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 1.24 2004/04/24 04:32:03 drhodus 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(struct proc *p1, int flags, struct proc **procp) 179 { 180 struct proc *p2, *pptr; 181 uid_t uid; 182 struct proc *newproc; 183 int ok; 184 static int curfail = 0, pidchecked = 0; 185 static struct timeval lastfail; 186 struct forklist *ep; 187 struct filedesc_to_leader *fdtol; 188 189 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 190 return (EINVAL); 191 192 /* 193 * Here we don't create a new process, but we divorce 194 * certain parts of a process from itself. 195 */ 196 if ((flags & RFPROC) == 0) { 197 198 vm_fork(p1, 0, flags); 199 200 /* 201 * Close all file descriptors. 202 */ 203 if (flags & RFCFDG) { 204 struct filedesc *fdtmp; 205 fdtmp = fdinit(p1); 206 fdfree(p1); 207 p1->p_fd = fdtmp; 208 } 209 210 /* 211 * Unshare file descriptors (from parent.) 212 */ 213 if (flags & RFFDG) { 214 if (p1->p_fd->fd_refcnt > 1) { 215 struct filedesc *newfd; 216 newfd = fdcopy(p1); 217 fdfree(p1); 218 p1->p_fd = newfd; 219 } 220 } 221 *procp = NULL; 222 return (0); 223 } 224 225 /* 226 * Although process entries are dynamically created, we still keep 227 * a global limit on the maximum number we will create. Don't allow 228 * a nonprivileged user to use the last ten processes; don't let root 229 * exceed the limit. The variable nprocs is the current number of 230 * processes, maxproc is the limit. 231 */ 232 uid = p1->p_ucred->cr_ruid; 233 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 234 if (ppsratecheck(&lastfail, &curfail, 1)) 235 printf("maxproc limit exceeded by uid %d, please " 236 "see tuning(7) and login.conf(5).\n", uid); 237 tsleep(&forksleep, 0, "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_ucred->cr_ruidinfo, 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 if (ppsratecheck(&lastfail, &curfail, 1)) 258 printf("maxproc limit exceeded by uid %d, please " 259 "see tuning(7) and login.conf(5).\n", uid); 260 tsleep(&forksleep, 0, "fork", hz / 2); 261 return (EAGAIN); 262 } 263 264 /* Allocate new proc. */ 265 newproc = zalloc(proc_zone); 266 267 /* 268 * Setup linkage for kernel based threading 269 */ 270 if ((flags & RFTHREAD) != 0) { 271 newproc->p_peers = p1->p_peers; 272 p1->p_peers = newproc; 273 newproc->p_leader = p1->p_leader; 274 } else { 275 newproc->p_peers = 0; 276 newproc->p_leader = newproc; 277 } 278 279 newproc->p_wakeup = 0; 280 newproc->p_vmspace = NULL; 281 282 /* 283 * Find an unused process ID. We remember a range of unused IDs 284 * ready to use (from nextpid+1 through pidchecked-1). 285 */ 286 nextpid++; 287 if (randompid) 288 nextpid += arc4random() % randompid; 289 retry: 290 /* 291 * If the process ID prototype has wrapped around, 292 * restart somewhat above 0, as the low-numbered procs 293 * tend to include daemons that don't exit. 294 */ 295 if (nextpid >= PID_MAX) { 296 nextpid = nextpid % PID_MAX; 297 if (nextpid < 100) 298 nextpid += 100; 299 pidchecked = 0; 300 } 301 if (nextpid >= pidchecked) { 302 int doingzomb = 0; 303 304 pidchecked = PID_MAX; 305 /* 306 * Scan the active and zombie procs to check whether this pid 307 * is in use. Remember the lowest pid that's greater 308 * than nextpid, so we can avoid checking for a while. 309 */ 310 p2 = LIST_FIRST(&allproc); 311 again: 312 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) { 313 while (p2->p_pid == nextpid || 314 p2->p_pgrp->pg_id == nextpid || 315 p2->p_session->s_sid == nextpid) { 316 nextpid++; 317 if (nextpid >= pidchecked) 318 goto retry; 319 } 320 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 321 pidchecked = p2->p_pid; 322 if (p2->p_pgrp->pg_id > nextpid && 323 pidchecked > p2->p_pgrp->pg_id) 324 pidchecked = p2->p_pgrp->pg_id; 325 if (p2->p_session->s_sid > nextpid && 326 pidchecked > p2->p_session->s_sid) 327 pidchecked = p2->p_session->s_sid; 328 } 329 if (!doingzomb) { 330 doingzomb = 1; 331 p2 = LIST_FIRST(&zombproc); 332 goto again; 333 } 334 } 335 336 p2 = newproc; 337 p2->p_stat = SIDL; /* protect against others */ 338 p2->p_pid = nextpid; 339 LIST_INSERT_HEAD(&allproc, p2, p_list); 340 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 341 342 /* 343 * Make a proc table entry for the new process. 344 * Start by zeroing the section of proc that is zero-initialized, 345 * then copy the section that is copied directly from the parent. 346 */ 347 bzero(&p2->p_startzero, 348 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 349 bcopy(&p1->p_startcopy, &p2->p_startcopy, 350 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 351 352 p2->p_aioinfo = NULL; 353 354 /* 355 * Duplicate sub-structures as needed. 356 * Increase reference counts on shared objects. 357 * The p_stats and p_sigacts substructs are set in vm_fork. 358 * 359 * P_CP_RELEASED indicates that the process is starting out in 360 * the kernel (in the fork trampoline). The flag will be cleared 361 * when the new process calls userret() and acquires its current 362 * process designation for the return to userland. 363 */ 364 p2->p_flag = P_INMEM | P_CP_RELEASED; 365 if (p1->p_flag & P_PROFIL) 366 startprofclock(p2); 367 p2->p_ucred = crhold(p1->p_ucred); 368 369 if (p2->p_ucred->cr_prison) { 370 p2->p_ucred->cr_prison->pr_ref++; 371 p2->p_flag |= P_JAILED; 372 } 373 374 if (p2->p_args) 375 p2->p_args->ar_ref++; 376 377 if (flags & RFSIGSHARE) { 378 p2->p_procsig = p1->p_procsig; 379 p2->p_procsig->ps_refcnt++; 380 if (p1->p_sigacts == &p1->p_addr->u_sigacts) { 381 struct sigacts *newsigacts; 382 int s; 383 384 /* Create the shared sigacts structure */ 385 MALLOC(newsigacts, struct sigacts *, 386 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 387 s = splhigh(); 388 /* 389 * Set p_sigacts to the new shared structure. 390 * Note that this is updating p1->p_sigacts at the 391 * same time, since p_sigacts is just a pointer to 392 * the shared p_procsig->ps_sigacts. 393 */ 394 p2->p_sigacts = newsigacts; 395 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts, 396 sizeof(*p2->p_sigacts)); 397 *p2->p_sigacts = p1->p_addr->u_sigacts; 398 splx(s); 399 } 400 } else { 401 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig), 402 M_SUBPROC, M_WAITOK); 403 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 404 p2->p_procsig->ps_refcnt = 1; 405 p2->p_sigacts = NULL; /* finished in vm_fork() */ 406 } 407 if (flags & RFLINUXTHPN) 408 p2->p_sigparent = SIGUSR1; 409 else 410 p2->p_sigparent = SIGCHLD; 411 412 /* bump references to the text vnode (for procfs) */ 413 p2->p_textvp = p1->p_textvp; 414 if (p2->p_textvp) 415 vref(p2->p_textvp); 416 417 if (flags & RFCFDG) { 418 p2->p_fd = fdinit(p1); 419 fdtol = NULL; 420 } else if (flags & RFFDG) { 421 p2->p_fd = fdcopy(p1); 422 fdtol = NULL; 423 } else { 424 p2->p_fd = fdshare(p1); 425 if (p1->p_fdtol == NULL) 426 p1->p_fdtol = 427 filedesc_to_leader_alloc(NULL, 428 p1->p_leader); 429 if ((flags & RFTHREAD) != 0) { 430 /* 431 * Shared file descriptor table and 432 * shared process leaders. 433 */ 434 fdtol = p1->p_fdtol; 435 fdtol->fdl_refcount++; 436 } else { 437 /* 438 * Shared file descriptor table, and 439 * different process leaders 440 */ 441 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2); 442 } 443 } 444 p2->p_fdtol = fdtol; 445 446 /* 447 * If p_limit is still copy-on-write, bump refcnt, 448 * otherwise get a copy that won't be modified. 449 * (If PL_SHAREMOD is clear, the structure is shared 450 * copy-on-write.) 451 */ 452 if (p1->p_limit->p_lflags & PL_SHAREMOD) { 453 p2->p_limit = limcopy(p1->p_limit); 454 } else { 455 p2->p_limit = p1->p_limit; 456 p2->p_limit->p_refcnt++; 457 } 458 459 /* 460 * Preserve some more flags in subprocess. P_PROFIL has already 461 * been preserved. 462 */ 463 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 464 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 465 p2->p_flag |= P_CONTROLT; 466 if (flags & RFPPWAIT) 467 p2->p_flag |= P_PPWAIT; 468 469 LIST_INSERT_AFTER(p1, p2, p_pglist); 470 471 /* 472 * Attach the new process to its parent. 473 * 474 * If RFNOWAIT is set, the newly created process becomes a child 475 * of init. This effectively disassociates the child from the 476 * parent. 477 */ 478 if (flags & RFNOWAIT) 479 pptr = initproc; 480 else 481 pptr = p1; 482 p2->p_pptr = pptr; 483 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 484 LIST_INIT(&p2->p_children); 485 varsymset_init(&p2->p_varsymset, &p1->p_varsymset); 486 487 #ifdef KTRACE 488 /* 489 * Copy traceflag and tracefile if enabled. If not inherited, 490 * these were zeroed above but we still could have a trace race 491 * so make sure p2's p_tracep is NULL. 492 */ 493 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) { 494 p2->p_traceflag = p1->p_traceflag; 495 if ((p2->p_tracep = p1->p_tracep) != NULL) 496 vref(p2->p_tracep); 497 } 498 #endif 499 500 /* 501 * Give the child process an estcpu skewed towards the batch side 502 * of the parent. This prevents batch programs from glitching 503 * interactive programs when they are first started. If the child 504 * is not a batch program it's priority will be corrected by the 505 * scheduler. 506 */ 507 p2->p_estcpu_fork = p2->p_estcpu = 508 ESTCPULIM(p1->p_estcpu + ESTCPURAMP); 509 510 /* 511 * This begins the section where we must prevent the parent 512 * from being swapped. 513 */ 514 PHOLD(p1); 515 516 /* 517 * Finish creating the child process. It will return via a different 518 * execution path later. (ie: directly into user mode) 519 */ 520 vm_fork(p1, p2, flags); 521 caps_fork(p1, p2, flags); 522 523 if (flags == (RFFDG | RFPROC)) { 524 mycpu->gd_cnt.v_forks++; 525 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 526 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 527 mycpu->gd_cnt.v_vforks++; 528 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 529 } else if (p1 == &proc0) { 530 mycpu->gd_cnt.v_kthreads++; 531 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 532 } else { 533 mycpu->gd_cnt.v_rforks++; 534 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize; 535 } 536 537 /* 538 * Both processes are set up, now check if any loadable modules want 539 * to adjust anything. 540 * What if they have an error? XXX 541 */ 542 TAILQ_FOREACH(ep, &fork_list, next) { 543 (*ep->function)(p1, p2, flags); 544 } 545 546 /* 547 * Make child runnable and add to run queue. 548 */ 549 microtime(&(p2->p_stats->p_start)); 550 p2->p_acflag = AFORK; 551 552 /* 553 * tell any interested parties about the new process 554 */ 555 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 556 557 /* 558 * Return child proc pointer to parent. 559 */ 560 *procp = p2; 561 return (0); 562 } 563 564 /* 565 * The next two functionms are general routines to handle adding/deleting 566 * items on the fork callout list. 567 * 568 * at_fork(): 569 * Take the arguments given and put them onto the fork callout list, 570 * However first make sure that it's not already there. 571 * Returns 0 on success or a standard error number. 572 */ 573 int 574 at_fork(forklist_fn function) 575 { 576 struct forklist *ep; 577 578 #ifdef INVARIANTS 579 /* let the programmer know if he's been stupid */ 580 if (rm_at_fork(function)) { 581 printf("WARNING: fork callout entry (%p) already present\n", 582 function); 583 } 584 #endif 585 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO); 586 ep->function = function; 587 TAILQ_INSERT_TAIL(&fork_list, ep, next); 588 return (0); 589 } 590 591 /* 592 * Scan the exit callout list for the given item and remove it.. 593 * Returns the number of items removed (0 or 1) 594 */ 595 int 596 rm_at_fork(forklist_fn function) 597 { 598 struct forklist *ep; 599 600 TAILQ_FOREACH(ep, &fork_list, next) { 601 if (ep->function == function) { 602 TAILQ_REMOVE(&fork_list, ep, next); 603 free(ep, M_ATFORK); 604 return(1); 605 } 606 } 607 return (0); 608 } 609 610 /* 611 * Add a forked process to the run queue after any remaining setup, such 612 * as setting the fork handler, has been completed. 613 */ 614 void 615 start_forked_proc(struct proc *p1, struct proc *p2) 616 { 617 /* 618 * Move from SIDL to RUN queue, and activate the process's thread. 619 * Activation of the thread effectively makes the process "a" 620 * current process, so we do not setrunqueue(). 621 */ 622 KASSERT(p2->p_stat == SIDL, 623 ("cannot start forked process, bad status: %p", p2)); 624 resetpriority(p2); 625 (void) splhigh(); 626 p2->p_stat = SRUN; 627 setrunqueue(p2); 628 (void) spl0(); 629 630 /* 631 * Now can be swapped. 632 */ 633 PRELE(p1); 634 635 /* 636 * Preserve synchronization semantics of vfork. If waiting for 637 * child to exec or exit, set P_PPWAIT on child, and sleep on our 638 * proc (in case of exit). 639 */ 640 while (p2->p_flag & P_PPWAIT) 641 tsleep(p1, 0, "ppwait", 0); 642 } 643 644