185 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 186 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 187 188 int 189 fork1(struct proc *p1, int flags, struct proc **procp) 190 { 191 struct proc *p2, *pptr; 192 uid_t uid; 193 struct proc *newproc; 194 int ok; 195 static int curfail = 0, pidchecked = 0; 196 static struct timeval lastfail; 197 struct forklist *ep; 198 struct filedesc_to_leader *fdtol; 199 200 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 201 return (EINVAL); 202 203 /* 204 * Here we don't create a new process, but we divorce 205 * certain parts of a process from itself. 206 */ 207 if ((flags & RFPROC) == 0) { 208 209 vm_fork(p1, 0, flags); 210 211 /* 212 * Close all file descriptors. 213 */ 214 if (flags & RFCFDG) { 215 struct filedesc *fdtmp; 216 fdtmp = fdinit(p1); 217 fdfree(p1); 218 p1->p_fd = fdtmp; 219 } 220 221 /* 222 * Unshare file descriptors (from parent.) 223 */ 224 if (flags & RFFDG) { 225 if (p1->p_fd->fd_refcnt > 1) { 226 struct filedesc *newfd; 227 newfd = fdcopy(p1); 228 fdfree(p1); 229 p1->p_fd = newfd; 230 } 231 } 232 *procp = NULL; 233 return (0); 234 } 235 236 /* 237 * Although process entries are dynamically created, we still keep 238 * a global limit on the maximum number we will create. Don't allow 239 * a nonprivileged user to use the last ten processes; don't let root 240 * exceed the limit. The variable nprocs is the current number of 241 * processes, maxproc is the limit. 242 */ 243 uid = p1->p_ucred->cr_ruid; 244 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 245 if (ppsratecheck(&lastfail, &curfail, 1)) 246 printf("maxproc limit exceeded by uid %d, please " 247 "see tuning(7) and login.conf(5).\n", uid); 248 tsleep(&forksleep, 0, "fork", hz / 2); 249 return (EAGAIN); 250 } 251 /* 252 * Increment the nprocs resource before blocking can occur. There 253 * are hard-limits as to the number of processes that can run. 254 */ 255 nprocs++; 256 257 /* 258 * Increment the count of procs running with this uid. Don't allow 259 * a nonprivileged user to exceed their current limit. 260 */ 261 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1, 262 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 263 if (!ok) { 264 /* 265 * Back out the process count 266 */ 267 nprocs--; 268 if (ppsratecheck(&lastfail, &curfail, 1)) 269 printf("maxproc limit exceeded by uid %d, please " 270 "see tuning(7) and login.conf(5).\n", uid); 271 tsleep(&forksleep, 0, "fork", hz / 2); 272 return (EAGAIN); 273 } 274 275 /* Allocate new proc. */ 276 newproc = zalloc(proc_zone); 277 278 /* 279 * Setup linkage for kernel based threading 280 */ 281 if ((flags & RFTHREAD) != 0) { 282 newproc->p_peers = p1->p_peers; 283 p1->p_peers = newproc; 284 newproc->p_leader = p1->p_leader; 285 } else { 286 newproc->p_peers = 0; 287 newproc->p_leader = newproc; 288 } 289 290 newproc->p_wakeup = 0; 291 newproc->p_vmspace = NULL; 292 TAILQ_INIT(&newproc->p_sysmsgq); 293 294 /* 295 * Find an unused process ID. We remember a range of unused IDs 296 * ready to use (from nextpid+1 through pidchecked-1). 297 */ 298 nextpid++; 299 if (randompid) 300 nextpid += arc4random() % randompid; 301 retry: 302 /* 303 * If the process ID prototype has wrapped around, 304 * restart somewhat above 0, as the low-numbered procs 305 * tend to include daemons that don't exit. 306 */ 307 if (nextpid >= PID_MAX) { 308 nextpid = nextpid % PID_MAX; 309 if (nextpid < 100) 310 nextpid += 100; 311 pidchecked = 0; 312 } 313 if (nextpid >= pidchecked) { 314 int doingzomb = 0; 315 316 pidchecked = PID_MAX; 317 /* 318 * Scan the active and zombie procs to check whether this pid 319 * is in use. Remember the lowest pid that's greater 320 * than nextpid, so we can avoid checking for a while. 321 */ 322 p2 = LIST_FIRST(&allproc); 323 again: 324 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 325 while (p2->p_pid == nextpid || 326 p2->p_pgrp->pg_id == nextpid || 327 p2->p_session->s_sid == nextpid) { 328 nextpid++; 329 if (nextpid >= pidchecked) 330 goto retry; 331 } 332 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 333 pidchecked = p2->p_pid; 334 if (p2->p_pgrp->pg_id > nextpid && 335 pidchecked > p2->p_pgrp->pg_id) 336 pidchecked = p2->p_pgrp->pg_id; 337 if (p2->p_session->s_sid > nextpid && 338 pidchecked > p2->p_session->s_sid) 339 pidchecked = p2->p_session->s_sid; 340 } 341 if (!doingzomb) { 342 doingzomb = 1; 343 p2 = LIST_FIRST(&zombproc); 344 goto again; 345 } 346 } 347 348 p2 = newproc; 349 p2->p_stat = SIDL; /* protect against others */ 350 p2->p_pid = nextpid; 351 LIST_INSERT_HEAD(&allproc, p2, p_list); 352 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 353 354 /* 355 * Make a proc table entry for the new process. 356 * Start by zeroing the section of proc that is zero-initialized, 357 * then copy the section that is copied directly from the parent. 358 */ 359 bzero(&p2->p_startzero, 360 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 361 bcopy(&p1->p_startcopy, &p2->p_startcopy, 362 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 363 364 p2->p_aioinfo = NULL; 365 366 /* 367 * Duplicate sub-structures as needed. 368 * Increase reference counts on shared objects. 369 * The p_stats and p_sigacts substructs are set in vm_fork. 370 */ 371 p2->p_flag = P_INMEM; 372 if (p1->p_flag & P_PROFIL) 373 startprofclock(p2); 374 p2->p_ucred = crhold(p1->p_ucred); 375 376 if (p2->p_ucred->cr_prison) { 377 p2->p_ucred->cr_prison->pr_ref++; 378 p2->p_flag |= P_JAILED; 379 } 380 381 if (p2->p_args) 382 p2->p_args->ar_ref++; 383 384 if (flags & RFSIGSHARE) { 385 p2->p_procsig = p1->p_procsig; 386 p2->p_procsig->ps_refcnt++; 387 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 388 struct sigacts *newsigacts; 389 int s; 390 391 /* Create the shared sigacts structure */ 392 MALLOC(newsigacts, struct sigacts *, 393 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 394 s = splhigh(); 395 /* 396 * Set p_sigacts to the new shared structure. 397 * Note that this is updating p1->p_sigacts at the 398 * same time, since p_sigacts is just a pointer to 399 * the shared p_procsig->ps_sigacts. 400 */ 401 p2->p_sigacts = newsigacts; 402 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 403 sizeof(*p2->p_sigacts)); 404 *p2->p_sigacts = p1->p_addr->u_sigacts; 405 splx(s); 406 } 407 } else { 408 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 409 M_SUBPROC, M_WAITOK); 410 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 411 p2->p_procsig->ps_refcnt = 1; 412 p2->p_sigacts = NULL; /* finished in vm_fork() */ 413 } 414 if (flags & RFLINUXTHPN) 415 p2->p_sigparent = SIGUSR1; 416 else 417 p2->p_sigparent = SIGCHLD; 418 419 /* bump references to the text vnode (for procfs) */ 420 p2->p_textvp = p1->p_textvp; 421 if (p2->p_textvp) 422 vref(p2->p_textvp); 423 424 if (flags & RFCFDG) { 425 p2->p_fd = fdinit(p1); 426 fdtol = NULL; 427 } else if (flags & RFFDG) { 428 p2->p_fd = fdcopy(p1); 429 fdtol = NULL; 430 } else { 431 p2->p_fd = fdshare(p1); 432 if (p1->p_fdtol == NULL) 433 p1->p_fdtol = 434 filedesc_to_leader_alloc(NULL, 435 p1->p_leader); 436 if ((flags & RFTHREAD) != 0) { 437 /* 438 * Shared file descriptor table and 439 * shared process leaders. 440 */ 441 fdtol = p1->p_fdtol; 442 fdtol->fdl_refcount++; 443 } else { 444 /* 445 * Shared file descriptor table, and 446 * different process leaders 447 */ 448 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 449 } 450 } 451 p2->p_fdtol = fdtol; 452 453 /* 454 * If p_limit is still copy-on-write, bump refcnt, 455 * otherwise get a copy that won't be modified. 456 * (If PL_SHAREMOD is clear, the structure is shared 457 * copy-on-write.) 458 */ 459 if (p1->p_limit->p_lflags & PL_SHAREMOD) { 460 p2->p_limit = limcopy(p1->p_limit); 461 } else { 462 p2->p_limit = p1->p_limit; 463 p2->p_limit->p_refcnt++; 464 } 465 466 /* 467 * Preserve some more flags in subprocess. P_PROFIL has already 468 * been preserved. 469 */ 470 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 471 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 472 p2->p_flag |= P_CONTROLT; 473 if (flags & RFPPWAIT) 474 p2->p_flag |= P_PPWAIT; 475 476 /* 477 * Once we are on a pglist we may receive signals. XXX we might 478 * race a ^C being sent to the process group by not receiving it 479 * at all prior to this line. 480 */ 481 LIST_INSERT_AFTER(p1, p2, p_pglist); 482 483 /* 484 * Attach the new process to its parent. 485 * 486 * If RFNOWAIT is set, the newly created process becomes a child 487 * of init. This effectively disassociates the child from the 488 * parent. 489 */ 490 if (flags & RFNOWAIT) 491 pptr = initproc; 492 else 493 pptr = p1; 494 p2->p_pptr = pptr; 495 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 496 LIST_INIT(&p2->p_children); 497 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 498 callout_init(&p2->p_ithandle); 499 500 #ifdef KTRACE 501 /* 502 * Copy traceflag and tracefile if enabled. If not inherited, 503 * these were zeroed above but we still could have a trace race 504 * so make sure p2's p_tracep is NULL. 505 */ 506 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 507 p2->p_traceflag = p1->p_traceflag; 508 if ((p2->p_tracep = p1->p_tracep) != NULL) 509 vref(p2->p_tracep); 510 } 511 #endif 512 513 /* 514 * Give the child process an estcpu skewed towards the batch side 515 * of the parent. This prevents batch programs from glitching 516 * interactive programs when they are first started. If the child 517 * is not a batch program it's priority will be corrected by the 518 * scheduler. 519 * 520 * The interactivity model always starts at 0 (par value). 521 */ 522 p2->p_estcpu_fork = p2->p_estcpu = 523 ESTCPULIM(p1->p_estcpu + ESTCPURAMP); 524 p2->p_interactive = 0; 525 526 /* 527 * This begins the section where we must prevent the parent 528 * from being swapped. 529 */ 530 PHOLD(p1); 531 532 /* 533 * Finish creating the child process. It will return via a different 534 * execution path later. (ie: directly into user mode) 535 */ 536 vm_fork(p1, p2, flags); 537 caps_fork(p1, p2, flags); 538 539 if (flags == (RFFDG | RFPROC)) { 540 mycpu->gd_cnt.v_forks++; 541 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 542 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 543 mycpu->gd_cnt.v_vforks++; 544 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 545 } else if (p1 == &proc0) { 546 mycpu->gd_cnt.v_kthreads++; 547 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 548 } else { 549 mycpu->gd_cnt.v_rforks++; 550 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 551 } 552 553 /* 554 * Both processes are set up, now check if any loadable modules want 555 * to adjust anything. 556 * What if they have an error? XXX 557 */ 558 TAILQ_FOREACH(ep, &fork_list, next) { 559 (*ep->function)(p1, p2, flags); 560 } 561 562 /* 563 * Make child runnable and add to run queue. 564 */ 565 microtime(&p2->p_thread->td_start); 566 p2->p_acflag = AFORK; 567 568 /* 569 * tell any interested parties about the new process 570 */ 571 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 572 573 /* 574 * Return child proc pointer to parent. 575 */ 576 *procp = p2; 577 return (0); 578 } 579 580 /* 581 * The next two functionms are general routines to handle adding/deleting 582 * items on the fork callout list. 583 * 584 * at_fork(): 585 * Take the arguments given and put them onto the fork callout list, 586 * However first make sure that it's not already there. 587 * Returns 0 on success or a standard error number. 588 */ 589 int 590 at_fork(forklist_fn function) 591 { 592 struct forklist *ep; 593 594 #ifdef INVARIANTS 595 /* let the programmer know if he's been stupid */ 596 if (rm_at_fork(function)) { 597 printf("WARNING: fork callout entry (%p) already present\n", 598 function); 599 } 600 #endif 601 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 602 ep->function = function; 603 TAILQ_INSERT_TAIL(&fork_list, ep, next); 604 return (0); 605 } 606 607 /* 608 * Scan the exit callout list for the given item and remove it.. 609 * Returns the number of items removed (0 or 1) 610 */ 611 int 612 rm_at_fork(forklist_fn function) 613 { 614 struct forklist *ep; 615 616 TAILQ_FOREACH(ep, &fork_list, next) { 617 if (ep->function == function) { 618 TAILQ_REMOVE(&fork_list, ep, next); 619 free(ep, M_ATFORK); 620 return(1); 621 } 622 } 623 return (0); 624 } 625 626 /* 627 * Add a forked process to the run queue after any remaining setup, such 628 * as setting the fork handler, has been completed. 629 */ 630 void 631 start_forked_proc(struct proc *p1, struct proc *p2) 632 { 633 /* 634 * Move from SIDL to RUN queue, and activate the process's thread. 635 * Activation of the thread effectively makes the process "a" 636 * current process, so we do not setrunqueue(). 637 * 638 * YYY setrunqueue works here but we should clean up the trampoline 639 * code so we just schedule the LWKT thread and let the trampoline 640 * deal with the userland scheduler on return to userland. 641 */ 642 KASSERT(p2 && p2->p_stat == SIDL, 643 ("cannot start forked process, bad status: %p", p2)); 644 resetpriority(p2); 645 (void) splhigh(); 646 p2->p_stat = SRUN; 647 setrunqueue(p2); 648 (void) spl0(); 649 650 /* 651 * Now can be swapped. 652 */ 653 PRELE(p1); 654 655 /* 656 * Preserve synchronization semantics of vfork. If waiting for 657 * child to exec or exit, set P_PPWAIT on child, and sleep on our 658 * proc (in case of exit). 659 */ 660 while (p2->p_flag & P_PPWAIT) 661 tsleep(p1, 0, "ppwait", 0); 662 } 663 664