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 1.21 2004/03/30 19:14:11 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 int 122 rfork(struct rfork_args *uap) 123 { 124 struct proc *p = curproc; 125 struct proc *p2; 126 int error; 127 128 /* Don't allow kernel only flags */ 129 if ((uap->flags & RFKERNELONLY) != 0) 130 return (EINVAL); 131 132 error = fork1(p, uap->flags, &p2); 133 if (error == 0) { 134 start_forked_proc(p, p2); 135 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0; 136 uap->sysmsg_fds[1] = 0; 137 } 138 return error; 139 } 140 141 142 int nprocs = 1; /* process 0 */ 143 static int nextpid = 0; 144 145 /* 146 * Random component to nextpid generation. We mix in a random factor to make 147 * it a little harder to predict. We sanity check the modulus value to avoid 148 * doing it in critical paths. Don't let it be too small or we pointlessly 149 * waste randomness entropy, and don't let it be impossibly large. Using a 150 * modulus that is too big causes a LOT more process table scans and slows 151 * down fork processing as the pidchecked caching is defeated. 152 */ 153 static int randompid = 0; 154 155 static int 156 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 157 { 158 int error, pid; 159 160 pid = randompid; 161 error = sysctl_handle_int(oidp, &pid, 0, req); 162 if (error || !req->newptr) 163 return (error); 164 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 165 pid = PID_MAX - 100; 166 else if (pid < 2) /* NOP */ 167 pid = 0; 168 else if (pid < 100) /* Make it reasonable */ 169 pid = 100; 170 randompid = pid; 171 return (error); 172 } 173 174 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 175 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 176 177 int 178 fork1(p1, flags, procp) 179 struct proc *p1; 180 int flags; 181 struct proc **procp; 182 { 183 struct proc *p2, *pptr; 184 uid_t uid; 185 struct proc *newproc; 186 int ok; 187 static int curfail = 0, pidchecked = 0; 188 static struct timeval lastfail; 189 struct forklist *ep; 190 struct filedesc_to_leader *fdtol; 191 192 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 193 return (EINVAL); 194 195 /* 196 * Here we don't create a new process, but we divorce 197 * certain parts of a process from itself. 198 */ 199 if ((flags & RFPROC) == 0) { 200 201 vm_fork(p1, 0, flags); 202 203 /* 204 * Close all file descriptors. 205 */ 206 if (flags & RFCFDG) { 207 struct filedesc *fdtmp; 208 fdtmp = fdinit(p1); 209 fdfree(p1); 210 p1->p_fd = fdtmp; 211 } 212 213 /* 214 * Unshare file descriptors (from parent.) 215 */ 216 if (flags & RFFDG) { 217 if (p1->p_fd->fd_refcnt > 1) { 218 struct filedesc *newfd; 219 newfd = fdcopy(p1); 220 fdfree(p1); 221 p1->p_fd = newfd; 222 } 223 } 224 *procp = NULL; 225 return (0); 226 } 227 228 /* 229 * Although process entries are dynamically created, we still keep 230 * a global limit on the maximum number we will create. Don't allow 231 * a nonprivileged user to use the last ten processes; don't let root 232 * exceed the limit. The variable nprocs is the current number of 233 * processes, maxproc is the limit. 234 */ 235 uid = p1->p_ucred->cr_ruid; 236 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 237 if (ppsratecheck(&lastfail, &curfail, 1)) 238 printf("maxproc limit exceeded by uid %d, please " 239 "see tuning(7) and login.conf(5).\n", uid); 240 tsleep(&forksleep, 0, "fork", hz / 2); 241 return (EAGAIN); 242 } 243 /* 244 * Increment the nprocs resource before blocking can occur. There 245 * are hard-limits as to the number of processes that can run. 246 */ 247 nprocs++; 248 249 /* 250 * Increment the count of procs running with this uid. Don't allow 251 * a nonprivileged user to exceed their current limit. 252 */ 253 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1, 254 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 255 if (!ok) { 256 /* 257 * Back out the process count 258 */ 259 nprocs--; 260 if (ppsratecheck(&lastfail, &curfail, 1)) 261 printf("maxproc limit exceeded by uid %d, please " 262 "see tuning(7) and login.conf(5).\n", uid); 263 tsleep(&forksleep, 0, "fork", hz / 2); 264 return (EAGAIN); 265 } 266 267 /* Allocate new proc. */ 268 newproc = zalloc(proc_zone); 269 270 /* 271 * Setup linkage for kernel based threading 272 */ 273 if((flags & RFTHREAD) != 0) { 274 newproc->p_peers = p1->p_peers; 275 p1->p_peers = newproc; 276 newproc->p_leader = p1->p_leader; 277 } else { 278 newproc->p_peers = 0; 279 newproc->p_leader = newproc; 280 } 281 282 newproc->p_wakeup = 0; 283 newproc->p_vmspace = NULL; 284 285 /* 286 * Find an unused process ID. We remember a range of unused IDs 287 * ready to use (from nextpid+1 through pidchecked-1). 288 */ 289 nextpid++; 290 if (randompid) 291 nextpid += arc4random() % randompid; 292 retry: 293 /* 294 * If the process ID prototype has wrapped around, 295 * restart somewhat above 0, as the low-numbered procs 296 * tend to include daemons that don't exit. 297 */ 298 if (nextpid >= PID_MAX) { 299 nextpid = nextpid % PID_MAX; 300 if (nextpid < 100) 301 nextpid += 100; 302 pidchecked = 0; 303 } 304 if (nextpid >= pidchecked) { 305 int doingzomb = 0; 306 307 pidchecked = PID_MAX; 308 /* 309 * Scan the active and zombie procs to check whether this pid 310 * is in use. Remember the lowest pid that's greater 311 * than nextpid, so we can avoid checking for a while. 312 */ 313 p2 = LIST_FIRST(&allproc); 314 again: 315 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 316 while (p2->p_pid == nextpid || 317 p2->p_pgrp->pg_id == nextpid || 318 p2->p_session->s_sid == nextpid) { 319 nextpid++; 320 if (nextpid >= pidchecked) 321 goto retry; 322 } 323 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 324 pidchecked = p2->p_pid; 325 if (p2->p_pgrp->pg_id > nextpid && 326 pidchecked > p2->p_pgrp->pg_id) 327 pidchecked = p2->p_pgrp->pg_id; 328 if (p2->p_session->s_sid > nextpid && 329 pidchecked > p2->p_session->s_sid) 330 pidchecked = p2->p_session->s_sid; 331 } 332 if (!doingzomb) { 333 doingzomb = 1; 334 p2 = LIST_FIRST(&zombproc); 335 goto again; 336 } 337 } 338 339 p2 = newproc; 340 p2->p_stat = SIDL; /* protect against others */ 341 p2->p_pid = nextpid; 342 LIST_INSERT_HEAD(&allproc, p2, p_list); 343 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 344 345 /* 346 * Make a proc table entry for the new process. 347 * Start by zeroing the section of proc that is zero-initialized, 348 * then copy the section that is copied directly from the parent. 349 */ 350 bzero(&p2->p_startzero, 351 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 352 bcopy(&p1->p_startcopy, &p2->p_startcopy, 353 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 354 355 p2->p_aioinfo = NULL; 356 357 /* 358 * Duplicate sub-structures as needed. 359 * Increase reference counts on shared objects. 360 * The p_stats and p_sigacts substructs are set in vm_fork. 361 * 362 * P_CP_RELEASED indicates that the process is starting out in 363 * the kernel (in the fork trampoline). The flag will be cleared 364 * when the new process calls userret() and acquires its current 365 * process designation for the return to userland. 366 */ 367 p2->p_flag = P_INMEM | P_CP_RELEASED; 368 if (p1->p_flag & P_PROFIL) 369 startprofclock(p2); 370 p2->p_ucred = crhold(p1->p_ucred); 371 372 if (p2->p_ucred->cr_prison) { 373 p2->p_ucred->cr_prison->pr_ref++; 374 p2->p_flag |= P_JAILED; 375 } 376 377 if (p2->p_args) 378 p2->p_args->ar_ref++; 379 380 if (flags & RFSIGSHARE) { 381 p2->p_procsig = p1->p_procsig; 382 p2->p_procsig->ps_refcnt++; 383 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 384 struct sigacts *newsigacts; 385 int s; 386 387 /* Create the shared sigacts structure */ 388 MALLOC(newsigacts, struct sigacts *, 389 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 390 s = splhigh(); 391 /* 392 * Set p_sigacts to the new shared structure. 393 * Note that this is updating p1->p_sigacts at the 394 * same time, since p_sigacts is just a pointer to 395 * the shared p_procsig->ps_sigacts. 396 */ 397 p2->p_sigacts = newsigacts; 398 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 399 sizeof(*p2->p_sigacts)); 400 *p2->p_sigacts = p1->p_addr->u_sigacts; 401 splx(s); 402 } 403 } else { 404 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 405 M_SUBPROC, M_WAITOK); 406 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 407 p2->p_procsig->ps_refcnt = 1; 408 p2->p_sigacts = NULL; /* finished in vm_fork() */ 409 } 410 if (flags & RFLINUXTHPN) 411 p2->p_sigparent = SIGUSR1; 412 else 413 p2->p_sigparent = SIGCHLD; 414 415 /* bump references to the text vnode (for procfs) */ 416 p2->p_textvp = p1->p_textvp; 417 if (p2->p_textvp) 418 VREF(p2->p_textvp); 419 420 if (flags & RFCFDG) { 421 p2->p_fd = fdinit(p1); 422 fdtol = NULL; 423 } else if (flags & RFFDG) { 424 p2->p_fd = fdcopy(p1); 425 fdtol = NULL; 426 } else { 427 p2->p_fd = fdshare(p1); 428 if (p1->p_fdtol == NULL) 429 p1->p_fdtol = 430 filedesc_to_leader_alloc(NULL, 431 p1->p_leader); 432 if ((flags & RFTHREAD) != 0) { 433 /* 434 * Shared file descriptor table and 435 * shared process leaders. 436 */ 437 fdtol = p1->p_fdtol; 438 fdtol->fdl_refcount++; 439 } else { 440 /* 441 * Shared file descriptor table, and 442 * different process leaders 443 */ 444 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 445 } 446 } 447 p2->p_fdtol = fdtol; 448 449 /* 450 * If p_limit is still copy-on-write, bump refcnt, 451 * otherwise get a copy that won't be modified. 452 * (If PL_SHAREMOD is clear, the structure is shared 453 * copy-on-write.) 454 */ 455 if (p1->p_limit->p_lflags & PL_SHAREMOD) 456 p2->p_limit = limcopy(p1->p_limit); 457 else { 458 p2->p_limit = p1->p_limit; 459 p2->p_limit->p_refcnt++; 460 } 461 462 /* 463 * Preserve some more flags in subprocess. P_PROFIL has already 464 * been preserved. 465 */ 466 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 467 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 468 p2->p_flag |= P_CONTROLT; 469 if (flags & RFPPWAIT) 470 p2->p_flag |= P_PPWAIT; 471 472 LIST_INSERT_AFTER(p1, p2, p_pglist); 473 474 /* 475 * Attach the new process to its parent. 476 * 477 * If RFNOWAIT is set, the newly created process becomes a child 478 * of init. This effectively disassociates the child from the 479 * parent. 480 */ 481 if (flags & RFNOWAIT) 482 pptr = initproc; 483 else 484 pptr = p1; 485 p2->p_pptr = pptr; 486 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 487 LIST_INIT(&p2->p_children); 488 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 489 490 #ifdef KTRACE 491 /* 492 * Copy traceflag and tracefile if enabled. If not inherited, 493 * these were zeroed above but we still could have a trace race 494 * so make sure p2's p_tracep is NULL. 495 */ 496 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 497 p2->p_traceflag = p1->p_traceflag; 498 if ((p2->p_tracep = p1->p_tracep) != NULL) 499 VREF(p2->p_tracep); 500 } 501 #endif 502 503 /* 504 * Give the child process an estcpu skewed towards the batch side 505 * of the parent. This prevents batch programs from glitching 506 * interactive programs when they are first started. If the child 507 * is not a batch program it's priority will be corrected by the 508 * scheduler. 509 */ 510 p2->p_estcpu_fork = p2->p_estcpu = 511 ESTCPULIM(p1->p_estcpu + ESTCPURAMP); 512 513 /* 514 * This begins the section where we must prevent the parent 515 * from being swapped. 516 */ 517 PHOLD(p1); 518 519 /* 520 * Finish creating the child process. It will return via a different 521 * execution path later. (ie: directly into user mode) 522 */ 523 vm_fork(p1, p2, flags); 524 caps_fork(p1, p2, flags); 525 526 if (flags == (RFFDG | RFPROC)) { 527 mycpu->gd_cnt.v_forks++; 528 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 529 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 530 mycpu->gd_cnt.v_vforks++; 531 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 532 } else if (p1 == &proc0) { 533 mycpu->gd_cnt.v_kthreads++; 534 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 535 } else { 536 mycpu->gd_cnt.v_rforks++; 537 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 538 } 539 540 /* 541 * Both processes are set up, now check if any loadable modules want 542 * to adjust anything. 543 * What if they have an error? XXX 544 */ 545 TAILQ_FOREACH(ep, &fork_list, next) { 546 (*ep->function)(p1, p2, flags); 547 } 548 549 /* 550 * Make child runnable and add to run queue. 551 */ 552 microtime(&(p2->p_stats->p_start)); 553 p2->p_acflag = AFORK; 554 555 /* 556 * tell any interested parties about the new process 557 */ 558 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 559 560 /* 561 * Return child proc pointer to parent. 562 */ 563 *procp = p2; 564 return (0); 565 } 566 567 /* 568 * The next two functionms are general routines to handle adding/deleting 569 * items on the fork callout list. 570 * 571 * at_fork(): 572 * Take the arguments given and put them onto the fork callout list, 573 * However first make sure that it's not already there. 574 * Returns 0 on success or a standard error number. 575 */ 576 577 int 578 at_fork(function) 579 forklist_fn function; 580 { 581 struct forklist *ep; 582 583 #ifdef INVARIANTS 584 /* let the programmer know if he's been stupid */ 585 if (rm_at_fork(function)) 586 printf("WARNING: fork callout entry (%p) already present\n", 587 function); 588 #endif 589 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT); 590 if (ep == NULL) 591 return (ENOMEM); 592 ep->function = function; 593 TAILQ_INSERT_TAIL(&fork_list, ep, next); 594 return (0); 595 } 596 597 /* 598 * Scan the exit callout list for the given item and remove it.. 599 * Returns the number of items removed (0 or 1) 600 */ 601 602 int 603 rm_at_fork(function) 604 forklist_fn function; 605 { 606 struct forklist *ep; 607 608 TAILQ_FOREACH(ep, &fork_list, next) { 609 if (ep->function == function) { 610 TAILQ_REMOVE(&fork_list, ep, next); 611 free(ep, M_ATFORK); 612 return(1); 613 } 614 } 615 return (0); 616 } 617 618 /* 619 * Add a forked process to the run queue after any remaining setup, such 620 * as setting the fork handler, has been completed. 621 */ 622 623 void 624 start_forked_proc(struct proc *p1, struct proc *p2) 625 { 626 /* 627 * Move from SIDL to RUN queue, and activate the process's thread. 628 * Activation of the thread effectively makes the process "a" 629 * current process, so we do not setrunqueue(). 630 */ 631 KASSERT(p2->p_stat == SIDL, 632 ("cannot start forked process, bad status: %p", p2)); 633 resetpriority(p2); 634 (void) splhigh(); 635 p2->p_stat = SRUN; 636 setrunqueue(p2); 637 (void) spl0(); 638 639 /* 640 * Now can be swapped. 641 */ 642 PRELE(p1); 643 644 /* 645 * Preserve synchronization semantics of vfork. If waiting for 646 * child to exec or exit, set P_PPWAIT on child, and sleep on our 647 * proc (in case of exit). 648 */ 649 while (p2->p_flag & P_PPWAIT) 650 tsleep(p1, 0, "ppwait", 0); 651 } 652 653