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 1.17 2003/11/27 19:57:37 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 int forksleep; /* Place for fork1() to sleep on. */ 85 86 /* ARGSUSED */ 87 int 88 fork(struct fork_args *uap) 89 { 90 struct proc *p = curproc; 91 struct proc *p2; 92 int error; 93 94 error = fork1(p, RFFDG | RFPROC, &p2); 95 if (error == 0) { 96 start_forked_proc(p, p2); 97 uap->sysmsg_fds[0] = p2->p_pid; 98 uap->sysmsg_fds[1] = 0; 99 } 100 return error; 101 } 102 103 /* ARGSUSED */ 104 int 105 vfork(struct vfork_args *uap) 106 { 107 struct proc *p = curproc; 108 struct proc *p2; 109 int error; 110 111 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2); 112 if (error == 0) { 113 start_forked_proc(p, p2); 114 uap->sysmsg_fds[0] = p2->p_pid; 115 uap->sysmsg_fds[1] = 0; 116 } 117 return error; 118 } 119 120 int 121 rfork(struct rfork_args *uap) 122 { 123 struct proc *p = curproc; 124 struct proc *p2; 125 int error; 126 127 /* Don't allow kernel only flags */ 128 if ((uap->flags & RFKERNELONLY) != 0) 129 return (EINVAL); 130 131 error = fork1(p, uap->flags, &p2); 132 if (error == 0) { 133 start_forked_proc(p, p2); 134 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0; 135 uap->sysmsg_fds[1] = 0; 136 } 137 return error; 138 } 139 140 141 int nprocs = 1; /* process 0 */ 142 static int nextpid = 0; 143 144 /* 145 * Random component to nextpid generation. We mix in a random factor to make 146 * it a little harder to predict. We sanity check the modulus value to avoid 147 * doing it in critical paths. Don't let it be too small or we pointlessly 148 * waste randomness entropy, and don't let it be impossibly large. Using a 149 * modulus that is too big causes a LOT more process table scans and slows 150 * down fork processing as the pidchecked caching is defeated. 151 */ 152 static int randompid = 0; 153 154 static int 155 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 156 { 157 int error, pid; 158 159 pid = randompid; 160 error = sysctl_handle_int(oidp, &pid, 0, req); 161 if (error || !req->newptr) 162 return (error); 163 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 164 pid = PID_MAX - 100; 165 else if (pid < 2) /* NOP */ 166 pid = 0; 167 else if (pid < 100) /* Make it reasonable */ 168 pid = 100; 169 randompid = pid; 170 return (error); 171 } 172 173 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 174 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 175 176 int 177 fork1(p1, flags, procp) 178 struct proc *p1; 179 int flags; 180 struct proc **procp; 181 { 182 struct proc *p2, *pptr; 183 uid_t uid; 184 struct proc *newproc; 185 int ok; 186 static int pidchecked = 0; 187 struct forklist *ep; 188 struct filedesc_to_leader *fdtol; 189 190 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 191 return (EINVAL); 192 193 /* 194 * Here we don't create a new process, but we divorce 195 * certain parts of a process from itself. 196 */ 197 if ((flags & RFPROC) == 0) { 198 199 vm_fork(p1, 0, flags); 200 201 /* 202 * Close all file descriptors. 203 */ 204 if (flags & RFCFDG) { 205 struct filedesc *fdtmp; 206 fdtmp = fdinit(p1); 207 fdfree(p1); 208 p1->p_fd = fdtmp; 209 } 210 211 /* 212 * Unshare file descriptors (from parent.) 213 */ 214 if (flags & RFFDG) { 215 if (p1->p_fd->fd_refcnt > 1) { 216 struct filedesc *newfd; 217 newfd = fdcopy(p1); 218 fdfree(p1); 219 p1->p_fd = newfd; 220 } 221 } 222 *procp = NULL; 223 return (0); 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 tsleep(&forksleep, 0, "fork", hz / 2); 236 return (EAGAIN); 237 } 238 /* 239 * Increment the nprocs resource before blocking can occur. There 240 * are hard-limits as to the number of processes that can run. 241 */ 242 nprocs++; 243 244 /* 245 * Increment the count of procs running with this uid. Don't allow 246 * a nonprivileged user to exceed their current limit. 247 */ 248 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1, 249 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 250 if (!ok) { 251 /* 252 * Back out the process count 253 */ 254 nprocs--; 255 tsleep(&forksleep, 0, "fork", hz / 2); 256 return (EAGAIN); 257 } 258 259 /* Allocate new proc. */ 260 newproc = zalloc(proc_zone); 261 262 /* 263 * Setup linkage for kernel based threading 264 */ 265 if((flags & RFTHREAD) != 0) { 266 newproc->p_peers = p1->p_peers; 267 p1->p_peers = newproc; 268 newproc->p_leader = p1->p_leader; 269 } else { 270 newproc->p_peers = 0; 271 newproc->p_leader = newproc; 272 } 273 274 newproc->p_wakeup = 0; 275 newproc->p_vmspace = NULL; 276 277 /* 278 * Find an unused process ID. We remember a range of unused IDs 279 * ready to use (from nextpid+1 through pidchecked-1). 280 */ 281 nextpid++; 282 if (randompid) 283 nextpid += arc4random() % randompid; 284 retry: 285 /* 286 * If the process ID prototype has wrapped around, 287 * restart somewhat above 0, as the low-numbered procs 288 * tend to include daemons that don't exit. 289 */ 290 if (nextpid >= PID_MAX) { 291 nextpid = nextpid % PID_MAX; 292 if (nextpid < 100) 293 nextpid += 100; 294 pidchecked = 0; 295 } 296 if (nextpid >= pidchecked) { 297 int doingzomb = 0; 298 299 pidchecked = PID_MAX; 300 /* 301 * Scan the active and zombie procs to check whether this pid 302 * is in use. Remember the lowest pid that's greater 303 * than nextpid, so we can avoid checking for a while. 304 */ 305 p2 = LIST_FIRST(&allproc); 306 again: 307 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 308 while (p2->p_pid == nextpid || 309 p2->p_pgrp->pg_id == nextpid || 310 p2->p_session->s_sid == nextpid) { 311 nextpid++; 312 if (nextpid >= pidchecked) 313 goto retry; 314 } 315 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 316 pidchecked = p2->p_pid; 317 if (p2->p_pgrp->pg_id > nextpid && 318 pidchecked > p2->p_pgrp->pg_id) 319 pidchecked = p2->p_pgrp->pg_id; 320 if (p2->p_session->s_sid > nextpid && 321 pidchecked > p2->p_session->s_sid) 322 pidchecked = p2->p_session->s_sid; 323 } 324 if (!doingzomb) { 325 doingzomb = 1; 326 p2 = LIST_FIRST(&zombproc); 327 goto again; 328 } 329 } 330 331 p2 = newproc; 332 p2->p_stat = SIDL; /* protect against others */ 333 p2->p_pid = nextpid; 334 LIST_INSERT_HEAD(&allproc, p2, p_list); 335 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 336 337 /* 338 * Make a proc table entry for the new process. 339 * Start by zeroing the section of proc that is zero-initialized, 340 * then copy the section that is copied directly from the parent. 341 */ 342 bzero(&p2->p_startzero, 343 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 344 bcopy(&p1->p_startcopy, &p2->p_startcopy, 345 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 346 347 p2->p_aioinfo = NULL; 348 349 /* 350 * Duplicate sub-structures as needed. 351 * Increase reference counts on shared objects. 352 * The p_stats and p_sigacts substructs are set in vm_fork. 353 * 354 * P_CP_RELEASED indicates that the process is starting out in 355 * the kernel (in the fork trampoline). The flag will be converted 356 * to P_CURPROC when the new process calls userret() and attempts 357 * to return to userland 358 */ 359 p2->p_flag = P_INMEM | P_CP_RELEASED; 360 if (p1->p_flag & P_PROFIL) 361 startprofclock(p2); 362 p2->p_ucred = crhold(p1->p_ucred); 363 364 if (p2->p_ucred->cr_prison) { 365 p2->p_ucred->cr_prison->pr_ref++; 366 p2->p_flag |= P_JAILED; 367 } 368 369 if (p2->p_args) 370 p2->p_args->ar_ref++; 371 372 if (flags & RFSIGSHARE) { 373 p2->p_procsig = p1->p_procsig; 374 p2->p_procsig->ps_refcnt++; 375 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 376 struct sigacts *newsigacts; 377 int s; 378 379 /* Create the shared sigacts structure */ 380 MALLOC(newsigacts, struct sigacts *, 381 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 382 s = splhigh(); 383 /* 384 * Set p_sigacts to the new shared structure. 385 * Note that this is updating p1->p_sigacts at the 386 * same time, since p_sigacts is just a pointer to 387 * the shared p_procsig->ps_sigacts. 388 */ 389 p2->p_sigacts = newsigacts; 390 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 391 sizeof(*p2->p_sigacts)); 392 *p2->p_sigacts = p1->p_addr->u_sigacts; 393 splx(s); 394 } 395 } else { 396 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 397 M_SUBPROC, M_WAITOK); 398 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 399 p2->p_procsig->ps_refcnt = 1; 400 p2->p_sigacts = NULL; /* finished in vm_fork() */ 401 } 402 if (flags & RFLINUXTHPN) 403 p2->p_sigparent = SIGUSR1; 404 else 405 p2->p_sigparent = SIGCHLD; 406 407 /* bump references to the text vnode (for procfs) */ 408 p2->p_textvp = p1->p_textvp; 409 if (p2->p_textvp) 410 VREF(p2->p_textvp); 411 412 if (flags & RFCFDG) { 413 p2->p_fd = fdinit(p1); 414 fdtol = NULL; 415 } else if (flags & RFFDG) { 416 p2->p_fd = fdcopy(p1); 417 fdtol = NULL; 418 } else { 419 p2->p_fd = fdshare(p1); 420 if (p1->p_fdtol == NULL) 421 p1->p_fdtol = 422 filedesc_to_leader_alloc(NULL, 423 p1->p_leader); 424 if ((flags & RFTHREAD) != 0) { 425 /* 426 * Shared file descriptor table and 427 * shared process leaders. 428 */ 429 fdtol = p1->p_fdtol; 430 fdtol->fdl_refcount++; 431 } else { 432 /* 433 * Shared file descriptor table, and 434 * different process leaders 435 */ 436 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 437 } 438 } 439 p2->p_fdtol = fdtol; 440 441 /* 442 * If p_limit is still copy-on-write, bump refcnt, 443 * otherwise get a copy that won't be modified. 444 * (If PL_SHAREMOD is clear, the structure is shared 445 * copy-on-write.) 446 */ 447 if (p1->p_limit->p_lflags & PL_SHAREMOD) 448 p2->p_limit = limcopy(p1->p_limit); 449 else { 450 p2->p_limit = p1->p_limit; 451 p2->p_limit->p_refcnt++; 452 } 453 454 /* 455 * Preserve some more flags in subprocess. P_PROFIL has already 456 * been preserved. 457 */ 458 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 459 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 460 p2->p_flag |= P_CONTROLT; 461 if (flags & RFPPWAIT) 462 p2->p_flag |= P_PPWAIT; 463 464 LIST_INSERT_AFTER(p1, p2, p_pglist); 465 466 /* 467 * Attach the new process to its parent. 468 * 469 * If RFNOWAIT is set, the newly created process becomes a child 470 * of init. This effectively disassociates the child from the 471 * parent. 472 */ 473 if (flags & RFNOWAIT) 474 pptr = initproc; 475 else 476 pptr = p1; 477 p2->p_pptr = pptr; 478 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 479 LIST_INIT(&p2->p_children); 480 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 481 482 #ifdef KTRACE 483 /* 484 * Copy traceflag and tracefile if enabled. If not inherited, 485 * these were zeroed above but we still could have a trace race 486 * so make sure p2's p_tracep is NULL. 487 */ 488 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 489 p2->p_traceflag = p1->p_traceflag; 490 if ((p2->p_tracep = p1->p_tracep) != NULL) 491 VREF(p2->p_tracep); 492 } 493 #endif 494 495 /* 496 * set priority of child to be that of parent 497 */ 498 p2->p_estcpu = p1->p_estcpu; 499 500 /* 501 * This begins the section where we must prevent the parent 502 * from being swapped. 503 */ 504 PHOLD(p1); 505 506 /* 507 * Finish creating the child process. It will return via a different 508 * execution path later. (ie: directly into user mode) 509 */ 510 vm_fork(p1, p2, flags); 511 512 if (flags == (RFFDG | RFPROC)) { 513 mycpu->gd_cnt.v_forks++; 514 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 515 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 516 mycpu->gd_cnt.v_vforks++; 517 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 518 } else if (p1 == &proc0) { 519 mycpu->gd_cnt.v_kthreads++; 520 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 521 } else { 522 mycpu->gd_cnt.v_rforks++; 523 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 524 } 525 526 /* 527 * Both processes are set up, now check if any loadable modules want 528 * to adjust anything. 529 * What if they have an error? XXX 530 */ 531 TAILQ_FOREACH(ep, &fork_list, next) { 532 (*ep->function)(p1, p2, flags); 533 } 534 535 /* 536 * Make child runnable and add to run queue. 537 */ 538 microtime(&(p2->p_stats->p_start)); 539 p2->p_acflag = AFORK; 540 541 /* 542 * tell any interested parties about the new process 543 */ 544 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 545 546 /* 547 * Return child proc pointer to parent. 548 */ 549 *procp = p2; 550 return (0); 551 } 552 553 /* 554 * The next two functionms are general routines to handle adding/deleting 555 * items on the fork callout list. 556 * 557 * at_fork(): 558 * Take the arguments given and put them onto the fork callout list, 559 * However first make sure that it's not already there. 560 * Returns 0 on success or a standard error number. 561 */ 562 563 int 564 at_fork(function) 565 forklist_fn function; 566 { 567 struct forklist *ep; 568 569 #ifdef INVARIANTS 570 /* let the programmer know if he's been stupid */ 571 if (rm_at_fork(function)) 572 printf("WARNING: fork callout entry (%p) already present\n", 573 function); 574 #endif 575 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT); 576 if (ep == NULL) 577 return (ENOMEM); 578 ep->function = function; 579 TAILQ_INSERT_TAIL(&fork_list, ep, next); 580 return (0); 581 } 582 583 /* 584 * Scan the exit callout list for the given item and remove it.. 585 * Returns the number of items removed (0 or 1) 586 */ 587 588 int 589 rm_at_fork(function) 590 forklist_fn function; 591 { 592 struct forklist *ep; 593 594 TAILQ_FOREACH(ep, &fork_list, next) { 595 if (ep->function == function) { 596 TAILQ_REMOVE(&fork_list, ep, next); 597 free(ep, M_ATFORK); 598 return(1); 599 } 600 } 601 return (0); 602 } 603 604 /* 605 * Add a forked process to the run queue after any remaining setup, such 606 * as setting the fork handler, has been completed. 607 */ 608 609 void 610 start_forked_proc(struct proc *p1, struct proc *p2) 611 { 612 /* 613 * Move from SIDL to RUN queue, and activate the process's thread. 614 * Activation of the thread effectively makes the process "a" 615 * current process, so we do not setrunqueue(). 616 */ 617 KASSERT(p2->p_stat == SIDL, 618 ("cannot start forked process, bad status: %p", p2)); 619 (void) splhigh(); 620 p2->p_stat = SRUN; 621 setrunqueue(p2); 622 (void) spl0(); 623 624 /* 625 * Now can be swapped. 626 */ 627 PRELE(p1); 628 629 /* 630 * Preserve synchronization semantics of vfork. If waiting for 631 * child to exec or exit, set P_PPWAIT on child, and sleep on our 632 * proc (in case of exit). 633 */ 634 while (p2->p_flag & P_PPWAIT) 635 tsleep(p1, 0, "ppwait", 0); 636 } 637 638