1 /* $NetBSD: kern_sysctl.c,v 1.271 2024/09/08 09:36:51 rillig Exp $ */ 2 3 /*- 4 * Copyright (c) 2003, 2007, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Brown. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1982, 1986, 1989, 1993 34 * The Regents of the University of California. All rights reserved. 35 * 36 * This code is derived from software contributed to Berkeley by 37 * Mike Karels at Berkeley Software Design, Inc. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * @(#)kern_sysctl.c 8.9 (Berkeley) 5/20/95 64 */ 65 66 /* 67 * sysctl system call. 68 */ 69 70 #define __COMPAT_SYSCTL 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: kern_sysctl.c,v 1.271 2024/09/08 09:36:51 rillig Exp $"); 74 75 #ifdef _KERNEL_OPT 76 #include "opt_defcorename.h" 77 #endif 78 79 #include "ksyms.h" 80 81 #include <sys/param.h> 82 #include <sys/types.h> 83 84 #include <sys/buf.h> 85 #include <sys/cprng.h> 86 #include <sys/kauth.h> 87 #include <sys/ksyms.h> 88 #include <sys/ktrace.h> 89 #include <sys/malloc.h> 90 #include <sys/mount.h> 91 #include <sys/once.h> 92 #include <sys/rndsource.h> 93 #include <sys/syscallargs.h> 94 #include <sys/sysctl.h> 95 #include <sys/systm.h> 96 97 #include <crypto/blake2/blake2s.h> 98 99 #define MAXDESCLEN 1024 100 MALLOC_DEFINE(M_SYSCTLNODE, "sysctlnode", "sysctl node structures"); 101 MALLOC_DEFINE(M_SYSCTLDATA, "sysctldata", "misc sysctl data"); 102 103 static int sysctl_mmap(SYSCTLFN_PROTO); 104 static int sysctl_alloc(struct sysctlnode *, int); 105 static int sysctl_realloc(struct sysctlnode *); 106 107 static int sysctl_cvt_in(struct lwp *, int *, const void *, size_t, 108 struct sysctlnode *); 109 static int sysctl_cvt_out(struct lwp *, int, const struct sysctlnode *, 110 void *, size_t, size_t *); 111 112 static int sysctl_log_add(struct sysctllog **, const struct sysctlnode *); 113 static int sysctl_log_realloc(struct sysctllog *); 114 115 typedef void sysctl_setup_func(struct sysctllog **); 116 117 #ifdef SYSCTL_DEBUG 118 #define DPRINTF(a) printf a 119 #else 120 #define DPRINTF(a) 121 #endif 122 123 struct sysctllog { 124 const struct sysctlnode *log_root; 125 int *log_num; 126 int log_size, log_left; 127 }; 128 129 /* 130 * the "root" of the new sysctl tree 131 */ 132 struct sysctlnode sysctl_root = { 133 .sysctl_flags = SYSCTL_VERSION| 134 CTLFLAG_ROOT|CTLFLAG_READWRITE| 135 CTLTYPE_NODE, 136 .sysctl_num = 0, 137 .sysctl_size = sizeof(struct sysctlnode), 138 .sysctl_name = "(root)", 139 }; 140 141 /* 142 * link set of functions that add nodes at boot time (see also 143 * sysctl_buildtree()) 144 */ 145 __link_set_decl(sysctl_funcs, sysctl_setup_func); 146 147 /* 148 * The `sysctl_treelock' is intended to serialize access to the sysctl 149 * tree. XXX This has serious problems; allocating memory and 150 * copying data out with the lock held is insane. 151 */ 152 krwlock_t sysctl_treelock; 153 154 kmutex_t sysctl_file_marker_lock; 155 156 /* 157 * Attributes stored in the kernel. 158 */ 159 char hostname[MAXHOSTNAMELEN]; 160 int hostnamelen; 161 162 char domainname[MAXHOSTNAMELEN]; 163 int domainnamelen; 164 165 long hostid; 166 167 #ifndef DEFCORENAME 168 #define DEFCORENAME "%n.core" 169 #endif 170 char defcorename[MAXPATHLEN] = DEFCORENAME; 171 172 /* 173 * ******************************************************************** 174 * Section 0: Some simple glue 175 * ******************************************************************** 176 * By wrapping copyin(), copyout(), and copyinstr() like this, we can 177 * stop caring about who's calling us and simplify some code a bunch. 178 * ******************************************************************** 179 */ 180 int 181 sysctl_copyin(struct lwp *l, const void *uaddr, void *kaddr, size_t len) 182 { 183 int error; 184 185 if (l != NULL) { 186 error = copyin(uaddr, kaddr, len); 187 ktrmibio(-1, UIO_WRITE, uaddr, len, error); 188 } else { 189 error = kcopy(uaddr, kaddr, len); 190 } 191 192 return error; 193 } 194 195 int 196 sysctl_copyout(struct lwp *l, const void *kaddr, void *uaddr, size_t len) 197 { 198 int error; 199 200 if (l != NULL) { 201 error = copyout(kaddr, uaddr, len); 202 ktrmibio(-1, UIO_READ, uaddr, len, error); 203 } else { 204 error = kcopy(kaddr, uaddr, len); 205 } 206 207 return error; 208 } 209 210 int 211 sysctl_copyinstr(struct lwp *l, const void *uaddr, void *kaddr, 212 size_t len, size_t *done) 213 { 214 int error; 215 216 if (l != NULL) { 217 error = copyinstr(uaddr, kaddr, len, done); 218 ktrmibio(-1, UIO_WRITE, uaddr, len, error); 219 } else { 220 error = copystr(uaddr, kaddr, len, done); 221 } 222 223 return error; 224 } 225 226 /* 227 * ******************************************************************** 228 * Initialize sysctl subsystem. 229 * ******************************************************************** 230 */ 231 void 232 sysctl_init(void) 233 { 234 sysctl_setup_func *const *sysctl_setup; 235 236 rw_init(&sysctl_treelock); 237 238 /* 239 * dynamic mib numbers start here 240 */ 241 sysctl_root.sysctl_num = CREATE_BASE; 242 sysctl_basenode_init(); 243 244 __link_set_foreach(sysctl_setup, sysctl_funcs) { 245 (**sysctl_setup)(NULL); 246 } 247 248 mutex_init(&sysctl_file_marker_lock, MUTEX_DEFAULT, IPL_NONE); 249 } 250 251 /* 252 * Setting this means no more permanent nodes can be added, 253 * trees that claim to be readonly at the root now are, and if 254 * the main tree is readonly, *everything* is. 255 * 256 * Also starts up the PRNG used for the "random" sysctl: it's 257 * better to start it later than sooner. 258 * 259 * Call this at the end of kernel init. 260 */ 261 void 262 sysctl_finalize(void) 263 { 264 265 sysctl_root.sysctl_flags |= CTLFLAG_PERMANENT; 266 } 267 268 /* 269 * ******************************************************************** 270 * The main native sysctl system call itself. 271 * ******************************************************************** 272 */ 273 int 274 sys___sysctl(struct lwp *l, const struct sys___sysctl_args *uap, register_t *retval) 275 { 276 /* { 277 syscallarg(const int *) name; 278 syscallarg(u_int) namelen; 279 syscallarg(void *) old; 280 syscallarg(size_t *) oldlenp; 281 syscallarg(const void *) new; 282 syscallarg(size_t) newlen; 283 } */ 284 int error, nerror, name[CTL_MAXNAME]; 285 size_t oldlen, savelen, *oldlenp; 286 287 /* 288 * get oldlen 289 */ 290 oldlen = 0; 291 oldlenp = SCARG(uap, oldlenp); 292 if (oldlenp != NULL) { 293 error = copyin(oldlenp, &oldlen, sizeof(oldlen)); 294 if (error) 295 return (error); 296 } 297 savelen = oldlen; 298 299 /* 300 * top-level sysctl names may or may not be non-terminal, but 301 * we don't care 302 */ 303 if (SCARG(uap, namelen) > CTL_MAXNAME || SCARG(uap, namelen) < 1) 304 return (EINVAL); 305 error = copyin(SCARG(uap, name), &name, 306 SCARG(uap, namelen) * sizeof(int)); 307 if (error) 308 return (error); 309 310 ktrmib(name, SCARG(uap, namelen)); 311 312 sysctl_lock(SCARG(uap, newv) != NULL); 313 314 /* 315 * do sysctl work (NULL means main built-in default tree) 316 */ 317 error = sysctl_dispatch(&name[0], SCARG(uap, namelen), 318 SCARG(uap, oldv), &oldlen, 319 SCARG(uap, newv), SCARG(uap, newlen), 320 &name[0], l, NULL); 321 322 /* 323 * release the sysctl lock 324 */ 325 sysctl_unlock(); 326 327 /* 328 * set caller's oldlen to new value even in the face of an 329 * error (if this gets an error and they didn't have one, they 330 * get this one) 331 */ 332 if (oldlenp) { 333 nerror = copyout(&oldlen, oldlenp, sizeof(oldlen)); 334 if (error == 0) 335 error = nerror; 336 } 337 338 /* 339 * if the only problem is that we weren't given enough space, 340 * that's an ENOMEM error 341 */ 342 if (error == 0 && SCARG(uap, oldv) != NULL && savelen < oldlen) 343 error = ENOMEM; 344 345 return (error); 346 } 347 348 /* 349 * ******************************************************************** 350 * Section 1: How the tree is used 351 * ******************************************************************** 352 * Implementations of sysctl for emulations should typically need only 353 * these three functions in this order: lock the tree, dispatch 354 * request into it, unlock the tree. 355 * ******************************************************************** 356 */ 357 void 358 sysctl_lock(bool write) 359 { 360 361 if (write) { 362 rw_enter(&sysctl_treelock, RW_WRITER); 363 curlwp->l_pflag |= LP_SYSCTLWRITE; 364 } else { 365 rw_enter(&sysctl_treelock, RW_READER); 366 curlwp->l_pflag &= ~LP_SYSCTLWRITE; 367 } 368 } 369 370 void 371 sysctl_relock(void) 372 { 373 374 if ((curlwp->l_pflag & LP_SYSCTLWRITE) != 0) { 375 rw_enter(&sysctl_treelock, RW_WRITER); 376 } else { 377 rw_enter(&sysctl_treelock, RW_READER); 378 } 379 } 380 381 /* 382 * ******************************************************************** 383 * the main sysctl dispatch routine. scans the given tree and picks a 384 * function to call based on what it finds. 385 * ******************************************************************** 386 */ 387 int 388 sysctl_dispatch(SYSCTLFN_ARGS) 389 { 390 int error; 391 sysctlfn fn; 392 int ni; 393 394 KASSERT(rw_lock_held(&sysctl_treelock)); 395 396 if (rnode && SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 397 printf("sysctl_dispatch: rnode %p wrong version\n", rnode); 398 error = EINVAL; 399 goto out; 400 } 401 402 fn = NULL; 403 error = sysctl_locate(l, name, namelen, &rnode, &ni); 404 405 if (rnode->sysctl_func != NULL) { 406 /* 407 * the node we ended up at has a function, so call it. it can 408 * hand off to query or create if it wants to. 409 */ 410 fn = rnode->sysctl_func; 411 } else if (error == 0) { 412 /* 413 * we found the node they were looking for, so do a lookup. 414 */ 415 fn = (sysctlfn)sysctl_lookup; /* XXX may write to rnode */ 416 } else if (error == ENOENT && (ni + 1) == namelen && name[ni] < 0) { 417 /* 418 * prospective parent node found, but the terminal node was 419 * not. generic operations associate with the parent. 420 */ 421 switch (name[ni]) { 422 case CTL_QUERY: 423 fn = sysctl_query; 424 break; 425 case CTL_CREATE: 426 #if NKSYMS > 0 427 case CTL_CREATESYM: 428 #endif /* NKSYMS > 0 */ 429 if (newp == NULL) { 430 error = EINVAL; 431 break; 432 } 433 KASSERT(rw_write_held(&sysctl_treelock)); 434 fn = (sysctlfn)sysctl_create; /* we own the rnode */ 435 break; 436 case CTL_DESTROY: 437 if (newp == NULL) { 438 error = EINVAL; 439 break; 440 } 441 KASSERT(rw_write_held(&sysctl_treelock)); 442 fn = (sysctlfn)sysctl_destroy; /* we own the rnode */ 443 break; 444 case CTL_MMAP: 445 fn = (sysctlfn)sysctl_mmap; /* we own the rnode */ 446 break; 447 case CTL_DESCRIBE: 448 fn = sysctl_describe; 449 break; 450 default: 451 error = EOPNOTSUPP; 452 break; 453 } 454 } 455 456 /* 457 * after all of that, maybe we found someone who knows how to 458 * get us what we want? 459 */ 460 if (fn != NULL) 461 error = (*fn)(name + ni, namelen - ni, oldp, oldlenp, 462 newp, newlen, name, l, rnode); 463 else if (error == 0) 464 error = EOPNOTSUPP; 465 466 out: 467 return (error); 468 } 469 470 /* 471 * ******************************************************************** 472 * Releases the tree lock. 473 * ******************************************************************** 474 */ 475 void 476 sysctl_unlock(void) 477 { 478 479 rw_exit(&sysctl_treelock); 480 } 481 482 /* 483 * ******************************************************************** 484 * Section 2: The main tree interfaces 485 * ******************************************************************** 486 * This is how sysctl_dispatch() does its work, and you can too, by 487 * calling these routines from helpers (though typically only 488 * sysctl_lookup() will be used). The tree MUST BE LOCKED when these 489 * are called. 490 * ******************************************************************** 491 */ 492 493 /* 494 * sysctl_locate -- Finds the node matching the given mib under the 495 * given tree (via rv). If no tree is given, we fall back to the 496 * native tree. The current process (via l) is used for access 497 * control on the tree (some nodes may be traversable only by root) and 498 * on return, nip will show how many numbers in the mib were consumed. 499 */ 500 int 501 sysctl_locate(struct lwp *l, const int *name, u_int namelen, 502 const struct sysctlnode **rnode, int *nip) 503 { 504 const struct sysctlnode *node, *pnode; 505 int tn, si, ni, error, alias; 506 507 KASSERT(rw_lock_held(&sysctl_treelock)); 508 509 /* 510 * basic checks and setup 511 */ 512 if (*rnode == NULL) 513 *rnode = &sysctl_root; 514 if (nip) 515 *nip = 0; 516 if (namelen == 0) 517 return (0); 518 519 /* 520 * search starts from "root" 521 */ 522 pnode = *rnode; 523 if (SYSCTL_VERS(pnode->sysctl_flags) != SYSCTL_VERSION) { 524 printf("sysctl_locate: pnode %p wrong version\n", pnode); 525 return (EINVAL); 526 } 527 node = pnode->sysctl_child; 528 error = 0; 529 530 /* 531 * scan for node to which new node should be attached 532 */ 533 for (ni = 0; ni < namelen; ni++) { 534 /* 535 * walked off bottom of tree 536 */ 537 if (node == NULL) { 538 if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE) 539 error = ENOENT; 540 else 541 error = ENOTDIR; 542 break; 543 } 544 /* 545 * can anyone traverse this node or only root? 546 */ 547 if (l != NULL && (pnode->sysctl_flags & CTLFLAG_PRIVATE) && 548 (error = kauth_authorize_system(l->l_cred, 549 KAUTH_SYSTEM_SYSCTL, KAUTH_REQ_SYSTEM_SYSCTL_PRVT, 550 NULL, NULL, NULL)) != 0) 551 return (error); 552 /* 553 * find a child node with the right number 554 */ 555 tn = name[ni]; 556 alias = 0; 557 558 si = 0; 559 /* 560 * Note: ANYNUMBER only matches positive integers. 561 * Since ANYNUMBER is only permitted on single-node 562 * sub-trees (eg proc), check before the loop and skip 563 * it if we can. 564 */ 565 if ((node[si].sysctl_flags & CTLFLAG_ANYNUMBER) && (tn >= 0)) 566 goto foundit; 567 for (; si < pnode->sysctl_clen; si++) { 568 if (node[si].sysctl_num == tn) { 569 if (node[si].sysctl_flags & CTLFLAG_ALIAS) { 570 if (alias++ == 4) 571 break; 572 else { 573 tn = node[si].sysctl_alias; 574 si = -1; 575 } 576 } else 577 goto foundit; 578 } 579 } 580 /* 581 * if we ran off the end, it obviously doesn't exist 582 */ 583 error = ENOENT; 584 break; 585 586 /* 587 * so far so good, move on down the line 588 */ 589 foundit: 590 pnode = &node[si]; 591 if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE) 592 node = node[si].sysctl_child; 593 else 594 node = NULL; 595 } 596 597 *rnode = pnode; 598 if (nip) 599 *nip = ni; 600 601 return (error); 602 } 603 604 /* 605 * sysctl_query -- The auto-discovery engine. Copies out the structs 606 * describing nodes under the given node and handles overlay trees. 607 */ 608 int 609 sysctl_query(SYSCTLFN_ARGS) 610 { 611 int error, ni, elim, v; 612 size_t out, left, t; 613 const struct sysctlnode *enode, *onode; 614 struct sysctlnode qnode; 615 616 KASSERT(rw_lock_held(&sysctl_treelock)); 617 618 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 619 printf("sysctl_query: rnode %p wrong version\n", rnode); 620 return (EINVAL); 621 } 622 623 if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE) 624 return (ENOTDIR); 625 if (namelen != 1 || name[0] != CTL_QUERY) 626 return (EINVAL); 627 628 error = 0; 629 out = 0; 630 left = *oldlenp; 631 elim = 0; 632 enode = NULL; 633 634 /* 635 * translate the given request to a current node 636 */ 637 error = sysctl_cvt_in(l, &v, newp, newlen, &qnode); 638 if (error) 639 return (error); 640 641 /* 642 * if the request specifies a version, check it 643 */ 644 if (qnode.sysctl_ver != 0) { 645 enode = rnode; 646 if (qnode.sysctl_ver != enode->sysctl_ver && 647 qnode.sysctl_ver != sysctl_rootof(enode)->sysctl_ver) 648 return (EINVAL); 649 } 650 651 /* 652 * process has overlay tree 653 */ 654 if (l && l->l_proc->p_emul->e_sysctlovly) { 655 enode = l->l_proc->p_emul->e_sysctlovly; 656 elim = (name - oname); 657 error = sysctl_locate(l, oname, elim, &enode, NULL); 658 if (error == 0) { 659 /* ah, found parent in overlay */ 660 elim = enode->sysctl_clen; 661 enode = enode->sysctl_child; 662 } else { 663 error = 0; 664 elim = 0; 665 enode = NULL; 666 } 667 } 668 669 for (ni = 0; ni < rnode->sysctl_clen; ni++) { 670 onode = &rnode->sysctl_child[ni]; 671 if (enode && enode->sysctl_num == onode->sysctl_num) { 672 if (SYSCTL_TYPE(enode->sysctl_flags) != CTLTYPE_NODE) 673 onode = enode; 674 if (--elim > 0) 675 enode++; 676 else 677 enode = NULL; 678 } 679 error = sysctl_cvt_out(l, v, onode, oldp, left, &t); 680 if (error) 681 return (error); 682 if (oldp != NULL) 683 oldp = (char*)oldp + t; 684 out += t; 685 left -= MIN(left, t); 686 } 687 688 /* 689 * overlay trees *MUST* be entirely consumed 690 */ 691 KASSERT(enode == NULL); 692 693 *oldlenp = out; 694 695 return (error); 696 } 697 698 /* 699 * sysctl_create -- Adds a node (the description of which is taken 700 * from newp) to the tree, returning a copy of it in the space pointed 701 * to by oldp. In the event that the requested slot is already taken 702 * (either by name or by number), the offending node is returned 703 * instead. Yes, this is complex, but we want to make sure everything 704 * is proper. 705 */ 706 #ifdef SYSCTL_DEBUG_CREATE 707 int _sysctl_create(SYSCTLFN_ARGS); 708 int 709 _sysctl_create(SYSCTLFN_ARGS) 710 #else 711 int 712 sysctl_create(SYSCTLFN_ARGS) 713 #endif 714 { 715 struct sysctlnode nnode, *node, *pnode; 716 int error, ni, at, nm, type, nsz, sz, flags, anum, v; 717 void *own; 718 719 KASSERT(rw_write_held(&sysctl_treelock)); 720 721 error = 0; 722 own = NULL; 723 anum = -1; 724 725 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 726 printf("sysctl_create: rnode %p wrong version\n", rnode); 727 return (EINVAL); 728 } 729 730 if (namelen != 1 || (name[namelen - 1] != CTL_CREATE 731 #if NKSYMS > 0 732 && name[namelen - 1] != CTL_CREATESYM 733 #endif /* NKSYMS > 0 */ 734 )) 735 return (EINVAL); 736 737 /* 738 * processes can only add nodes at securelevel 0, must be 739 * root, and can't add nodes to a parent that's not writeable 740 */ 741 if (l != NULL) { 742 #ifndef SYSCTL_DISALLOW_CREATE 743 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SYSCTL, 744 KAUTH_REQ_SYSTEM_SYSCTL_ADD, NULL, NULL, NULL); 745 if (error) 746 return (error); 747 if (!(rnode->sysctl_flags & CTLFLAG_READWRITE)) 748 #endif /* SYSCTL_DISALLOW_CREATE */ 749 return (EPERM); 750 } 751 752 /* 753 * nothing can add a node if: 754 * we've finished initial set up of this tree and 755 * (the tree itself is not writeable or 756 * the entire sysctl system is not writeable) 757 */ 758 if ((sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_PERMANENT) && 759 (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) || 760 !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE))) 761 return (EPERM); 762 763 /* 764 * it must be a "node", not a "int" or something 765 */ 766 if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE) 767 return (ENOTDIR); 768 if (rnode->sysctl_flags & CTLFLAG_ALIAS) { 769 printf("sysctl_create: attempt to add node to aliased " 770 "node %p\n", rnode); 771 return (EINVAL); 772 } 773 pnode = __UNCONST(rnode); /* we are adding children to this node */ 774 775 if (newp == NULL) 776 return (EINVAL); 777 error = sysctl_cvt_in(l, &v, newp, newlen, &nnode); 778 if (error) 779 return (error); 780 781 /* 782 * nodes passed in don't *have* parents 783 */ 784 if (nnode.sysctl_parent != NULL) 785 return (EINVAL); 786 787 /* 788 * if we are indeed adding it, it should be a "good" name and 789 * number 790 */ 791 nm = nnode.sysctl_num; 792 #if NKSYMS > 0 793 if (nm == CTL_CREATESYM) 794 nm = CTL_CREATE; 795 #endif /* NKSYMS > 0 */ 796 if (nm < 0 && nm != CTL_CREATE) 797 return (EINVAL); 798 799 /* 800 * the name can't start with a digit 801 */ 802 if (nnode.sysctl_name[0] >= '0' && 803 nnode.sysctl_name[0] <= '9') 804 return (EINVAL); 805 806 /* 807 * the name must be only alphanumerics or - or _, longer than 808 * 0 bytes and less than SYSCTL_NAMELEN 809 */ 810 nsz = 0; 811 while (nsz < SYSCTL_NAMELEN && nnode.sysctl_name[nsz] != '\0') { 812 if ((nnode.sysctl_name[nsz] >= '0' && 813 nnode.sysctl_name[nsz] <= '9') || 814 (nnode.sysctl_name[nsz] >= 'A' && 815 nnode.sysctl_name[nsz] <= 'Z') || 816 (nnode.sysctl_name[nsz] >= 'a' && 817 nnode.sysctl_name[nsz] <= 'z') || 818 nnode.sysctl_name[nsz] == '-' || 819 nnode.sysctl_name[nsz] == '_') 820 nsz++; 821 else 822 return (EINVAL); 823 } 824 if (nsz == 0 || nsz == SYSCTL_NAMELEN) 825 return (EINVAL); 826 827 /* 828 * various checks revolve around size vs type, etc 829 */ 830 type = SYSCTL_TYPE(nnode.sysctl_flags); 831 flags = SYSCTL_FLAGS(nnode.sysctl_flags); 832 sz = nnode.sysctl_size; 833 834 /* 835 * find out if there's a collision, and if so, let the caller 836 * know what they collided with 837 */ 838 node = pnode->sysctl_child; 839 at = 0; 840 if (node) { 841 if ((flags | node->sysctl_flags) & CTLFLAG_ANYNUMBER) 842 /* No siblings for a CTLFLAG_ANYNUMBER node */ 843 return EINVAL; 844 for (ni = 0; ni < pnode->sysctl_clen; ni++) { 845 if (nm == node[ni].sysctl_num || 846 strcmp(nnode.sysctl_name, node[ni].sysctl_name) == 0) { 847 /* 848 * ignore error here, since we 849 * are already fixed on EEXIST 850 */ 851 (void)sysctl_cvt_out(l, v, &node[ni], oldp, 852 *oldlenp, oldlenp); 853 return (EEXIST); 854 } 855 if (nm > node[ni].sysctl_num) 856 at++; 857 } 858 } 859 860 /* 861 * use sysctl_ver to add to the tree iff it hasn't changed 862 */ 863 if (nnode.sysctl_ver != 0) { 864 /* 865 * a specified value must match either the parent 866 * node's version or the root node's version 867 */ 868 if (nnode.sysctl_ver != sysctl_rootof(rnode)->sysctl_ver && 869 nnode.sysctl_ver != rnode->sysctl_ver) { 870 return (EINVAL); 871 } 872 } 873 874 /* 875 * only the kernel can assign functions to entries 876 */ 877 if (l != NULL && nnode.sysctl_func != NULL) 878 return (EPERM); 879 880 /* 881 * only the kernel can create permanent entries, and only then 882 * before the kernel is finished setting itself up 883 */ 884 if (l != NULL && (flags & ~SYSCTL_USERFLAGS)) 885 return (EPERM); 886 if ((flags & CTLFLAG_PERMANENT) & 887 (sysctl_root.sysctl_flags & CTLFLAG_PERMANENT)) 888 return (EPERM); 889 if ((flags & (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE)) == 890 (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE)) 891 return (EINVAL); 892 if ((flags & CTLFLAG_IMMEDIATE) && 893 type != CTLTYPE_INT && type != CTLTYPE_QUAD && type != CTLTYPE_BOOL) 894 return (EINVAL); 895 896 /* 897 * check size, or set it if unset and we can figure it out. 898 * kernel created nodes are allowed to have a function instead 899 * of a size (or a data pointer). 900 */ 901 switch (type) { 902 case CTLTYPE_NODE: 903 /* 904 * only *i* can assert the size of a node 905 */ 906 if (flags & CTLFLAG_ALIAS) { 907 anum = nnode.sysctl_alias; 908 if (anum < 0) 909 return (EINVAL); 910 nnode.sysctl_alias = 0; 911 } 912 if (sz != 0 || nnode.sysctl_data != NULL) 913 return (EINVAL); 914 if (nnode.sysctl_csize != 0 || 915 nnode.sysctl_clen != 0 || 916 nnode.sysctl_child != 0) 917 return (EINVAL); 918 if (flags & CTLFLAG_OWNDATA) 919 return (EINVAL); 920 sz = sizeof(struct sysctlnode); 921 break; 922 case CTLTYPE_INT: 923 /* 924 * since an int is an int, if the size is not given or 925 * is wrong, we can "int-uit" it. 926 */ 927 if (sz != 0 && sz != sizeof(int)) 928 return (EINVAL); 929 sz = sizeof(int); 930 break; 931 case CTLTYPE_STRING: 932 /* 933 * strings are a little more tricky 934 */ 935 if (sz == 0) { 936 if (l == NULL) { 937 if (nnode.sysctl_func == NULL) { 938 if (nnode.sysctl_data == NULL) 939 return (EINVAL); 940 else 941 sz = strlen(nnode.sysctl_data) + 942 1; 943 } 944 } else if (nnode.sysctl_data == NULL && 945 flags & CTLFLAG_OWNDATA) { 946 return (EINVAL); 947 } else { 948 char *vp, *e; 949 size_t s; 950 951 /* 952 * we want a rough idea of what the 953 * size is now 954 */ 955 vp = malloc(PAGE_SIZE, M_SYSCTLDATA, M_WAITOK); 956 if (vp == NULL) 957 return (ENOMEM); 958 e = nnode.sysctl_data; 959 do { 960 error = copyinstr(e, vp, PAGE_SIZE, &s); 961 if (error) { 962 if (error != ENAMETOOLONG) { 963 free(vp, M_SYSCTLDATA); 964 return (error); 965 } 966 e += PAGE_SIZE; 967 if ((e - 32 * PAGE_SIZE) > 968 (char*)nnode.sysctl_data) { 969 free(vp, M_SYSCTLDATA); 970 return (ERANGE); 971 } 972 } 973 } while (error != 0); 974 sz = s + (e - (char*)nnode.sysctl_data); 975 free(vp, M_SYSCTLDATA); 976 } 977 } 978 break; 979 case CTLTYPE_QUAD: 980 if (sz != 0 && sz != sizeof(u_quad_t)) 981 return (EINVAL); 982 sz = sizeof(u_quad_t); 983 break; 984 case CTLTYPE_BOOL: 985 /* 986 * since a bool is a bool, if the size is not given or 987 * is wrong, we can "intuit" it. 988 */ 989 if (sz != 0 && sz != sizeof(bool)) 990 return (EINVAL); 991 sz = sizeof(bool); 992 break; 993 case CTLTYPE_STRUCT: 994 if (sz == 0) { 995 if (l != NULL || nnode.sysctl_func == NULL) 996 return (EINVAL); 997 if (flags & CTLFLAG_OWNDATA) 998 return (EINVAL); 999 } 1000 break; 1001 default: 1002 return (EINVAL); 1003 } 1004 1005 /* 1006 * at this point, if sz is zero, we *must* have a 1007 * function to go with it and we can't own it. 1008 */ 1009 1010 /* 1011 * l ptr own 1012 * 0 0 0 -> EINVAL (if no func) 1013 * 0 0 1 -> own 1014 * 0 1 0 -> kptr 1015 * 0 1 1 -> kptr 1016 * 1 0 0 -> EINVAL 1017 * 1 0 1 -> own 1018 * 1 1 0 -> kptr, no own (fault on lookup) 1019 * 1 1 1 -> uptr, own 1020 */ 1021 if (type != CTLTYPE_NODE) { 1022 if (sz != 0) { 1023 if (flags & CTLFLAG_OWNDATA) { 1024 own = malloc(sz, M_SYSCTLDATA, M_WAITOK); 1025 if (own == NULL) 1026 return ENOMEM; 1027 if (nnode.sysctl_data == NULL) 1028 memset(own, 0, sz); 1029 else { 1030 error = sysctl_copyin(l, 1031 nnode.sysctl_data, own, sz); 1032 if (error != 0) { 1033 free(own, M_SYSCTLDATA); 1034 return (error); 1035 } 1036 } 1037 } else if ((nnode.sysctl_data != NULL) && 1038 !(flags & CTLFLAG_IMMEDIATE)) { 1039 #if NKSYMS > 0 1040 if (name[namelen - 1] == CTL_CREATESYM) { 1041 char symname[128]; /* XXX enough? */ 1042 u_long symaddr; 1043 size_t symlen; 1044 1045 error = sysctl_copyinstr(l, 1046 nnode.sysctl_data, symname, 1047 sizeof(symname), &symlen); 1048 if (error) 1049 return (error); 1050 error = ksyms_getval(NULL, symname, 1051 &symaddr, KSYMS_EXTERN); 1052 if (error) 1053 return (error); /* EINVAL? */ 1054 nnode.sysctl_data = (void*)symaddr; 1055 } 1056 #endif /* NKSYMS > 0 */ 1057 /* 1058 * Ideally, we'd like to verify here 1059 * that this address is acceptable, 1060 * but... 1061 * 1062 * - it might be valid now, only to 1063 * become invalid later 1064 * 1065 * - it might be invalid only for the 1066 * moment and valid later 1067 * 1068 * - or something else. 1069 * 1070 * Since we can't get a good answer, 1071 * we'll just accept the address as 1072 * given, and fault on individual 1073 * lookups. 1074 */ 1075 } 1076 } else if (nnode.sysctl_func == NULL) 1077 return (EINVAL); 1078 } 1079 1080 /* 1081 * a process can't assign a function to a node, and the kernel 1082 * can't create a node that has no function or data. 1083 * (XXX somewhat redundant check) 1084 */ 1085 if (l != NULL || nnode.sysctl_func == NULL) { 1086 if (type != CTLTYPE_NODE && 1087 !(flags & CTLFLAG_IMMEDIATE) && 1088 nnode.sysctl_data == NULL && 1089 own == NULL) 1090 return (EINVAL); 1091 } 1092 1093 #ifdef SYSCTL_DISALLOW_KWRITE 1094 /* 1095 * a process can't create a writable node unless it refers to 1096 * new data. 1097 */ 1098 if (l != NULL && own == NULL && type != CTLTYPE_NODE && 1099 (flags & CTLFLAG_READWRITE) != CTLFLAG_READONLY && 1100 !(flags & CTLFLAG_IMMEDIATE)) 1101 return (EPERM); 1102 #endif /* SYSCTL_DISALLOW_KWRITE */ 1103 1104 /* 1105 * make sure there's somewhere to put the new stuff. 1106 */ 1107 if (pnode->sysctl_child == NULL) { 1108 if (flags & CTLFLAG_ANYNUMBER) 1109 error = sysctl_alloc(pnode, 1); 1110 else 1111 error = sysctl_alloc(pnode, 0); 1112 if (error) { 1113 if (own != NULL) 1114 free(own, M_SYSCTLDATA); 1115 return (error); 1116 } 1117 } 1118 node = pnode->sysctl_child; 1119 1120 /* 1121 * no collisions, so pick a good dynamic number if we need to. 1122 */ 1123 if (nm == CTL_CREATE) { 1124 nm = ++sysctl_root.sysctl_num; 1125 for (ni = 0; ni < pnode->sysctl_clen; ni++) { 1126 if (nm == node[ni].sysctl_num) { 1127 nm++; 1128 ni = -1; 1129 } else if (nm > node[ni].sysctl_num) 1130 at = ni + 1; 1131 } 1132 } 1133 1134 /* 1135 * oops...ran out of space 1136 */ 1137 if (pnode->sysctl_clen == pnode->sysctl_csize) { 1138 error = sysctl_realloc(pnode); 1139 if (error) { 1140 if (own != NULL) 1141 free(own, M_SYSCTLDATA); 1142 return (error); 1143 } 1144 node = pnode->sysctl_child; 1145 } 1146 1147 /* 1148 * insert new node data 1149 */ 1150 if (at < pnode->sysctl_clen) { 1151 int t; 1152 1153 /* 1154 * move the nodes that should come after the new one 1155 */ 1156 memmove(&node[at + 1], &node[at], 1157 (pnode->sysctl_clen - at) * sizeof(struct sysctlnode)); 1158 memset(&node[at], 0, sizeof(struct sysctlnode)); 1159 node[at].sysctl_parent = pnode; 1160 /* 1161 * and...reparent any children of any moved nodes 1162 */ 1163 for (ni = at; ni <= pnode->sysctl_clen; ni++) 1164 if (node[ni].sysctl_child != NULL) 1165 for (t = 0; t < node[ni].sysctl_csize; t++) 1166 node[ni].sysctl_child[t].sysctl_parent = 1167 &node[ni]; 1168 } 1169 node = &node[at]; 1170 pnode->sysctl_clen++; 1171 1172 strlcpy(node->sysctl_name, nnode.sysctl_name, 1173 sizeof(node->sysctl_name)); 1174 node->sysctl_num = nm; 1175 node->sysctl_size = sz; 1176 node->sysctl_flags = SYSCTL_VERSION|type|flags; /* XXX other trees */ 1177 node->sysctl_csize = 0; 1178 node->sysctl_clen = 0; 1179 if (own) { 1180 node->sysctl_data = own; 1181 node->sysctl_flags |= CTLFLAG_OWNDATA; 1182 } else if (flags & CTLFLAG_ALIAS) { 1183 node->sysctl_alias = anum; 1184 } else if (flags & CTLFLAG_IMMEDIATE) { 1185 switch (type) { 1186 case CTLTYPE_BOOL: 1187 node->sysctl_bdata = nnode.sysctl_bdata; 1188 break; 1189 case CTLTYPE_INT: 1190 node->sysctl_idata = nnode.sysctl_idata; 1191 break; 1192 case CTLTYPE_QUAD: 1193 node->sysctl_qdata = nnode.sysctl_qdata; 1194 break; 1195 } 1196 } else { 1197 node->sysctl_data = nnode.sysctl_data; 1198 node->sysctl_flags &= ~CTLFLAG_OWNDATA; 1199 } 1200 node->sysctl_func = nnode.sysctl_func; 1201 node->sysctl_child = NULL; 1202 /* node->sysctl_parent should already be done */ 1203 1204 /* 1205 * update "version" on path to "root" 1206 */ 1207 for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent) 1208 ; 1209 pnode = node; 1210 for (nm = rnode->sysctl_ver + 1; pnode != NULL; 1211 pnode = pnode->sysctl_parent) 1212 pnode->sysctl_ver = nm; 1213 1214 /* If this fails, the node is already added - the user won't know! */ 1215 error = sysctl_cvt_out(l, v, node, oldp, *oldlenp, oldlenp); 1216 1217 return (error); 1218 } 1219 1220 /* 1221 * ******************************************************************** 1222 * A wrapper around sysctl_create() that prints the thing we're trying 1223 * to add. 1224 * ******************************************************************** 1225 */ 1226 #ifdef SYSCTL_DEBUG_CREATE 1227 int 1228 sysctl_create(SYSCTLFN_ARGS) 1229 { 1230 const struct sysctlnode *node; 1231 int k, v, rc, ni, nl = namelen + (name - oname); 1232 struct sysctlnode nnode; 1233 1234 if (newp == NULL) 1235 return EINVAL; 1236 int error = sysctl_cvt_in(l, &v, newp, newlen, &nnode); 1237 if (error) 1238 return error; 1239 1240 node = &nnode; 1241 1242 printf("namelen %d (", nl); 1243 for (ni = 0; ni < nl - 1; ni++) 1244 printf(" %d", oname[ni]); 1245 printf(" %d )\t[%s]\tflags %08x (%08x %d %zu)\n", 1246 k = node->sysctl_num, 1247 node->sysctl_name, 1248 node->sysctl_flags, 1249 SYSCTL_FLAGS(node->sysctl_flags), 1250 SYSCTL_TYPE(node->sysctl_flags), 1251 node->sysctl_size); 1252 1253 node = rnode; 1254 rc = _sysctl_create(SYSCTLFN_CALL(rnode)); 1255 1256 printf("sysctl_create("); 1257 for (ni = 0; ni < nl - 1; ni++) 1258 printf(" %d", oname[ni]); 1259 printf(" %d ) returned %d\n", k, rc); 1260 1261 return (rc); 1262 } 1263 #endif /* SYSCTL_DEBUG_CREATE */ 1264 1265 /* 1266 * sysctl_destroy -- Removes a node (as described by newp) from the 1267 * given tree, returning (if successful) a copy of the dead node in 1268 * oldp. Since we're removing stuff, there's not much to check. 1269 */ 1270 int 1271 sysctl_destroy(SYSCTLFN_ARGS) 1272 { 1273 struct sysctlnode *node, *pnode, onode, nnode; 1274 int ni, error, v; 1275 1276 KASSERT(rw_write_held(&sysctl_treelock)); 1277 1278 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 1279 printf("sysctl_destroy: rnode %p wrong version\n", rnode); 1280 return (EINVAL); 1281 } 1282 1283 error = 0; 1284 1285 if (namelen != 1 || name[namelen - 1] != CTL_DESTROY) 1286 return (EINVAL); 1287 1288 /* 1289 * processes can only destroy nodes at securelevel 0, must be 1290 * root, and can't remove nodes from a parent that's not 1291 * writeable 1292 */ 1293 if (l != NULL) { 1294 #ifndef SYSCTL_DISALLOW_CREATE 1295 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SYSCTL, 1296 KAUTH_REQ_SYSTEM_SYSCTL_DELETE, NULL, NULL, NULL); 1297 if (error) 1298 return (error); 1299 if (!(rnode->sysctl_flags & CTLFLAG_READWRITE)) 1300 #endif /* SYSCTL_DISALLOW_CREATE */ 1301 return (EPERM); 1302 } 1303 1304 /* 1305 * nothing can remove a node if: 1306 * the node is permanent (checked later) or 1307 * the tree itself is not writeable or 1308 * the entire sysctl system is not writeable 1309 * 1310 * note that we ignore whether setup is complete or not, 1311 * because these rules always apply. 1312 */ 1313 if (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) || 1314 !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE)) 1315 return (EPERM); 1316 1317 if (newp == NULL) 1318 return (EINVAL); 1319 error = sysctl_cvt_in(l, &v, newp, newlen, &nnode); 1320 if (error) 1321 return (error); 1322 memset(&onode, 0, sizeof(struct sysctlnode)); 1323 1324 node = rnode->sysctl_child; 1325 for (ni = 0; ni < rnode->sysctl_clen; ni++) { 1326 if (nnode.sysctl_num == node[ni].sysctl_num) { 1327 /* 1328 * if name specified, must match 1329 */ 1330 if (nnode.sysctl_name[0] != '\0' && 1331 strcmp(nnode.sysctl_name, node[ni].sysctl_name)) 1332 continue; 1333 /* 1334 * if version specified, must match 1335 */ 1336 if (nnode.sysctl_ver != 0 && 1337 nnode.sysctl_ver != node[ni].sysctl_ver) 1338 continue; 1339 /* 1340 * this must be the one 1341 */ 1342 break; 1343 } 1344 } 1345 if (ni == rnode->sysctl_clen) 1346 return (ENOENT); 1347 node = &node[ni]; 1348 pnode = node->sysctl_parent; 1349 1350 /* 1351 * if the kernel says permanent, it is, so there. nyah. 1352 */ 1353 if (SYSCTL_FLAGS(node->sysctl_flags) & CTLFLAG_PERMANENT) 1354 return (EPERM); 1355 1356 /* 1357 * can't delete non-empty nodes 1358 */ 1359 if (SYSCTL_TYPE(node->sysctl_flags) == CTLTYPE_NODE && 1360 node->sysctl_clen != 0) 1361 return (ENOTEMPTY); 1362 1363 /* 1364 * if the node "owns" data, release it now 1365 */ 1366 if (node->sysctl_flags & CTLFLAG_OWNDATA) { 1367 if (node->sysctl_data != NULL) 1368 free(node->sysctl_data, M_SYSCTLDATA); 1369 node->sysctl_data = NULL; 1370 } 1371 if (node->sysctl_flags & CTLFLAG_OWNDESC) { 1372 if (node->sysctl_desc != NULL) 1373 /*XXXUNCONST*/ 1374 free(__UNCONST(node->sysctl_desc), M_SYSCTLDATA); 1375 node->sysctl_desc = NULL; 1376 } 1377 1378 /* 1379 * if the node to be removed is not the last one on the list, 1380 * move the remaining nodes up, and reparent any grandchildren 1381 */ 1382 onode = *node; 1383 if (ni < pnode->sysctl_clen - 1) { 1384 int t; 1385 1386 memmove(&pnode->sysctl_child[ni], &pnode->sysctl_child[ni + 1], 1387 (pnode->sysctl_clen - ni - 1) * 1388 sizeof(struct sysctlnode)); 1389 for (; ni < pnode->sysctl_clen - 1; ni++) 1390 if (SYSCTL_TYPE(pnode->sysctl_child[ni].sysctl_flags) == 1391 CTLTYPE_NODE) 1392 for (t = 0; 1393 t < pnode->sysctl_child[ni].sysctl_clen; 1394 t++) 1395 pnode->sysctl_child[ni].sysctl_child[t]. 1396 sysctl_parent = 1397 &pnode->sysctl_child[ni]; 1398 ni = pnode->sysctl_clen - 1; 1399 node = &pnode->sysctl_child[ni]; 1400 } 1401 1402 /* 1403 * reset the space we just vacated 1404 */ 1405 memset(node, 0, sizeof(struct sysctlnode)); 1406 node->sysctl_parent = pnode; 1407 pnode->sysctl_clen--; 1408 1409 /* 1410 * if this parent just lost its last child, nuke the creche 1411 */ 1412 if (pnode->sysctl_clen == 0) { 1413 free(pnode->sysctl_child, M_SYSCTLNODE); 1414 pnode->sysctl_csize = 0; 1415 pnode->sysctl_child = NULL; 1416 } 1417 1418 /* 1419 * update "version" on path to "root" 1420 */ 1421 for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent) 1422 ; 1423 for (ni = rnode->sysctl_ver + 1; pnode != NULL; 1424 pnode = pnode->sysctl_parent) 1425 pnode->sysctl_ver = ni; 1426 1427 error = sysctl_cvt_out(l, v, &onode, oldp, *oldlenp, oldlenp); 1428 1429 return (error); 1430 } 1431 1432 /* 1433 * sysctl_lookup -- Handles copyin/copyout of new and old values. 1434 * Partial reads are globally allowed. Only root can write to things 1435 * unless the node says otherwise. 1436 */ 1437 int 1438 sysctl_lookup(SYSCTLFN_ARGS) 1439 { 1440 int error, rw; 1441 size_t sz, len; 1442 void *d; 1443 1444 KASSERT(rw_lock_held(&sysctl_treelock)); 1445 1446 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 1447 printf("%s: rnode %p wrong version\n", __func__, rnode); 1448 return EINVAL; 1449 } 1450 1451 if (newlen == 0) 1452 newp = NULL; 1453 1454 error = 0; 1455 1456 /* 1457 * you can't "look up" a node. you can "query" it, but you 1458 * can't "look it up". 1459 */ 1460 if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_NODE || namelen != 0) { 1461 DPRINTF(("%s: can't lookup a node\n", __func__)); 1462 return EINVAL; 1463 } 1464 1465 /* 1466 * some nodes are private, so only root can look into them. 1467 */ 1468 if (l != NULL && (rnode->sysctl_flags & CTLFLAG_PRIVATE) && 1469 (error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SYSCTL, 1470 KAUTH_REQ_SYSTEM_SYSCTL_PRVT, NULL, NULL, NULL)) != 0) { 1471 DPRINTF(("%s: private node\n", __func__)); 1472 return error; 1473 } 1474 1475 /* 1476 * if a node wants to be writable according to different rules 1477 * other than "only root can write to stuff unless a flag is 1478 * set", then it needs its own function which should have been 1479 * called and not us. 1480 */ 1481 if (l != NULL && newp != NULL && 1482 !(rnode->sysctl_flags & CTLFLAG_ANYWRITE) && 1483 (error = kauth_authorize_system(l->l_cred, 1484 KAUTH_SYSTEM_SYSCTL, KAUTH_REQ_SYSTEM_SYSCTL_MODIFY, NULL, NULL, 1485 NULL)) != 0) { 1486 DPRINTF(("%s: can't modify\n", __func__)); 1487 return error; 1488 } 1489 1490 /* 1491 * is this node supposedly writable? 1492 */ 1493 rw = (rnode->sysctl_flags & CTLFLAG_READWRITE) ? 1 : 0; 1494 1495 /* 1496 * it appears not to be writable at this time, so if someone 1497 * tried to write to it, we must tell them to go away 1498 */ 1499 if (!rw && newp != NULL) { 1500 DPRINTF(("%s: not writable\n", __func__)); 1501 return EPERM; 1502 } 1503 1504 /* 1505 * step one, copy out the stuff we have presently 1506 */ 1507 if (rnode->sysctl_flags & CTLFLAG_IMMEDIATE) { 1508 /* 1509 * note that we discard const here because we are 1510 * modifying the contents of the node (which is okay 1511 * because it's ours) 1512 * 1513 * It also doesn't matter which field of the union we pick. 1514 */ 1515 d = __UNCONST(&rnode->sysctl_qdata); 1516 } else 1517 d = rnode->sysctl_data; 1518 1519 if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_STRING) 1520 sz = strlen(d) + 1; /* XXX@@@ possible fault here */ 1521 else 1522 sz = rnode->sysctl_size; 1523 if (oldp != NULL) { 1524 error = sysctl_copyout(l, d, oldp, MIN(sz, *oldlenp)); 1525 if (error) { 1526 DPRINTF(("%s: bad copyout %d\n", __func__, error)); 1527 return error; 1528 } 1529 } 1530 *oldlenp = sz; 1531 1532 /* 1533 * are we done? 1534 */ 1535 if (newp == NULL) 1536 return 0; 1537 1538 /* 1539 * hmm...not done. must now "copy in" new value. re-adjust 1540 * sz to maximum value (strings are "weird"). 1541 */ 1542 sz = rnode->sysctl_size; 1543 switch (SYSCTL_TYPE(rnode->sysctl_flags)) { 1544 case CTLTYPE_BOOL: { 1545 bool tmp; 1546 /* 1547 * these data must be *exactly* the same size coming 1548 * in. bool may only be true or false. 1549 */ 1550 if (newlen != sz) { 1551 DPRINTF(("%s: bad size %zu != %zu\n", __func__, newlen, 1552 sz)); 1553 return EINVAL; 1554 } 1555 error = sysctl_copyin(l, newp, &tmp, sz); 1556 if (error) 1557 break; 1558 if (tmp != true && tmp != false) { 1559 DPRINTF(("%s: tmp %d\n", __func__, tmp)); 1560 return EINVAL; 1561 } 1562 *(bool *)d = tmp; 1563 break; 1564 } 1565 case CTLTYPE_INT: 1566 case CTLTYPE_QUAD: 1567 case CTLTYPE_STRUCT: 1568 /* 1569 * these data must be *exactly* the same size coming 1570 * in. 1571 */ 1572 if (newlen != sz) 1573 goto bad_size; 1574 error = sysctl_copyin(l, newp, d, sz); 1575 rnd_add_data(NULL, d, sz, 0); 1576 break; 1577 case CTLTYPE_STRING: { 1578 /* 1579 * strings, on the other hand, can be shorter, and we 1580 * let userland be sloppy about the trailing nul. 1581 */ 1582 char *newbuf; 1583 1584 /* 1585 * too much new string? 1586 */ 1587 if (newlen > sz) 1588 goto bad_size; 1589 1590 /* 1591 * temporary copy of new inbound string 1592 */ 1593 len = MIN(sz, newlen); 1594 newbuf = malloc(len, M_SYSCTLDATA, M_WAITOK); 1595 if (newbuf == NULL) { 1596 DPRINTF(("%s: oomem %zu\n", __func__, len)); 1597 return ENOMEM; 1598 } 1599 error = sysctl_copyin(l, newp, newbuf, len); 1600 if (error) { 1601 free(newbuf, M_SYSCTLDATA); 1602 DPRINTF(("%s: copyin %d\n", __func__, error)); 1603 return error; 1604 } 1605 1606 /* 1607 * did they NUL terminate it, or do we have space 1608 * left to do it ourselves? 1609 */ 1610 if (newbuf[len - 1] != '\0' && len == sz) { 1611 free(newbuf, M_SYSCTLDATA); 1612 DPRINTF(("%s: string too long\n", __func__)); 1613 return EINVAL; 1614 } 1615 1616 /* 1617 * looks good, so pop it into place and zero the rest. 1618 */ 1619 if (len > 0) { 1620 memcpy(d, newbuf, len); 1621 rnd_add_data(NULL, d, len, 0); 1622 } 1623 if (sz != len) 1624 memset((char*)d + len, 0, sz - len); 1625 free(newbuf, M_SYSCTLDATA); 1626 break; 1627 } 1628 default: 1629 DPRINTF(("%s: bad type\n", __func__)); 1630 return EINVAL; 1631 } 1632 if (error) { 1633 DPRINTF(("%s: copyin %d\n", __func__, error)); 1634 } 1635 1636 return error; 1637 1638 bad_size: 1639 DPRINTF(("%s: bad size %zu > %zu\n", __func__, newlen, sz)); 1640 return EINVAL; 1641 } 1642 1643 /* 1644 * sysctl_mmap -- Dispatches sysctl mmap requests to those nodes that 1645 * purport to handle it. This interface isn't fully fleshed out yet, 1646 * unfortunately. 1647 */ 1648 static int 1649 sysctl_mmap(SYSCTLFN_ARGS) 1650 { 1651 const struct sysctlnode *node; 1652 struct sysctlnode nnode; 1653 int error; 1654 int sysctl_num; 1655 1656 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 1657 printf("sysctl_mmap: rnode %p wrong version\n", rnode); 1658 return (EINVAL); 1659 } 1660 1661 /* 1662 * let's just pretend that didn't happen, m'kay? 1663 */ 1664 if (l == NULL) 1665 return (EPERM); 1666 1667 /* 1668 * is this a sysctlnode description of an mmap request? 1669 */ 1670 if (newp == NULL || newlen != sizeof(struct sysctlnode)) 1671 return (EINVAL); 1672 error = sysctl_copyin(l, newp, &nnode, sizeof(nnode)); 1673 if (error) 1674 return (error); 1675 1676 /* 1677 * does the node they asked for exist? 1678 */ 1679 if (namelen != 1) 1680 return (EOPNOTSUPP); 1681 node = rnode; 1682 sysctl_num = nnode.sysctl_num; 1683 error = sysctl_locate(l, &sysctl_num, 1, &node, NULL); 1684 if (error) 1685 return (error); 1686 1687 /* 1688 * does this node that we have found purport to handle mmap? 1689 */ 1690 if (node->sysctl_func == NULL || 1691 !(node->sysctl_flags & CTLFLAG_MMAP)) 1692 return (EOPNOTSUPP); 1693 1694 /* 1695 * well...okay, they asked for it. 1696 */ 1697 return ((*node->sysctl_func)(SYSCTLFN_CALL(node))); 1698 } 1699 1700 int 1701 sysctl_describe(SYSCTLFN_ARGS) 1702 { 1703 struct sysctldesc *d; 1704 void *bf; 1705 size_t sz, left, tot; 1706 int i, error, v = -1; 1707 struct sysctlnode *node; 1708 struct sysctlnode dnode; 1709 1710 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 1711 printf("sysctl_query: rnode %p wrong version\n", rnode); 1712 return (EINVAL); 1713 } 1714 1715 if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE) 1716 return (ENOTDIR); 1717 if (namelen != 1 || name[0] != CTL_DESCRIBE) 1718 return (EINVAL); 1719 1720 /* 1721 * get ready... 1722 */ 1723 error = 0; 1724 d = bf = malloc(MAXDESCLEN, M_TEMP, M_WAITOK); 1725 if (bf == NULL) 1726 return ENOMEM; 1727 tot = 0; 1728 node = rnode->sysctl_child; 1729 left = *oldlenp; 1730 1731 /* 1732 * no request -> all descriptions at this level 1733 * request with desc unset -> just this node 1734 * request with desc set -> set descr for this node 1735 */ 1736 if (newp != NULL) { 1737 error = sysctl_cvt_in(l, &v, newp, newlen, &dnode); 1738 if (error) 1739 goto out; 1740 if (dnode.sysctl_desc != NULL) { 1741 /* 1742 * processes cannot set descriptions above 1743 * securelevel 0. and must be root. blah 1744 * blah blah. a couple more checks are made 1745 * once we find the node we want. 1746 */ 1747 if (l != NULL) { 1748 #ifndef SYSCTL_DISALLOW_CREATE 1749 error = kauth_authorize_system(l->l_cred, 1750 KAUTH_SYSTEM_SYSCTL, 1751 KAUTH_REQ_SYSTEM_SYSCTL_DESC, NULL, 1752 NULL, NULL); 1753 if (error) 1754 goto out; 1755 #else /* SYSCTL_DISALLOW_CREATE */ 1756 error = EPERM; 1757 goto out; 1758 #endif /* SYSCTL_DISALLOW_CREATE */ 1759 } 1760 1761 /* 1762 * find node and try to set the description on it 1763 */ 1764 for (i = 0; i < rnode->sysctl_clen; i++) 1765 if (node[i].sysctl_num == dnode.sysctl_num) 1766 break; 1767 if (i == rnode->sysctl_clen) { 1768 error = ENOENT; 1769 goto out; 1770 } 1771 node = &node[i]; 1772 1773 /* 1774 * did the caller specify a node version? 1775 */ 1776 if (dnode.sysctl_ver != 0 && 1777 dnode.sysctl_ver != node->sysctl_ver) { 1778 error = EINVAL; 1779 goto out; 1780 } 1781 1782 /* 1783 * okay...some rules: 1784 * (1) if setup is done and the tree is 1785 * read-only or the whole system is 1786 * read-only 1787 * (2) no one can set a description on a 1788 * permanent node (it must be set when 1789 * using createv) 1790 * (3) processes cannot *change* a description 1791 * (4) processes *can*, however, set a 1792 * description on a read-only node so that 1793 * one can be created and then described 1794 * in two steps 1795 * anything else come to mind? 1796 */ 1797 if ((sysctl_root.sysctl_flags & CTLFLAG_PERMANENT) && 1798 (!(sysctl_rootof(node)->sysctl_flags & 1799 CTLFLAG_READWRITE) || 1800 !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE))) { 1801 error = EPERM; 1802 goto out; 1803 } 1804 if (node->sysctl_flags & CTLFLAG_PERMANENT) { 1805 error = EPERM; 1806 goto out; 1807 } 1808 if (l != NULL && node->sysctl_desc != NULL) { 1809 error = EPERM; 1810 goto out; 1811 } 1812 1813 /* 1814 * right, let's go ahead. the first step is 1815 * making the description into something the 1816 * node can "own", if need be. 1817 */ 1818 if (l != NULL || 1819 dnode.sysctl_flags & CTLFLAG_OWNDESC) { 1820 char *nd, *k; 1821 1822 k = malloc(MAXDESCLEN, M_TEMP, M_WAITOK); 1823 if (k == NULL) { 1824 error = ENOMEM; 1825 goto out; 1826 } 1827 error = sysctl_copyinstr(l, dnode.sysctl_desc, 1828 k, MAXDESCLEN, &sz); 1829 if (error) { 1830 free(k, M_TEMP); 1831 goto out; 1832 } 1833 nd = malloc(sz, M_SYSCTLDATA, M_WAITOK); 1834 if (nd == NULL) { 1835 free(k, M_TEMP); 1836 error = ENOMEM; 1837 goto out; 1838 } 1839 memcpy(nd, k, sz); 1840 dnode.sysctl_flags |= CTLFLAG_OWNDESC; 1841 dnode.sysctl_desc = nd; 1842 free(k, M_TEMP); 1843 } 1844 1845 /* 1846 * now "release" the old description and 1847 * attach the new one. ta-da. 1848 */ 1849 if ((node->sysctl_flags & CTLFLAG_OWNDESC) && 1850 node->sysctl_desc != NULL) 1851 /*XXXUNCONST*/ 1852 free(__UNCONST(node->sysctl_desc), M_SYSCTLDATA); 1853 node->sysctl_desc = dnode.sysctl_desc; 1854 node->sysctl_flags |= 1855 (dnode.sysctl_flags & CTLFLAG_OWNDESC); 1856 1857 /* 1858 * now we "fall out" and into the loop which 1859 * will copy the new description back out for 1860 * those interested parties 1861 */ 1862 } 1863 } 1864 1865 /* 1866 * scan for one description or just retrieve all descriptions 1867 */ 1868 for (i = 0; i < rnode->sysctl_clen; i++) { 1869 /* 1870 * did they ask for the description of only one node? 1871 */ 1872 if (v != -1 && node[i].sysctl_num != dnode.sysctl_num) 1873 continue; 1874 1875 /* 1876 * don't describe "private" nodes to non-suser users 1877 */ 1878 if ((node[i].sysctl_flags & CTLFLAG_PRIVATE) && (l != NULL) && 1879 !(kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SYSCTL, 1880 KAUTH_REQ_SYSTEM_SYSCTL_PRVT, NULL, NULL, NULL))) 1881 continue; 1882 1883 /* 1884 * is this description "valid"? 1885 */ 1886 memset(bf, 0, MAXDESCLEN); 1887 if (node[i].sysctl_desc == NULL) 1888 sz = 1; 1889 else if (copystr(node[i].sysctl_desc, &d->descr_str[0], 1890 MAXDESCLEN - sizeof(*d), &sz) != 0) { 1891 /* 1892 * erase possible partial description 1893 */ 1894 memset(bf, 0, MAXDESCLEN); 1895 sz = 1; 1896 } 1897 1898 /* 1899 * we've got it, stuff it into the caller's buffer 1900 */ 1901 d->descr_num = node[i].sysctl_num; 1902 d->descr_ver = node[i].sysctl_ver; 1903 d->descr_len = sz; /* includes trailing nul */ 1904 sz = (char *)NEXT_DESCR(d) - (char *)d; 1905 if (oldp != NULL && left >= sz) { 1906 error = sysctl_copyout(l, d, oldp, sz); 1907 if (error) 1908 goto out; 1909 left -= sz; 1910 oldp = (void *)__sysc_desc_adv(oldp, d->descr_len); 1911 } 1912 tot += sz; 1913 1914 /* 1915 * if we get this far with v not "unset", they asked 1916 * for a specific node and we found it 1917 */ 1918 if (v != -1) 1919 break; 1920 } 1921 1922 /* 1923 * did we find it after all? 1924 */ 1925 if (v != -1 && tot == 0) 1926 error = ENOENT; 1927 else 1928 *oldlenp = tot; 1929 1930 out: 1931 free(bf, M_TEMP); 1932 return (error); 1933 } 1934 1935 /* 1936 * ******************************************************************** 1937 * Section 3: Create and destroy from inside the kernel 1938 * ******************************************************************** 1939 * sysctl_createv() and sysctl_destroyv() are simpler-to-use 1940 * interfaces for the kernel to fling new entries into the mib and rip 1941 * them out later. In the case of sysctl_createv(), the returned copy 1942 * of the node (see sysctl_create()) will be translated back into a 1943 * pointer to the actual node. 1944 * 1945 * Note that sysctl_createv() will return 0 if the create request 1946 * matches an existing node (ala mkdir -p), and that sysctl_destroyv() 1947 * will return 0 if the node to be destroyed already does not exist 1948 * (aka rm -f) or if it is a parent of other nodes. 1949 * 1950 * This allows two (or more) different subsystems to assert sub-tree 1951 * existence before populating their own nodes, and to remove their 1952 * own nodes without orphaning the others when they are done. 1953 * ******************************************************************** 1954 */ 1955 #undef sysctl_createv 1956 int 1957 sysctl_createv(struct sysctllog **log, int cflags, 1958 const struct sysctlnode **rnode, const struct sysctlnode **cnode, 1959 int flags, int type, const char *namep, const char *descr, 1960 sysctlfn func, u_quad_t qv, void *newp, size_t newlen, 1961 ...) 1962 { 1963 va_list ap; 1964 int error, ni, namelen, name[CTL_MAXNAME]; 1965 const struct sysctlnode *root, *pnode; 1966 struct sysctlnode nnode, onode, *dnode; 1967 size_t sz; 1968 const struct sysctlnode *snode __diagused; 1969 1970 /* 1971 * where are we putting this? 1972 */ 1973 if (rnode != NULL && *rnode == NULL) { 1974 printf("sysctl_createv: rnode NULL\n"); 1975 return (EINVAL); 1976 } 1977 root = rnode ? *rnode : NULL; 1978 if (cnode != NULL) 1979 *cnode = NULL; 1980 if (cflags != 0) 1981 return (EINVAL); 1982 1983 /* 1984 * what is it? 1985 */ 1986 flags = SYSCTL_VERSION|SYSCTL_TYPE(type)|SYSCTL_FLAGS(flags); 1987 if (log != NULL) 1988 flags &= ~CTLFLAG_PERMANENT; 1989 1990 /* 1991 * where do we put it? 1992 */ 1993 va_start(ap, newlen); 1994 namelen = 0; 1995 error = 0; 1996 ni = -1; 1997 do { 1998 if (++ni == CTL_MAXNAME) { 1999 error = ENAMETOOLONG; 2000 break; 2001 } 2002 name[ni] = va_arg(ap, int); 2003 /* 2004 * sorry, this is not supported from here 2005 */ 2006 if (name[ni] == CTL_CREATESYM) { 2007 error = EINVAL; 2008 break; 2009 } 2010 } while (name[ni] != CTL_EOL && name[ni] != CTL_CREATE); 2011 va_end(ap); 2012 if (error) 2013 return error; 2014 namelen = ni + (name[ni] == CTL_CREATE ? 1 : 0); 2015 2016 /* 2017 * what's it called 2018 */ 2019 if (strlcpy(nnode.sysctl_name, namep, sizeof(nnode.sysctl_name)) >= 2020 sizeof(nnode.sysctl_name)) 2021 return (ENAMETOOLONG); 2022 2023 /* 2024 * cons up the description of the new node 2025 */ 2026 nnode.sysctl_num = name[namelen - 1]; 2027 name[namelen - 1] = CTL_CREATE; 2028 nnode.sysctl_size = newlen; 2029 nnode.sysctl_flags = flags; 2030 if (type == CTLTYPE_NODE) { 2031 nnode.sysctl_csize = 0; 2032 nnode.sysctl_clen = 0; 2033 nnode.sysctl_child = NULL; 2034 if (flags & CTLFLAG_ALIAS) 2035 nnode.sysctl_alias = qv; 2036 } else if (flags & CTLFLAG_IMMEDIATE) { 2037 switch (type) { 2038 case CTLTYPE_BOOL: 2039 nnode.sysctl_bdata = qv; 2040 break; 2041 case CTLTYPE_INT: 2042 nnode.sysctl_idata = qv; 2043 break; 2044 case CTLTYPE_QUAD: 2045 nnode.sysctl_qdata = qv; 2046 break; 2047 default: 2048 return (EINVAL); 2049 } 2050 } else { 2051 nnode.sysctl_data = newp; 2052 } 2053 nnode.sysctl_func = func; 2054 nnode.sysctl_parent = NULL; 2055 nnode.sysctl_ver = 0; 2056 2057 /* 2058 * initialize lock state -- we need locks if the main tree has 2059 * been marked as complete, but since we could be called from 2060 * either there, or from a device driver (say, at device 2061 * insertion), or from a module (at module load time, say), we 2062 * don't really want to "wait"... 2063 */ 2064 sysctl_lock(true); 2065 2066 /* 2067 * locate the prospective parent of the new node, and if we 2068 * find it, add the new node. 2069 */ 2070 sz = sizeof(onode); 2071 pnode = root; 2072 error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni); 2073 if (error) { 2074 /* 2075 * XXX: If you are seeing this printf in early bringup 2076 * stages, perhaps your setfault is not functioning and 2077 * thus kcopy() is mis-behaving. 2078 */ 2079 printf("sysctl_createv: sysctl_locate(%s) returned %d\n", 2080 nnode.sysctl_name, error); 2081 sysctl_unlock(); 2082 return (error); 2083 } 2084 error = sysctl_create(&name[ni], namelen - ni, &onode, &sz, 2085 &nnode, sizeof(nnode), &name[0], NULL, 2086 pnode); 2087 2088 /* 2089 * unfortunately the node we wanted to create is already 2090 * there. if the node that's already there is a reasonable 2091 * facsimile of the node we wanted to create, just pretend 2092 * (for the caller's benefit) that we managed to create the 2093 * node they wanted. 2094 */ 2095 if (error == EEXIST) { 2096 /* name is the same as requested... */ 2097 if (strcmp(nnode.sysctl_name, onode.sysctl_name) == 0 && 2098 /* they want the same function... */ 2099 nnode.sysctl_func == onode.sysctl_func && 2100 /* number is the same as requested, or... */ 2101 (nnode.sysctl_num == onode.sysctl_num || 2102 /* they didn't pick a number... */ 2103 nnode.sysctl_num == CTL_CREATE)) { 2104 /* 2105 * collision here from trying to create 2106 * something that already existed; let's give 2107 * our customers a hand and tell them they got 2108 * what they wanted. 2109 */ 2110 #ifdef SYSCTL_DEBUG_CREATE 2111 printf("cleared\n"); 2112 #endif /* SYSCTL_DEBUG_CREATE */ 2113 error = 0; 2114 } 2115 } 2116 2117 if (error == 0 && 2118 (cnode != NULL || log != NULL || descr != NULL)) { 2119 /* 2120 * sysctl_create() gave us back a copy of the node, 2121 * but we need to know where it actually is... 2122 */ 2123 pnode = root; 2124 error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni); 2125 snode = pnode; 2126 2127 /* 2128 * manual scan of last layer so that aliased nodes 2129 * aren't followed. 2130 */ 2131 if (error == 0) { 2132 for (ni = 0; ni < pnode->sysctl_clen; ni++) 2133 if (pnode->sysctl_child[ni].sysctl_num == 2134 onode.sysctl_num) 2135 break; 2136 if (ni < pnode->sysctl_clen) 2137 pnode = &pnode->sysctl_child[ni]; 2138 else 2139 error = ENOENT; 2140 } 2141 2142 /* 2143 * not expecting an error here, but... 2144 */ 2145 if (error == 0) { 2146 KASSERTMSG(pnode->sysctl_parent == snode, 2147 "sysctl parent mis-match pnode %s, snode %s", 2148 pnode->sysctl_name, snode->sysctl_name); 2149 if (log != NULL) 2150 sysctl_log_add(log, pnode); 2151 if (cnode != NULL) 2152 *cnode = pnode; 2153 if (descr != NULL) { 2154 /* 2155 * allow first caller to *set* a 2156 * description actually to set it 2157 * 2158 * discard const here so we can attach 2159 * the description 2160 */ 2161 dnode = __UNCONST(pnode); 2162 if (pnode->sysctl_desc != NULL) 2163 /* skip it...we've got one */; 2164 else if (flags & CTLFLAG_OWNDESC) { 2165 size_t l = strlen(descr) + 1; 2166 char *d = malloc(l, M_SYSCTLDATA, 2167 M_WAITOK); 2168 if (d != NULL) { 2169 memcpy(d, descr, l); 2170 dnode->sysctl_desc = d; 2171 dnode->sysctl_flags |= 2172 CTLFLAG_OWNDESC; 2173 } 2174 } else 2175 dnode->sysctl_desc = descr; 2176 } 2177 } else { 2178 printf("sysctl_create succeeded but node not found?!\n"); 2179 /* 2180 * confusing, but the create said it 2181 * succeeded, so... 2182 */ 2183 error = 0; 2184 } 2185 } 2186 2187 /* 2188 * now it should be safe to release the lock state. note that 2189 * the pointer to the newly created node being passed back may 2190 * not be "good" for very long. 2191 */ 2192 sysctl_unlock(); 2193 2194 if (error != 0) { 2195 printf("sysctl_createv: sysctl_create(%s) returned %d\n", 2196 nnode.sysctl_name, error); 2197 #if 0 2198 if (error != ENOENT) 2199 sysctl_dump(&onode); 2200 #endif 2201 } 2202 2203 return (error); 2204 } 2205 2206 int 2207 sysctl_destroyv(struct sysctlnode *rnode, ...) 2208 { 2209 va_list ap; 2210 int error, name[CTL_MAXNAME], namelen, ni; 2211 const struct sysctlnode *pnode, *node; 2212 struct sysctlnode dnode, *onode; 2213 size_t sz; 2214 2215 va_start(ap, rnode); 2216 namelen = 0; 2217 ni = 0; 2218 do { 2219 if (ni == CTL_MAXNAME) { 2220 va_end(ap); 2221 return (ENAMETOOLONG); 2222 } 2223 name[ni] = va_arg(ap, int); 2224 } while (name[ni++] != CTL_EOL); 2225 namelen = ni - 1; 2226 va_end(ap); 2227 2228 /* 2229 * i can't imagine why we'd be destroying a node when the tree 2230 * wasn't complete, but who knows? 2231 */ 2232 sysctl_lock(true); 2233 2234 /* 2235 * where is it? 2236 */ 2237 node = rnode; 2238 error = sysctl_locate(NULL, &name[0], namelen - 1, &node, &ni); 2239 if (error) { 2240 /* they want it gone and it's not there, so... */ 2241 sysctl_unlock(); 2242 return (error == ENOENT ? 0 : error); 2243 } 2244 2245 /* 2246 * set up the deletion 2247 */ 2248 pnode = node; 2249 node = &dnode; 2250 memset(&dnode, 0, sizeof(dnode)); 2251 dnode.sysctl_flags = SYSCTL_VERSION; 2252 dnode.sysctl_num = name[namelen - 1]; 2253 2254 /* 2255 * we found it, now let's nuke it 2256 */ 2257 name[namelen - 1] = CTL_DESTROY; 2258 sz = 0; 2259 error = sysctl_destroy(&name[namelen - 1], 1, NULL, &sz, 2260 node, sizeof(*node), &name[0], NULL, 2261 pnode); 2262 if (error == ENOTEMPTY) { 2263 /* 2264 * think of trying to delete "foo" when "foo.bar" 2265 * (which someone else put there) is still in 2266 * existence 2267 */ 2268 error = 0; 2269 2270 /* 2271 * dunno who put the description there, but if this 2272 * node can ever be removed, we need to make sure the 2273 * string doesn't go out of context. that means we 2274 * need to find the node that's still there (don't use 2275 * sysctl_locate() because that follows aliasing). 2276 */ 2277 node = pnode->sysctl_child; 2278 for (ni = 0; ni < pnode->sysctl_clen; ni++) 2279 if (node[ni].sysctl_num == dnode.sysctl_num) 2280 break; 2281 node = (ni < pnode->sysctl_clen) ? &node[ni] : NULL; 2282 2283 /* 2284 * if we found it, and this node has a description, 2285 * and this node can be released, and it doesn't 2286 * already own its own description...sigh. :) 2287 */ 2288 if (node != NULL && node->sysctl_desc != NULL && 2289 !(node->sysctl_flags & CTLFLAG_PERMANENT) && 2290 !(node->sysctl_flags & CTLFLAG_OWNDESC)) { 2291 char *d; 2292 2293 sz = strlen(node->sysctl_desc) + 1; 2294 d = malloc(sz, M_SYSCTLDATA, M_WAITOK); 2295 if (d != NULL) { 2296 /* 2297 * discard const so that we can 2298 * re-attach the description 2299 */ 2300 memcpy(d, node->sysctl_desc, sz); 2301 onode = __UNCONST(node); 2302 onode->sysctl_desc = d; 2303 onode->sysctl_flags |= CTLFLAG_OWNDESC; 2304 } else { 2305 /* 2306 * XXX drop the description? be 2307 * afraid? don't care? 2308 */ 2309 } 2310 } 2311 } 2312 2313 sysctl_unlock(); 2314 2315 return (error); 2316 } 2317 2318 /* 2319 * ******************************************************************** 2320 * Deletes an entire n-ary tree. Not recommended unless you know why 2321 * you're doing it. Personally, I don't know why you'd even think 2322 * about it. 2323 * ******************************************************************** 2324 */ 2325 void 2326 sysctl_free(struct sysctlnode *rnode) 2327 { 2328 struct sysctlnode *node, *pnode; 2329 2330 rw_enter(&sysctl_treelock, RW_WRITER); 2331 2332 if (rnode == NULL) 2333 rnode = &sysctl_root; 2334 2335 if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) { 2336 printf("sysctl_free: rnode %p wrong version\n", rnode); 2337 rw_exit(&sysctl_treelock); 2338 return; 2339 } 2340 2341 pnode = rnode; 2342 2343 node = pnode->sysctl_child; 2344 do { 2345 while (node != NULL && pnode->sysctl_csize > 0) { 2346 while (node < 2347 &pnode->sysctl_child[pnode->sysctl_clen] && 2348 (SYSCTL_TYPE(node->sysctl_flags) != 2349 CTLTYPE_NODE || 2350 node->sysctl_csize == 0)) { 2351 if (SYSCTL_FLAGS(node->sysctl_flags) & 2352 CTLFLAG_OWNDATA) { 2353 if (node->sysctl_data != NULL) { 2354 free(node->sysctl_data, 2355 M_SYSCTLDATA); 2356 node->sysctl_data = NULL; 2357 } 2358 } 2359 if (SYSCTL_FLAGS(node->sysctl_flags) & 2360 CTLFLAG_OWNDESC) { 2361 if (node->sysctl_desc != NULL) { 2362 /*XXXUNCONST*/ 2363 free(__UNCONST(node->sysctl_desc), 2364 M_SYSCTLDATA); 2365 node->sysctl_desc = NULL; 2366 } 2367 } 2368 node++; 2369 } 2370 if (node < &pnode->sysctl_child[pnode->sysctl_clen]) { 2371 pnode = node; 2372 node = node->sysctl_child; 2373 } else 2374 break; 2375 } 2376 if (pnode->sysctl_child != NULL) 2377 free(pnode->sysctl_child, M_SYSCTLNODE); 2378 pnode->sysctl_clen = 0; 2379 pnode->sysctl_csize = 0; 2380 pnode->sysctl_child = NULL; 2381 node = pnode; 2382 pnode = node->sysctl_parent; 2383 } while (pnode != NULL && node != rnode); 2384 2385 rw_exit(&sysctl_treelock); 2386 } 2387 2388 void 2389 sysctl_log_print(const struct sysctllog *slog) 2390 { 2391 int i, len; 2392 2393 printf("root %p left %d size %d content", (const void *)slog->log_root, 2394 slog->log_left, slog->log_size); 2395 2396 for (len = 0, i = slog->log_left; i < slog->log_size; i++) { 2397 switch (len) { 2398 case 0: 2399 len = -1; 2400 printf(" version %d", slog->log_num[i]); 2401 break; 2402 case -1: 2403 len = -2; 2404 printf(" type %d", slog->log_num[i]); 2405 break; 2406 case -2: 2407 len = slog->log_num[i]; 2408 printf(" len %d:", slog->log_num[i]); 2409 if (len <= 0) 2410 len = -1; 2411 break; 2412 default: 2413 len--; 2414 printf(" %d", slog->log_num[i]); 2415 break; 2416 } 2417 } 2418 printf(" end\n"); 2419 } 2420 2421 int 2422 sysctl_log_add(struct sysctllog **logp, const struct sysctlnode *node) 2423 { 2424 const int size0 = 16; 2425 int name[CTL_MAXNAME], namelen, i; 2426 const struct sysctlnode *pnode; 2427 struct sysctllog *log; 2428 2429 if (node->sysctl_flags & CTLFLAG_PERMANENT) 2430 return (0); 2431 2432 if (logp == NULL) 2433 return (0); 2434 2435 if (*logp == NULL) { 2436 log = malloc(sizeof(struct sysctllog), 2437 M_SYSCTLDATA, M_WAITOK); 2438 if (log == NULL) { 2439 /* XXX print error message? */ 2440 return (-1); 2441 } 2442 log->log_num = malloc(size0 * sizeof(int), 2443 M_SYSCTLDATA, M_WAITOK); 2444 if (log->log_num == NULL) { 2445 /* XXX print error message? */ 2446 free(log, M_SYSCTLDATA); 2447 return (-1); 2448 } 2449 memset(log->log_num, 0, size0 * sizeof(int)); 2450 log->log_root = NULL; 2451 log->log_size = size0; 2452 log->log_left = size0; 2453 *logp = log; 2454 } else 2455 log = *logp; 2456 2457 /* 2458 * check that the root is proper. it's okay to record the 2459 * address of the root of a tree. it's the only thing that's 2460 * guaranteed not to shift around as nodes come and go. 2461 */ 2462 if (log->log_root == NULL) 2463 log->log_root = sysctl_rootof(node); 2464 else if (log->log_root != sysctl_rootof(node)) { 2465 printf("sysctl: log %p root mismatch (%p)\n", 2466 log->log_root, sysctl_rootof(node)); 2467 return (-1); 2468 } 2469 2470 /* 2471 * we will copy out name in reverse order 2472 */ 2473 for (pnode = node, namelen = 0; 2474 pnode != NULL && !(pnode->sysctl_flags & CTLFLAG_ROOT); 2475 pnode = pnode->sysctl_parent) 2476 name[namelen++] = pnode->sysctl_num; 2477 2478 /* 2479 * do we have space? 2480 */ 2481 if (log->log_left < (namelen + 3)) 2482 sysctl_log_realloc(log); 2483 if (log->log_left < (namelen + 3)) 2484 return (-1); 2485 2486 /* 2487 * stuff name in, then namelen, then node type, and finally, 2488 * the version for non-node nodes. 2489 */ 2490 for (i = 0; i < namelen && i < CTL_MAXNAME; i++) 2491 log->log_num[--log->log_left] = name[i]; 2492 log->log_num[--log->log_left] = namelen; 2493 log->log_num[--log->log_left] = SYSCTL_TYPE(node->sysctl_flags); 2494 if (log->log_num[log->log_left] != CTLTYPE_NODE) 2495 log->log_num[--log->log_left] = node->sysctl_ver; 2496 else 2497 log->log_num[--log->log_left] = 0; 2498 2499 return (0); 2500 } 2501 2502 void 2503 sysctl_teardown(struct sysctllog **logp) 2504 { 2505 const struct sysctlnode *rnode; 2506 struct sysctlnode node; 2507 struct sysctllog *log; 2508 uint namelen; 2509 int *name, t, v, error, ni; 2510 size_t sz; 2511 2512 if (logp == NULL || *logp == NULL) 2513 return; 2514 log = *logp; 2515 2516 rw_enter(&sysctl_treelock, RW_WRITER); 2517 memset(&node, 0, sizeof(node)); 2518 2519 while (log->log_left < log->log_size) { 2520 KASSERT(log->log_left + 3 < log->log_size); 2521 KASSERT(log->log_left + log->log_num[log->log_left + 2] <= 2522 log->log_size); 2523 v = log->log_num[log->log_left++]; 2524 t = log->log_num[log->log_left++]; 2525 namelen = log->log_num[log->log_left++]; 2526 name = &log->log_num[log->log_left]; 2527 2528 node.sysctl_num = name[namelen - 1]; 2529 node.sysctl_flags = SYSCTL_VERSION|t; 2530 node.sysctl_ver = v; 2531 2532 rnode = log->log_root; 2533 error = sysctl_locate(NULL, &name[0], namelen, &rnode, &ni); 2534 if (error == 0) { 2535 name[namelen - 1] = CTL_DESTROY; 2536 rnode = rnode->sysctl_parent; 2537 sz = 0; 2538 (void)sysctl_destroy(&name[namelen - 1], 1, NULL, 2539 &sz, &node, sizeof(node), 2540 &name[0], NULL, rnode); 2541 } 2542 2543 log->log_left += namelen; 2544 } 2545 2546 KASSERT(log->log_size == log->log_left); 2547 free(log->log_num, M_SYSCTLDATA); 2548 free(log, M_SYSCTLDATA); 2549 *logp = NULL; 2550 2551 rw_exit(&sysctl_treelock); 2552 } 2553 2554 /* 2555 * ******************************************************************** 2556 * old_sysctl -- A routine to bridge old-style internal calls to the 2557 * new infrastructure. 2558 * ******************************************************************** 2559 */ 2560 int 2561 old_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, 2562 void *newp, size_t newlen, struct lwp *l) 2563 { 2564 int error; 2565 size_t oldlen = 0; 2566 size_t savelen; 2567 2568 if (oldlenp) { 2569 oldlen = *oldlenp; 2570 } 2571 savelen = oldlen; 2572 2573 sysctl_lock(newp != NULL); 2574 error = sysctl_dispatch(name, namelen, oldp, &oldlen, 2575 newp, newlen, name, l, NULL); 2576 sysctl_unlock(); 2577 if (error == 0 && oldp != NULL && savelen < oldlen) 2578 error = ENOMEM; 2579 if (oldlenp) { 2580 *oldlenp = oldlen; 2581 } 2582 2583 return (error); 2584 } 2585 2586 /* 2587 * ******************************************************************** 2588 * Section 4: Generic helper routines 2589 * ******************************************************************** 2590 * "helper" routines that can do more finely grained access control, 2591 * construct structures from disparate information, create the 2592 * appearance of more nodes and sub-trees, etc. for example, if 2593 * CTL_PROC wanted a helper function, it could respond to a CTL_QUERY 2594 * with a dynamically created list of nodes that represented the 2595 * currently running processes at that instant. 2596 * ******************************************************************** 2597 */ 2598 2599 /* 2600 * first, a few generic helpers that provide: 2601 * 2602 * sysctl_needfunc() a readonly interface that emits a warning 2603 * sysctl_notavail() returns EOPNOTSUPP (generic error) 2604 * sysctl_null() an empty return buffer with no error 2605 */ 2606 int 2607 sysctl_needfunc(SYSCTLFN_ARGS) 2608 { 2609 int error; 2610 2611 printf("!!SYSCTL_NEEDFUNC!!\n"); 2612 2613 if (newp != NULL || namelen != 0) 2614 return (EOPNOTSUPP); 2615 2616 error = 0; 2617 if (oldp != NULL) 2618 error = sysctl_copyout(l, rnode->sysctl_data, oldp, 2619 MIN(rnode->sysctl_size, *oldlenp)); 2620 *oldlenp = rnode->sysctl_size; 2621 2622 return (error); 2623 } 2624 2625 int 2626 sysctl_notavail(SYSCTLFN_ARGS) 2627 { 2628 2629 if (namelen == 1 && name[0] == CTL_QUERY) 2630 return (sysctl_query(SYSCTLFN_CALL(rnode))); 2631 2632 return (EOPNOTSUPP); 2633 } 2634 2635 int 2636 sysctl_null(SYSCTLFN_ARGS) 2637 { 2638 2639 *oldlenp = 0; 2640 2641 return (0); 2642 } 2643 2644 u_int 2645 sysctl_map_flags(const u_int *map, u_int word) 2646 { 2647 u_int rv; 2648 2649 for (rv = 0; *map != 0; map += 2) 2650 if ((word & map[0]) != 0) 2651 rv |= map[1]; 2652 2653 return rv; 2654 } 2655 2656 /* 2657 * ******************************************************************** 2658 * Section 5: The machinery that makes it all go 2659 * ******************************************************************** 2660 * Memory "manglement" routines. Not much to this, eh? 2661 * ******************************************************************** 2662 */ 2663 static int 2664 sysctl_alloc(struct sysctlnode *p, int x) 2665 { 2666 int i; 2667 struct sysctlnode *n; 2668 2669 assert(p->sysctl_child == NULL); 2670 2671 if (x == 1) 2672 n = malloc(sizeof(struct sysctlnode), 2673 M_SYSCTLNODE, M_WAITOK); 2674 else 2675 n = malloc(SYSCTL_DEFSIZE * sizeof(struct sysctlnode), 2676 M_SYSCTLNODE, M_WAITOK); 2677 if (n == NULL) 2678 return (ENOMEM); 2679 2680 if (x == 1) { 2681 memset(n, 0, sizeof(struct sysctlnode)); 2682 p->sysctl_csize = 1; 2683 } else { 2684 memset(n, 0, SYSCTL_DEFSIZE * sizeof(struct sysctlnode)); 2685 p->sysctl_csize = SYSCTL_DEFSIZE; 2686 } 2687 p->sysctl_clen = 0; 2688 2689 for (i = 0; i < p->sysctl_csize; i++) 2690 n[i].sysctl_parent = p; 2691 2692 p->sysctl_child = n; 2693 return (0); 2694 } 2695 2696 static int 2697 sysctl_realloc(struct sysctlnode *p) 2698 { 2699 int i, j, olen; 2700 struct sysctlnode *n; 2701 2702 assert(p->sysctl_csize == p->sysctl_clen); 2703 2704 /* 2705 * how many do we have...how many should we make? 2706 */ 2707 olen = p->sysctl_clen; 2708 n = malloc(2 * olen * sizeof(struct sysctlnode), M_SYSCTLNODE, 2709 M_WAITOK); 2710 if (n == NULL) 2711 return (ENOMEM); 2712 2713 /* 2714 * move old children over...initialize new children 2715 */ 2716 memcpy(n, p->sysctl_child, olen * sizeof(struct sysctlnode)); 2717 memset(&n[olen], 0, olen * sizeof(struct sysctlnode)); 2718 p->sysctl_csize = 2 * olen; 2719 2720 /* 2721 * reattach moved (and new) children to parent; if a moved 2722 * child node has children, reattach the parent pointers of 2723 * grandchildren 2724 */ 2725 for (i = 0; i < p->sysctl_csize; i++) { 2726 n[i].sysctl_parent = p; 2727 if (n[i].sysctl_child != NULL) { 2728 for (j = 0; j < n[i].sysctl_csize; j++) 2729 n[i].sysctl_child[j].sysctl_parent = &n[i]; 2730 } 2731 } 2732 2733 /* 2734 * get out with the old and in with the new 2735 */ 2736 free(p->sysctl_child, M_SYSCTLNODE); 2737 p->sysctl_child = n; 2738 2739 return (0); 2740 } 2741 2742 static int 2743 sysctl_log_realloc(struct sysctllog *log) 2744 { 2745 int *n, s, d; 2746 2747 s = log->log_size * 2; 2748 d = log->log_size; 2749 2750 n = malloc(s * sizeof(int), M_SYSCTLDATA, M_WAITOK); 2751 if (n == NULL) 2752 return (-1); 2753 2754 memset(n, 0, s * sizeof(int)); 2755 memcpy(&n[d], log->log_num, d * sizeof(int)); 2756 free(log->log_num, M_SYSCTLDATA); 2757 log->log_num = n; 2758 if (d) 2759 log->log_left += d; 2760 else 2761 log->log_left = s; 2762 log->log_size = s; 2763 2764 return (0); 2765 } 2766 2767 /* 2768 * ******************************************************************** 2769 * Section 6: Conversion between API versions wrt the sysctlnode 2770 * ******************************************************************** 2771 */ 2772 static int 2773 sysctl_cvt_in(struct lwp *l, int *vp, const void *i, size_t sz, 2774 struct sysctlnode *node) 2775 { 2776 int error, flags; 2777 2778 if (i == NULL || sz < sizeof(flags)) 2779 return (EINVAL); 2780 2781 error = sysctl_copyin(l, i, &flags, sizeof(flags)); 2782 if (error) 2783 return (error); 2784 2785 #if (SYSCTL_VERSION != SYSCTL_VERS_1) 2786 #error sysctl_cvt_in: no support for SYSCTL_VERSION 2787 #endif /* (SYSCTL_VERSION != SYSCTL_VERS_1) */ 2788 2789 if (sz == sizeof(*node) && 2790 SYSCTL_VERS(flags) == SYSCTL_VERSION) { 2791 error = sysctl_copyin(l, i, node, sizeof(*node)); 2792 if (error) 2793 return (error); 2794 *vp = SYSCTL_VERSION; 2795 return (0); 2796 } 2797 2798 return (EINVAL); 2799 } 2800 2801 static int 2802 sysctl_cvt_out(struct lwp *l, int v, const struct sysctlnode *i, 2803 void *ovp, size_t left, size_t *szp) 2804 { 2805 size_t sz = sizeof(*i); 2806 const void *src = i; 2807 int error; 2808 2809 switch (v) { 2810 case SYSCTL_VERS_0: 2811 return (EINVAL); 2812 2813 #if (SYSCTL_VERSION != SYSCTL_VERS_1) 2814 #error sysctl_cvt_out: no support for SYSCTL_VERSION 2815 #endif /* (SYSCTL_VERSION != SYSCTL_VERS_1) */ 2816 2817 case SYSCTL_VERSION: 2818 /* nothing more to do here */ 2819 break; 2820 } 2821 2822 if (ovp != NULL && left >= sz) { 2823 error = sysctl_copyout(l, src, ovp, sz); 2824 if (error) 2825 return (error); 2826 } 2827 2828 if (szp != NULL) 2829 *szp = sz; 2830 2831 return (0); 2832 } 2833 2834 static uint8_t address_key[32]; /* key used in address hashing */ 2835 static ONCE_DECL(random_inithook); 2836 2837 static int 2838 random_address_init(void) 2839 { 2840 2841 cprng_strong(kern_cprng, address_key, sizeof(address_key), 0); 2842 return 0; 2843 } 2844 2845 void 2846 hash_value_ensure_initialized(void) 2847 { 2848 2849 RUN_ONCE(&random_inithook, random_address_init); 2850 } 2851 2852 void 2853 hash_value(void *d, size_t ds, const void *s, size_t ss) 2854 { 2855 2856 blake2s(d, ds, address_key, sizeof(address_key), s, ss); 2857 } 2858