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 1.72.2.13 2003/06/06 20:21:32 tegge Exp $ 40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.2 2003/06/17 04:28:41 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 60 #include <vm/vm.h> 61 #include <sys/lock.h> 62 #include <vm/pmap.h> 63 #include <vm/vm_map.h> 64 #include <vm/vm_extern.h> 65 #include <vm/vm_zone.h> 66 67 #include <sys/vmmeter.h> 68 #include <sys/user.h> 69 70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 71 72 /* 73 * These are the stuctures used to create a callout list for things to do 74 * when forking a process 75 */ 76 struct forklist { 77 forklist_fn function; 78 TAILQ_ENTRY(forklist) next; 79 }; 80 81 TAILQ_HEAD(forklist_head, forklist); 82 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 83 84 #ifndef _SYS_SYSPROTO_H_ 85 struct fork_args { 86 int dummy; 87 }; 88 #endif 89 90 int forksleep; /* Place for fork1() to sleep on. */ 91 92 /* ARGSUSED */ 93 int 94 fork(p, uap) 95 struct proc *p; 96 struct fork_args *uap; 97 { 98 int error; 99 struct proc *p2; 100 101 error = fork1(p, RFFDG | RFPROC, &p2); 102 if (error == 0) { 103 p->p_retval[0] = p2->p_pid; 104 p->p_retval[1] = 0; 105 } 106 return error; 107 } 108 109 /* ARGSUSED */ 110 int 111 vfork(p, uap) 112 struct proc *p; 113 struct vfork_args *uap; 114 { 115 int error; 116 struct proc *p2; 117 118 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2); 119 if (error == 0) { 120 p->p_retval[0] = p2->p_pid; 121 p->p_retval[1] = 0; 122 } 123 return error; 124 } 125 126 int 127 rfork(p, uap) 128 struct proc *p; 129 struct rfork_args *uap; 130 { 131 int error; 132 struct proc *p2; 133 134 error = fork1(p, uap->flags, &p2); 135 if (error == 0) { 136 p->p_retval[0] = p2 ? p2->p_pid : 0; 137 p->p_retval[1] = 0; 138 } 139 return error; 140 } 141 142 143 int nprocs = 1; /* process 0 */ 144 static int nextpid = 0; 145 146 /* 147 * Random component to nextpid generation. We mix in a random factor to make 148 * it a little harder to predict. We sanity check the modulus value to avoid 149 * doing it in critical paths. Don't let it be too small or we pointlessly 150 * waste randomness entropy, and don't let it be impossibly large. Using a 151 * modulus that is too big causes a LOT more process table scans and slows 152 * down fork processing as the pidchecked caching is defeated. 153 */ 154 static int randompid = 0; 155 156 static int 157 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 158 { 159 int error, pid; 160 161 pid = randompid; 162 error = sysctl_handle_int(oidp, &pid, 0, req); 163 if (error || !req->newptr) 164 return (error); 165 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 166 pid = PID_MAX - 100; 167 else if (pid < 2) /* NOP */ 168 pid = 0; 169 else if (pid < 100) /* Make it reasonable */ 170 pid = 100; 171 randompid = pid; 172 return (error); 173 } 174 175 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 176 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 177 178 int 179 fork1(p1, flags, procp) 180 struct proc *p1; 181 int flags; 182 struct proc **procp; 183 { 184 struct proc *p2, *pptr; 185 uid_t uid; 186 struct proc *newproc; 187 int ok; 188 static int pidchecked = 0; 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_cred->p_ruid; 236 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 237 tsleep(&forksleep, PUSER, "fork", hz / 2); 238 return (EAGAIN); 239 } 240 /* 241 * Increment the nprocs resource before blocking can occur. There 242 * are hard-limits as to the number of processes that can run. 243 */ 244 nprocs++; 245 246 /* 247 * Increment the count of procs running with this uid. Don't allow 248 * a nonprivileged user to exceed their current limit. 249 */ 250 ok = chgproccnt(p1->p_cred->p_uidinfo, 1, 251 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 252 if (!ok) { 253 /* 254 * Back out the process count 255 */ 256 nprocs--; 257 tsleep(&forksleep, PUSER, "fork", hz / 2); 258 return (EAGAIN); 259 } 260 261 /* Allocate new proc. */ 262 newproc = zalloc(proc_zone); 263 264 /* 265 * Setup linkage for kernel based threading 266 */ 267 if((flags & RFTHREAD) != 0) { 268 newproc->p_peers = p1->p_peers; 269 p1->p_peers = newproc; 270 newproc->p_leader = p1->p_leader; 271 } else { 272 newproc->p_peers = 0; 273 newproc->p_leader = newproc; 274 } 275 276 newproc->p_wakeup = 0; 277 278 newproc->p_vmspace = NULL; 279 280 /* 281 * Find an unused process ID. We remember a range of unused IDs 282 * ready to use (from nextpid+1 through pidchecked-1). 283 */ 284 nextpid++; 285 if (randompid) 286 nextpid += arc4random() % randompid; 287 retry: 288 /* 289 * If the process ID prototype has wrapped around, 290 * restart somewhat above 0, as the low-numbered procs 291 * tend to include daemons that don't exit. 292 */ 293 if (nextpid >= PID_MAX) { 294 nextpid = nextpid % PID_MAX; 295 if (nextpid < 100) 296 nextpid += 100; 297 pidchecked = 0; 298 } 299 if (nextpid >= pidchecked) { 300 int doingzomb = 0; 301 302 pidchecked = PID_MAX; 303 /* 304 * Scan the active and zombie procs to check whether this pid 305 * is in use. Remember the lowest pid that's greater 306 * than nextpid, so we can avoid checking for a while. 307 */ 308 p2 = LIST_FIRST(&allproc); 309 again: 310 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 311 while (p2->p_pid == nextpid || 312 p2->p_pgrp->pg_id == nextpid || 313 p2->p_session->s_sid == nextpid) { 314 nextpid++; 315 if (nextpid >= pidchecked) 316 goto retry; 317 } 318 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 319 pidchecked = p2->p_pid; 320 if (p2->p_pgrp->pg_id > nextpid && 321 pidchecked > p2->p_pgrp->pg_id) 322 pidchecked = p2->p_pgrp->pg_id; 323 if (p2->p_session->s_sid > nextpid && 324 pidchecked > p2->p_session->s_sid) 325 pidchecked = p2->p_session->s_sid; 326 } 327 if (!doingzomb) { 328 doingzomb = 1; 329 p2 = LIST_FIRST(&zombproc); 330 goto again; 331 } 332 } 333 334 p2 = newproc; 335 p2->p_stat = SIDL; /* protect against others */ 336 p2->p_pid = nextpid; 337 LIST_INSERT_HEAD(&allproc, p2, p_list); 338 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 339 340 /* 341 * Make a proc table entry for the new process. 342 * Start by zeroing the section of proc that is zero-initialized, 343 * then copy the section that is copied directly from the parent. 344 */ 345 bzero(&p2->p_startzero, 346 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 347 bcopy(&p1->p_startcopy, &p2->p_startcopy, 348 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 349 350 p2->p_aioinfo = NULL; 351 352 /* 353 * Duplicate sub-structures as needed. 354 * Increase reference counts on shared objects. 355 * The p_stats and p_sigacts substructs are set in vm_fork. 356 */ 357 p2->p_flag = P_INMEM; 358 if (p1->p_flag & P_PROFIL) 359 startprofclock(p2); 360 MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred), 361 M_SUBPROC, M_WAITOK); 362 bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred)); 363 p2->p_cred->p_refcnt = 1; 364 crhold(p1->p_ucred); 365 uihold(p1->p_cred->p_uidinfo); 366 367 if (p2->p_prison) { 368 p2->p_prison->pr_ref++; 369 p2->p_flag |= P_JAILED; 370 } 371 372 if (p2->p_args) 373 p2->p_args->ar_ref++; 374 375 if (flags & RFSIGSHARE) { 376 p2->p_procsig = p1->p_procsig; 377 p2->p_procsig->ps_refcnt++; 378 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 379 struct sigacts *newsigacts; 380 int s; 381 382 /* Create the shared sigacts structure */ 383 MALLOC(newsigacts, struct sigacts *, 384 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 385 s = splhigh(); 386 /* 387 * Set p_sigacts to the new shared structure. 388 * Note that this is updating p1->p_sigacts at the 389 * same time, since p_sigacts is just a pointer to 390 * the shared p_procsig->ps_sigacts. 391 */ 392 p2->p_sigacts = newsigacts; 393 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 394 sizeof(*p2->p_sigacts)); 395 *p2->p_sigacts = p1->p_addr->u_sigacts; 396 splx(s); 397 } 398 } else { 399 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 400 M_SUBPROC, M_WAITOK); 401 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 402 p2->p_procsig->ps_refcnt = 1; 403 p2->p_sigacts = NULL; /* finished in vm_fork() */ 404 } 405 if (flags & RFLINUXTHPN) 406 p2->p_sigparent = SIGUSR1; 407 else 408 p2->p_sigparent = SIGCHLD; 409 410 /* bump references to the text vnode (for procfs) */ 411 p2->p_textvp = p1->p_textvp; 412 if (p2->p_textvp) 413 VREF(p2->p_textvp); 414 415 if (flags & RFCFDG) { 416 p2->p_fd = fdinit(p1); 417 fdtol = NULL; 418 } else if (flags & RFFDG) { 419 p2->p_fd = fdcopy(p1); 420 fdtol = NULL; 421 } else { 422 p2->p_fd = fdshare(p1); 423 if (p1->p_fdtol == NULL) 424 p1->p_fdtol = 425 filedesc_to_leader_alloc(NULL, 426 p1->p_leader); 427 if ((flags & RFTHREAD) != 0) { 428 /* 429 * Shared file descriptor table and 430 * shared process leaders. 431 */ 432 fdtol = p1->p_fdtol; 433 fdtol->fdl_refcount++; 434 } else { 435 /* 436 * Shared file descriptor table, and 437 * different process leaders 438 */ 439 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 440 p2); 441 } 442 } 443 p2->p_fdtol = fdtol; 444 445 /* 446 * If p_limit is still copy-on-write, bump refcnt, 447 * otherwise get a copy that won't be modified. 448 * (If PL_SHAREMOD is clear, the structure is shared 449 * copy-on-write.) 450 */ 451 if (p1->p_limit->p_lflags & PL_SHAREMOD) 452 p2->p_limit = limcopy(p1->p_limit); 453 else { 454 p2->p_limit = p1->p_limit; 455 p2->p_limit->p_refcnt++; 456 } 457 458 /* 459 * Preserve some more flags in subprocess. P_PROFIL has already 460 * been preserved. 461 */ 462 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 463 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 464 p2->p_flag |= P_CONTROLT; 465 if (flags & RFPPWAIT) 466 p2->p_flag |= P_PPWAIT; 467 468 LIST_INSERT_AFTER(p1, p2, p_pglist); 469 470 /* 471 * Attach the new process to its parent. 472 * 473 * If RFNOWAIT is set, the newly created process becomes a child 474 * of init. This effectively disassociates the child from the 475 * parent. 476 */ 477 if (flags & RFNOWAIT) 478 pptr = initproc; 479 else 480 pptr = p1; 481 p2->p_pptr = pptr; 482 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 483 LIST_INIT(&p2->p_children); 484 485 #ifdef KTRACE 486 /* 487 * Copy traceflag and tracefile if enabled. If not inherited, 488 * these were zeroed above but we still could have a trace race 489 * so make sure p2's p_tracep is NULL. 490 */ 491 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 492 p2->p_traceflag = p1->p_traceflag; 493 if ((p2->p_tracep = p1->p_tracep) != NULL) 494 VREF(p2->p_tracep); 495 } 496 #endif 497 498 /* 499 * set priority of child to be that of parent 500 */ 501 p2->p_estcpu = p1->p_estcpu; 502 503 /* 504 * This begins the section where we must prevent the parent 505 * from being swapped. 506 */ 507 PHOLD(p1); 508 509 /* 510 * Finish creating the child process. It will return via a different 511 * execution path later. (ie: directly into user mode) 512 */ 513 vm_fork(p1, p2, flags); 514 515 if (flags == (RFFDG | RFPROC)) { 516 cnt.v_forks++; 517 cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 518 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 519 cnt.v_vforks++; 520 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 521 } else if (p1 == &proc0) { 522 cnt.v_kthreads++; 523 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 524 } else { 525 cnt.v_rforks++; 526 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 527 } 528 529 /* 530 * Both processes are set up, now check if any loadable modules want 531 * to adjust anything. 532 * What if they have an error? XXX 533 */ 534 TAILQ_FOREACH(ep, &fork_list, next) { 535 (*ep->function)(p1, p2, flags); 536 } 537 538 /* 539 * Make child runnable and add to run queue. 540 */ 541 microtime(&(p2->p_stats->p_start)); 542 p2->p_acflag = AFORK; 543 (void) splhigh(); 544 p2->p_stat = SRUN; 545 setrunqueue(p2); 546 (void) spl0(); 547 548 /* 549 * Now can be swapped. 550 */ 551 PRELE(p1); 552 553 /* 554 * tell any interested parties about the new process 555 */ 556 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 557 558 /* 559 * Preserve synchronization semantics of vfork. If waiting for 560 * child to exec or exit, set P_PPWAIT on child, and sleep on our 561 * proc (in case of exit). 562 */ 563 while (p2->p_flag & P_PPWAIT) 564 tsleep(p1, PWAIT, "ppwait", 0); 565 566 /* 567 * Return child proc pointer to parent. 568 */ 569 *procp = p2; 570 return (0); 571 } 572 573 /* 574 * The next two functionms are general routines to handle adding/deleting 575 * items on the fork callout list. 576 * 577 * at_fork(): 578 * Take the arguments given and put them onto the fork callout list, 579 * However first make sure that it's not already there. 580 * Returns 0 on success or a standard error number. 581 */ 582 583 int 584 at_fork(function) 585 forklist_fn function; 586 { 587 struct forklist *ep; 588 589 #ifdef INVARIANTS 590 /* let the programmer know if he's been stupid */ 591 if (rm_at_fork(function)) 592 printf("WARNING: fork callout entry (%p) already present\n", 593 function); 594 #endif 595 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT); 596 if (ep == NULL) 597 return (ENOMEM); 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 */ 607 608 int 609 rm_at_fork(function) 610 forklist_fn function; 611 { 612 struct forklist *ep; 613 614 TAILQ_FOREACH(ep, &fork_list, next) { 615 if (ep->function == function) { 616 TAILQ_REMOVE(&fork_list, ep, next); 617 free(ep, M_ATFORK); 618 return(1); 619 } 620 } 621 return (0); 622 } 623