1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Mike Karels at Berkeley Software Design, Inc. 7 * 8 * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD 9 * project, to make these variables more userfriendly. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)kern_sysctl.c 8.4 (Berkeley) 4/14/94 36 * $FreeBSD: src/sys/kern/kern_sysctl.c,v 1.92.2.9 2003/05/01 22:48:09 trhodes Exp $ 37 */ 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/buf.h> 43 #include <sys/sysctl.h> 44 #include <sys/malloc.h> 45 #include <sys/proc.h> 46 #include <sys/priv.h> 47 #include <sys/sysproto.h> 48 #include <sys/lock.h> 49 #include <sys/sbuf.h> 50 51 #include <vm/vm.h> 52 #include <vm/vm_extern.h> 53 54 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic"); 55 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids"); 56 57 /* 58 * The sysctllock protects the MIB tree. It also protects sysctl 59 * contexts used with dynamic sysctls. The sysctl_register_oid() and 60 * sysctl_unregister_oid() routines require the sysctllock to already 61 * be held, so the sysctl_lock() and sysctl_unlock() routines are 62 * provided for the few places in the kernel which need to use that 63 * API rather than using the dynamic API. Use of the dynamic API is 64 * strongly encouraged for most code. 65 */ 66 67 static int sysctl_root(SYSCTL_HANDLER_ARGS); 68 static void sysctl_register_oid_int(struct sysctl_oid *oipd); 69 static void sysctl_unregister_oid_int(struct sysctl_oid *oipd); 70 71 struct sysctl_oid_list sysctl__children; /* root list */ 72 73 static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, 74 int recurse); 75 76 static struct sysctl_oid * 77 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list, int lock) 78 { 79 struct sysctl_oid *oidp; 80 81 SLIST_FOREACH(oidp, list, oid_link) { 82 if (strcmp(oidp->oid_name, name) == 0) { 83 break; 84 } 85 } 86 return (oidp); 87 } 88 89 /* 90 * Initialization of the MIB tree. 91 * 92 * Order by number in each list. 93 */ 94 95 void 96 sysctl_register_oid(struct sysctl_oid *oidp) 97 { 98 SYSCTL_XLOCK(); 99 sysctl_register_oid_int(oidp); 100 SYSCTL_XUNLOCK(); 101 } 102 103 static void 104 sysctl_register_oid_int(struct sysctl_oid *oidp) 105 { 106 struct sysctl_oid_list *parent = oidp->oid_parent; 107 struct sysctl_oid *p; 108 struct sysctl_oid *q; 109 110 /* 111 * Finish initialization from sysctl_set or add. 112 */ 113 lockinit(&oidp->oid_lock, "oidlk", 0, LK_CANRECURSE); 114 115 /* 116 * First check if another oid with the same name already 117 * exists in the parent's list. 118 */ 119 p = sysctl_find_oidname(oidp->oid_name, parent, 0); 120 if (p != NULL) { 121 if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) 122 p->oid_refcnt++; 123 else 124 kprintf("can't re-use a leaf (%s)!\n", p->oid_name); 125 return; 126 } 127 128 /* 129 * If this oid has a number OID_AUTO, give it a number which 130 * is greater than any current oid. Make sure it is at least 131 * 256 to leave space for pre-assigned oid numbers. 132 */ 133 if (oidp->oid_number == OID_AUTO) { 134 int newoid = 0x100; /* minimum AUTO oid */ 135 136 /* 137 * Adjust based on highest oid in parent list 138 */ 139 SLIST_FOREACH(p, parent, oid_link) { 140 if (newoid <= p->oid_number) 141 newoid = p->oid_number + 1; 142 } 143 oidp->oid_number = newoid; 144 } 145 146 /* 147 * Insert the oid into the parent's list in order. 148 */ 149 q = NULL; 150 SLIST_FOREACH(p, parent, oid_link) { 151 if (oidp->oid_number < p->oid_number) 152 break; 153 q = p; 154 } 155 if (q) 156 SLIST_INSERT_AFTER(q, oidp, oid_link); 157 else 158 SLIST_INSERT_HEAD(parent, oidp, oid_link); 159 } 160 161 void 162 sysctl_unregister_oid(struct sysctl_oid *oidp) 163 { 164 SYSCTL_XLOCK(); 165 sysctl_unregister_oid_int(oidp); 166 SYSCTL_XUNLOCK(); 167 } 168 169 static void 170 sysctl_unregister_oid_int(struct sysctl_oid *oidp) 171 { 172 struct sysctl_oid *p; 173 174 if (oidp->oid_number == OID_AUTO) 175 panic("Trying to unregister OID_AUTO entry: %p", oidp); 176 177 SLIST_FOREACH(p, oidp->oid_parent, oid_link) { 178 if (p != oidp) 179 continue; 180 SLIST_REMOVE(oidp->oid_parent, oidp, sysctl_oid, oid_link); 181 return; 182 } 183 184 /* 185 * This can happen when a module fails to register and is 186 * being unloaded afterwards. It should not be a panic() 187 * for normal use. 188 */ 189 kprintf("%s: failed to unregister sysctl\n", __func__); 190 } 191 192 /* Initialize a new context to keep track of dynamically added sysctls. */ 193 int 194 sysctl_ctx_init(struct sysctl_ctx_list *c) 195 { 196 if (c == NULL) 197 return(EINVAL); 198 TAILQ_INIT(c); 199 return(0); 200 } 201 202 /* Free the context, and destroy all dynamic oids registered in this context */ 203 int 204 sysctl_ctx_free(struct sysctl_ctx_list *clist) 205 { 206 struct sysctl_ctx_entry *e, *e1; 207 int error; 208 209 error = 0; 210 /* 211 * First perform a "dry run" to check if it's ok to remove oids. 212 * XXX FIXME 213 * XXX This algorithm is a hack. But I don't know any 214 * XXX better solution for now... 215 */ 216 SYSCTL_XLOCK(); 217 TAILQ_FOREACH(e, clist, link) { 218 error = sysctl_remove_oid_locked(e->entry, 0, 0); 219 if (error) 220 break; 221 } 222 /* 223 * Restore deregistered entries, either from the end, 224 * or from the place where error occured. 225 * e contains the entry that was not unregistered 226 */ 227 if (error) 228 e1 = TAILQ_PREV(e, sysctl_ctx_list, link); 229 else 230 e1 = TAILQ_LAST(clist, sysctl_ctx_list); 231 while (e1 != NULL) { 232 sysctl_register_oid(e1->entry); 233 e1 = TAILQ_PREV(e1, sysctl_ctx_list, link); 234 } 235 if (error) { 236 SYSCTL_XUNLOCK(); 237 return(EBUSY); 238 } 239 /* Now really delete the entries */ 240 e = TAILQ_FIRST(clist); 241 while (e != NULL) { 242 e1 = TAILQ_NEXT(e, link); 243 error = sysctl_remove_oid_locked(e->entry, 1, 0); 244 if (error) 245 panic("sysctl_remove_oid: corrupt tree, entry: %s", 246 e->entry->oid_name); 247 kfree(e, M_SYSCTLOID); 248 e = e1; 249 } 250 SYSCTL_XUNLOCK(); 251 return (error); 252 } 253 254 /* Add an entry to the context */ 255 struct sysctl_ctx_entry * 256 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 257 { 258 struct sysctl_ctx_entry *e; 259 260 SYSCTL_ASSERT_LOCKED(); 261 if (clist == NULL || oidp == NULL) 262 return(NULL); 263 e = kmalloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK); 264 e->entry = oidp; 265 TAILQ_INSERT_HEAD(clist, e, link); 266 return (e); 267 } 268 269 /* Find an entry in the context */ 270 struct sysctl_ctx_entry * 271 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 272 { 273 struct sysctl_ctx_entry *e; 274 275 SYSCTL_ASSERT_LOCKED(); 276 if (clist == NULL || oidp == NULL) 277 return(NULL); 278 TAILQ_FOREACH(e, clist, link) { 279 if(e->entry == oidp) 280 return(e); 281 } 282 return (e); 283 } 284 285 /* 286 * Delete an entry from the context. 287 * NOTE: this function doesn't free oidp! You have to remove it 288 * with sysctl_remove_oid(). 289 */ 290 int 291 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 292 { 293 struct sysctl_ctx_entry *e; 294 295 if (clist == NULL || oidp == NULL) 296 return (EINVAL); 297 SYSCTL_XLOCK(); 298 e = sysctl_ctx_entry_find(clist, oidp); 299 if (e != NULL) { 300 TAILQ_REMOVE(clist, e, link); 301 SYSCTL_XUNLOCK(); 302 kfree(e, M_SYSCTLOID); 303 return (0); 304 } else { 305 SYSCTL_XUNLOCK(); 306 return (ENOENT); 307 } 308 } 309 310 /* 311 * Remove dynamically created sysctl trees. 312 * oidp - top of the tree to be removed 313 * del - if 0 - just deregister, otherwise free up entries as well 314 * recurse - if != 0 traverse the subtree to be deleted 315 */ 316 int 317 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse) 318 { 319 int error; 320 321 SYSCTL_XLOCK(); 322 error = sysctl_remove_oid_locked(oidp, del, recurse); 323 SYSCTL_XUNLOCK(); 324 return (error); 325 } 326 327 static int 328 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse) 329 { 330 struct sysctl_oid *p, *tmp; 331 int error; 332 333 SYSCTL_ASSERT_LOCKED(); 334 if (oidp == NULL) 335 return(EINVAL); 336 if ((oidp->oid_kind & CTLFLAG_DYN) == 0) { 337 kprintf("can't remove non-dynamic nodes!\n"); 338 return (EINVAL); 339 } 340 /* 341 * WARNING: normal method to do this should be through 342 * sysctl_ctx_free(). Use recursing as the last resort 343 * method to purge your sysctl tree of leftovers... 344 * However, if some other code still references these nodes, 345 * it will panic. 346 */ 347 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 348 if (oidp->oid_refcnt == 1) { 349 SLIST_FOREACH_MUTABLE(p, 350 SYSCTL_CHILDREN(oidp), oid_link, tmp) { 351 if (!recurse) { 352 kprintf("Warning: failed attempt to " 353 "remove oid %s with child %s\n", 354 oidp->oid_name, p->oid_name); 355 return (ENOTEMPTY); 356 } 357 error = sysctl_remove_oid_locked(p, del, 358 recurse); 359 if (error) 360 return (error); 361 } 362 if (del) 363 kfree(SYSCTL_CHILDREN(oidp), M_SYSCTLOID); 364 } 365 } 366 if (oidp->oid_refcnt > 1 ) { 367 oidp->oid_refcnt--; 368 } else { 369 if (oidp->oid_refcnt == 0) { 370 kprintf("Warning: bad oid_refcnt=%u (%s)!\n", 371 oidp->oid_refcnt, oidp->oid_name); 372 return (EINVAL); 373 } 374 sysctl_unregister_oid(oidp); 375 if (del) { 376 /* 377 * Wait for all threads running the handler to drain. 378 * This preserves the previous behavior when the 379 * sysctl lock was held across a handler invocation, 380 * and is necessary for module unload correctness. 381 */ 382 while (oidp->oid_running > 0) { 383 oidp->oid_kind |= CTLFLAG_DYING; 384 tsleep_interlock(&oidp->oid_running, 0); 385 SYSCTL_XUNLOCK(); 386 tsleep(&oidp->oid_running, PINTERLOCKED, 387 "oidrm", 0); 388 SYSCTL_XLOCK(); 389 } 390 if (oidp->oid_descr) 391 kfree(__DECONST(char *, oidp->oid_descr), 392 M_SYSCTLOID); 393 kfree(__DECONST(char *, oidp->oid_name), M_SYSCTLOID); 394 lockuninit(&oidp->oid_lock); 395 kfree(oidp, M_SYSCTLOID); 396 } 397 } 398 return (0); 399 } 400 401 int 402 sysctl_remove_name(struct sysctl_oid *parent, const char *name, 403 int del, int recurse) 404 { 405 struct sysctl_oid *p, *tmp; 406 int error; 407 408 error = ENOENT; 409 SYSCTL_XLOCK(); 410 SLIST_FOREACH_MUTABLE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) { 411 if (strcmp(p->oid_name, name) == 0) { 412 error = sysctl_remove_oid_locked(p, del, recurse); 413 break; 414 } 415 } 416 SYSCTL_XUNLOCK(); 417 418 return (error); 419 } 420 421 /* 422 * Create new sysctls at run time. 423 * clist may point to a valid context initialized with sysctl_ctx_init(). 424 */ 425 struct sysctl_oid * 426 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent, 427 int number, const char *name, int kind, void *arg1, int arg2, 428 int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr) 429 { 430 struct sysctl_oid *oidp; 431 ssize_t len; 432 char *newname; 433 434 /* You have to hook up somewhere.. */ 435 if (parent == NULL) 436 return(NULL); 437 SYSCTL_XLOCK(); 438 /* Check if the node already exists, otherwise create it */ 439 oidp = sysctl_find_oidname(name, parent, 0); 440 if (oidp != NULL) { 441 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 442 oidp->oid_refcnt++; 443 /* Update the context */ 444 if (clist != NULL) 445 sysctl_ctx_entry_add(clist, oidp); 446 SYSCTL_XUNLOCK(); 447 return (oidp); 448 } else { 449 kprintf("can't re-use a leaf (%s)!\n", name); 450 SYSCTL_XUNLOCK(); 451 return (NULL); 452 } 453 } 454 oidp = kmalloc(sizeof(struct sysctl_oid), M_SYSCTLOID, 455 M_WAITOK | M_ZERO); 456 oidp->oid_parent = parent; 457 SLIST_NEXT(oidp, oid_link) = NULL; 458 oidp->oid_number = number; 459 oidp->oid_refcnt = 1; 460 len = strlen(name); 461 newname = kmalloc(len + 1, M_SYSCTLOID, M_WAITOK); 462 bcopy(name, newname, len + 1); 463 newname[len] = '\0'; 464 oidp->oid_name = newname; 465 oidp->oid_handler = handler; 466 oidp->oid_kind = CTLFLAG_DYN | kind; 467 if ((kind & CTLTYPE) == CTLTYPE_NODE) { 468 struct sysctl_oid_list *children; 469 470 /* Allocate space for children */ 471 children = kmalloc(sizeof(*children), M_SYSCTLOID, M_WAITOK); 472 SYSCTL_SET_CHILDREN(oidp, children); 473 SLIST_INIT(children); 474 } else { 475 oidp->oid_arg1 = arg1; 476 oidp->oid_arg2 = arg2; 477 } 478 oidp->oid_fmt = fmt; 479 if (descr) { 480 int len = strlen(descr) + 1; 481 oidp->oid_descr = kmalloc(len, M_SYSCTLOID, M_WAITOK); 482 strcpy((char *)(uintptr_t)(const void *)oidp->oid_descr, descr); 483 }; 484 /* Update the context, if used */ 485 if (clist != NULL) 486 sysctl_ctx_entry_add(clist, oidp); 487 /* Register this oid */ 488 sysctl_register_oid_int(oidp); 489 SYSCTL_XUNLOCK(); 490 return (oidp); 491 } 492 493 /* 494 * Rename an existing oid. 495 */ 496 void 497 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name) 498 { 499 char *newname; 500 char *oldname; 501 502 newname = kstrdup(name, M_SYSCTLOID); 503 SYSCTL_XLOCK(); 504 oldname = __DECONST(char *, oidp->oid_name); 505 oidp->oid_name = newname; 506 SYSCTL_XUNLOCK(); 507 kfree(oldname, M_SYSCTLOID); 508 } 509 510 /* 511 * Register the kernel's oids on startup. 512 */ 513 SET_DECLARE(sysctl_set, struct sysctl_oid); 514 515 static void 516 sysctl_register_all(void *arg) 517 { 518 struct sysctl_oid **oidp; 519 520 SYSCTL_XLOCK(); 521 SET_FOREACH(oidp, sysctl_set) 522 sysctl_register_oid(*oidp); 523 SYSCTL_XUNLOCK(); 524 } 525 SYSINIT(sysctl, SI_BOOT1_POST, SI_ORDER_ANY, sysctl_register_all, 0); 526 527 /* 528 * "Staff-functions" 529 * 530 * These functions implement a presently undocumented interface 531 * used by the sysctl program to walk the tree, and get the type 532 * so it can print the value. 533 * This interface is under work and consideration, and should probably 534 * be killed with a big axe by the first person who can find the time. 535 * (be aware though, that the proper interface isn't as obvious as it 536 * may seem, there are various conflicting requirements. 537 * 538 * {0,0} kprintf the entire MIB-tree. 539 * {0,1,...} return the name of the "..." OID. 540 * {0,2,...} return the next OID. 541 * {0,3} return the OID of the name in "new" 542 * {0,4,...} return the kind & format info for the "..." OID. 543 */ 544 545 static void 546 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i) 547 { 548 int k; 549 struct sysctl_oid *oidp; 550 551 SLIST_FOREACH(oidp, l, oid_link) { 552 553 for (k=0; k<i; k++) 554 kprintf(" "); 555 556 kprintf("%d %s ", oidp->oid_number, oidp->oid_name); 557 558 kprintf("%c%c", 559 oidp->oid_kind & CTLFLAG_RD ? 'R':' ', 560 oidp->oid_kind & CTLFLAG_WR ? 'W':' '); 561 562 if (oidp->oid_handler) 563 kprintf(" *Handler"); 564 565 switch (oidp->oid_kind & CTLTYPE) { 566 case CTLTYPE_NODE: 567 kprintf(" Node\n"); 568 if (!oidp->oid_handler) { 569 sysctl_sysctl_debug_dump_node( 570 oidp->oid_arg1, i+2); 571 } 572 break; 573 case CTLTYPE_INT: kprintf(" Int\n"); break; 574 case CTLTYPE_STRING: kprintf(" String\n"); break; 575 case CTLTYPE_QUAD: kprintf(" Quad\n"); break; 576 case CTLTYPE_U32: kprintf(" uint32_t\n"); break; 577 case CTLTYPE_OPAQUE: kprintf(" Opaque/struct\n"); break; 578 default: kprintf("\n"); 579 } 580 581 } 582 } 583 584 static int 585 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS) 586 { 587 int error; 588 589 error = priv_check(req->td, PRIV_SYSCTL_DEBUG); 590 if (error) 591 return (error); 592 sysctl_sysctl_debug_dump_node(&sysctl__children, 0); 593 594 return (ENOENT); 595 } 596 597 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING | CTLFLAG_RD, 598 0, 0, sysctl_sysctl_debug, "-", ""); 599 600 static int 601 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS) 602 { 603 int *name = (int *) arg1; 604 u_int namelen = arg2; 605 int error = 0; 606 struct sysctl_oid *oid; 607 struct sysctl_oid_list *lsp = &sysctl__children, *lsp2; 608 char buf[10]; 609 610 while (namelen) { 611 if (!lsp) { 612 ksnprintf(buf, sizeof(buf), "%d", *name); 613 if (req->oldidx) 614 error = SYSCTL_OUT(req, ".", 1); 615 if (!error) 616 error = SYSCTL_OUT(req, buf, strlen(buf)); 617 if (error) 618 goto out; 619 namelen--; 620 name++; 621 continue; 622 } 623 lsp2 = NULL; 624 SLIST_FOREACH(oid, lsp, oid_link) { 625 if (oid->oid_number != *name) 626 continue; 627 628 if (req->oldidx) 629 error = SYSCTL_OUT(req, ".", 1); 630 if (!error) 631 error = SYSCTL_OUT(req, oid->oid_name, 632 strlen(oid->oid_name)); 633 if (error) 634 goto out; 635 636 namelen--; 637 name++; 638 639 if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE) 640 break; 641 642 if (oid->oid_handler) 643 break; 644 645 lsp2 = SYSCTL_CHILDREN(oid); 646 break; 647 } 648 lsp = lsp2; 649 } 650 error = SYSCTL_OUT(req, "", 1); 651 out: 652 return (error); 653 } 654 655 SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_NOLOCK, 656 sysctl_sysctl_name, ""); 657 658 static int 659 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen, 660 int *next, int *len, int level, struct sysctl_oid **oidpp) 661 { 662 struct sysctl_oid *oidp; 663 664 *len = level; 665 SLIST_FOREACH(oidp, lsp, oid_link) { 666 *next = oidp->oid_number; 667 *oidpp = oidp; 668 669 if (oidp->oid_kind & CTLFLAG_SKIP) 670 continue; 671 672 if (!namelen) { 673 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 674 return (0); 675 if (oidp->oid_handler) 676 /* We really should call the handler here...*/ 677 return (0); 678 lsp = SYSCTL_CHILDREN(oidp); 679 if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1, 680 len, level+1, oidpp)) 681 return (0); 682 goto emptynode; 683 } 684 685 if (oidp->oid_number < *name) 686 continue; 687 688 if (oidp->oid_number > *name) { 689 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 690 return (0); 691 if (oidp->oid_handler) 692 return (0); 693 lsp = SYSCTL_CHILDREN(oidp); 694 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, 695 next+1, len, level+1, oidpp)) 696 return (0); 697 goto next; 698 } 699 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 700 continue; 701 702 if (oidp->oid_handler) 703 continue; 704 705 lsp = SYSCTL_CHILDREN(oidp); 706 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, 707 len, level+1, oidpp)) 708 return (0); 709 next: 710 namelen = 1; 711 emptynode: 712 *len = level; 713 } 714 return (1); 715 } 716 717 static int 718 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS) 719 { 720 int *name = (int *) arg1; 721 u_int namelen = arg2; 722 int i, j, error; 723 struct sysctl_oid *oid; 724 struct sysctl_oid_list *lsp = &sysctl__children; 725 int newoid[CTL_MAXNAME]; 726 727 i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid); 728 if (i) 729 return ENOENT; 730 error = SYSCTL_OUT(req, newoid, j * sizeof (int)); 731 732 return (error); 733 } 734 735 SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_NOLOCK, 736 sysctl_sysctl_next, ""); 737 738 static int 739 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp) 740 { 741 struct sysctl_oid *oidp; 742 struct sysctl_oid_list *lsp = &sysctl__children; 743 char *p; 744 745 SYSCTL_ASSERT_LOCKED(); 746 747 for (*len = 0; *len < CTL_MAXNAME;) { 748 p = strsep(&name, "."); 749 750 oidp = SLIST_FIRST(lsp); 751 for (;; oidp = SLIST_NEXT(oidp, oid_link)) { 752 if (oidp == NULL) 753 return (ENOENT); 754 if (strcmp(p, oidp->oid_name) == 0) 755 break; 756 } 757 *oid++ = oidp->oid_number; 758 (*len)++; 759 760 if (name == NULL || *name == '\0') { 761 if (oidpp) 762 *oidpp = oidp; 763 return (0); 764 } 765 766 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 767 break; 768 769 if (oidp->oid_handler) 770 break; 771 772 lsp = SYSCTL_CHILDREN(oidp); 773 } 774 return (ENOENT); 775 } 776 777 static int 778 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS) 779 { 780 char *p; 781 int error, oid[CTL_MAXNAME], len; 782 struct sysctl_oid *op = NULL; 783 784 if (!req->newlen) 785 return ENOENT; 786 if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */ 787 return (ENAMETOOLONG); 788 789 p = kmalloc(req->newlen+1, M_SYSCTL, M_WAITOK); 790 791 error = SYSCTL_IN(req, p, req->newlen); 792 if (error) { 793 kfree(p, M_SYSCTL); 794 return (error); 795 } 796 797 p [req->newlen] = '\0'; 798 799 error = name2oid(p, oid, &len, &op); 800 801 kfree(p, M_SYSCTL); 802 803 if (error) 804 return (error); 805 806 error = SYSCTL_OUT(req, oid, len * sizeof *oid); 807 return (error); 808 } 809 810 SYSCTL_PROC(_sysctl, 3, name2oid, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK, 811 0, 0, sysctl_sysctl_name2oid, "I", ""); 812 813 static int 814 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS) 815 { 816 struct sysctl_oid *oid; 817 int error; 818 819 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 820 if (error) 821 return (error); 822 823 if (!oid->oid_fmt) 824 return (ENOENT); 825 error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind)); 826 if (error) 827 return (error); 828 error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1); 829 return (error); 830 } 831 832 833 SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD | CTLFLAG_NOLOCK, 834 sysctl_sysctl_oidfmt, ""); 835 836 static int 837 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS) 838 { 839 struct sysctl_oid *oid; 840 int error; 841 842 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 843 if (error) 844 return (error); 845 846 if (!oid->oid_descr) 847 return (ENOENT); 848 error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1); 849 return (error); 850 } 851 852 SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD | CTLFLAG_NOLOCK, 853 sysctl_sysctl_oiddescr, ""); 854 855 /* 856 * Default "handler" functions. 857 */ 858 859 /* 860 * Handle a 32-bit number, signed or unsigned. arg1 points to it. 861 */ 862 863 int 864 sysctl_handle_32(SYSCTL_HANDLER_ARGS) 865 { 866 int error = 0; 867 868 if (!arg1) 869 return (EINVAL); 870 error = SYSCTL_OUT(req, arg1, sizeof(int32_t)); 871 872 if (error || !req->newptr) 873 return (error); 874 875 error = SYSCTL_IN(req, arg1, sizeof(int32_t)); 876 return (error); 877 } 878 879 /* 880 * Handle an int, signed or unsigned. 881 * Two cases: 882 * a variable: point arg1 at it. 883 * a constant: pass it in arg2. 884 */ 885 886 int 887 sysctl_handle_int(SYSCTL_HANDLER_ARGS) 888 { 889 int error = 0; 890 891 if (arg1) 892 error = SYSCTL_OUT(req, arg1, sizeof(int)); 893 else 894 error = SYSCTL_OUT(req, &arg2, sizeof(int)); 895 896 if (error || !req->newptr) 897 return (error); 898 899 if (!arg1) 900 error = EPERM; 901 else 902 error = SYSCTL_IN(req, arg1, sizeof(int)); 903 return (error); 904 } 905 906 /* 907 * Handle a long, signed or unsigned. arg1 points to it. 908 */ 909 910 int 911 sysctl_handle_long(SYSCTL_HANDLER_ARGS) 912 { 913 int error = 0; 914 915 if (!arg1) 916 return (EINVAL); 917 if (req->oldlen == sizeof(int) && 918 *(long *)arg1 >= INT_MIN && 919 *(long *)arg1 <= INT_MAX) { 920 /* 921 * Backwards compatibility for read-only fields promoted 922 * from int to long. Allow userland to request the field 923 * as an integer if the value is in-range. 924 */ 925 int val = (int)*(long *)arg1; 926 error = SYSCTL_OUT(req, &val, sizeof(int)); 927 } else { 928 /* 929 * Normal operation fo a long 930 */ 931 error = SYSCTL_OUT(req, arg1, sizeof(long)); 932 } 933 934 if (error || !req->newptr) 935 return (error); 936 937 error = SYSCTL_IN(req, arg1, sizeof(long)); 938 939 return (error); 940 } 941 942 /* 943 * Handle a quad, signed or unsigned. arg1 points to it. 944 */ 945 946 int 947 sysctl_handle_quad(SYSCTL_HANDLER_ARGS) 948 { 949 int error = 0; 950 951 if (!arg1) 952 return (EINVAL); 953 error = SYSCTL_OUT(req, arg1, sizeof(quad_t)); 954 955 if (error || !req->newptr) 956 return (error); 957 958 error = SYSCTL_IN(req, arg1, sizeof(quad_t)); 959 return (error); 960 } 961 962 /* 963 * Handle our generic '\0' terminated 'C' string. 964 * Two cases: 965 * a variable string: point arg1 at it, arg2 is max length. 966 * a constant string: point arg1 at it, arg2 is zero. 967 */ 968 969 int 970 sysctl_handle_string(SYSCTL_HANDLER_ARGS) 971 { 972 int error=0; 973 974 error = SYSCTL_OUT(req, arg1, strlen((char *)arg1)+1); 975 976 if (error || !req->newptr) 977 return (error); 978 979 if ((req->newlen - req->newidx) >= arg2) { 980 error = EINVAL; 981 } else { 982 arg2 = (req->newlen - req->newidx); 983 error = SYSCTL_IN(req, arg1, arg2); 984 ((char *)arg1)[arg2] = '\0'; 985 } 986 987 return (error); 988 } 989 990 /* 991 * Handle any kind of opaque data. 992 * arg1 points to it, arg2 is the size. 993 */ 994 995 int 996 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS) 997 { 998 int error; 999 1000 error = SYSCTL_OUT(req, arg1, arg2); 1001 1002 if (error || !req->newptr) 1003 return (error); 1004 1005 error = SYSCTL_IN(req, arg1, arg2); 1006 1007 return (error); 1008 } 1009 1010 /* 1011 * Transfer functions to/from kernel space. 1012 * XXX: rather untested at this point 1013 */ 1014 static int 1015 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l) 1016 { 1017 size_t i = 0; 1018 1019 if (req->oldptr) { 1020 i = l; 1021 if (i > req->oldlen - req->oldidx) 1022 i = req->oldlen - req->oldidx; 1023 if (i > 0) 1024 bcopy(p, (char *)req->oldptr + req->oldidx, i); 1025 } 1026 req->oldidx += l; 1027 if (req->oldptr && i != l) 1028 return (ENOMEM); 1029 return (0); 1030 } 1031 1032 static int 1033 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l) 1034 { 1035 1036 if (!req->newptr) 1037 return 0; 1038 if (req->newlen - req->newidx < l) 1039 return (EINVAL); 1040 bcopy((char *)req->newptr + req->newidx, p, l); 1041 req->newidx += l; 1042 return (0); 1043 } 1044 1045 int 1046 kernel_sysctl(int *name, u_int namelen, 1047 void *old, size_t *oldlenp, 1048 void *new, size_t newlen, size_t *retval) 1049 { 1050 int error = 0; 1051 struct sysctl_req req; 1052 1053 bzero(&req, sizeof req); 1054 1055 req.td = curthread; 1056 1057 if (oldlenp) { 1058 req.oldlen = *oldlenp; 1059 } 1060 req.validlen = req.oldlen; 1061 1062 if (old) { 1063 req.oldptr= old; 1064 } 1065 1066 if (new != NULL) { 1067 req.newlen = newlen; 1068 req.newptr = new; 1069 } 1070 1071 req.oldfunc = sysctl_old_kernel; 1072 req.newfunc = sysctl_new_kernel; 1073 #if 0 1074 req.lock = REQ_UNWIRED; 1075 #endif 1076 1077 SYSCTL_SLOCK(); 1078 error = sysctl_root(0, name, namelen, &req); 1079 SYSCTL_SUNLOCK(); 1080 1081 #if 0 1082 if (req.lock == REQ_WIRED && req.validlen > 0) 1083 vsunlock(req.oldptr, req.validlen); 1084 #endif 1085 1086 if (error && error != ENOMEM) 1087 return (error); 1088 1089 if (retval) { 1090 if (req.oldptr && req.oldidx > req.validlen) 1091 *retval = req.validlen; 1092 else 1093 *retval = req.oldidx; 1094 } 1095 return (error); 1096 } 1097 1098 int 1099 kernel_sysctlbyname(char *name, 1100 void *old, size_t *oldlenp, 1101 void *new, size_t newlen, size_t *retval) 1102 { 1103 int oid[CTL_MAXNAME]; 1104 size_t oidlen, plen; 1105 int error; 1106 1107 oid[0] = 0; /* sysctl internal magic */ 1108 oid[1] = 3; /* name2oid */ 1109 oidlen = sizeof(oid); 1110 1111 error = kernel_sysctl(oid, 2, oid, &oidlen, name, strlen(name), &plen); 1112 if (error) 1113 return (error); 1114 1115 error = kernel_sysctl(oid, plen / sizeof(int), old, oldlenp, 1116 new, newlen, retval); 1117 return (error); 1118 } 1119 1120 /* 1121 * Transfer function to/from user space. 1122 */ 1123 static int 1124 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l) 1125 { 1126 int error = 0; 1127 size_t i = 0; 1128 1129 #if 0 1130 if (req->lock == 1 && req->oldptr) { 1131 vslock(req->oldptr, req->oldlen); 1132 req->lock = 2; 1133 } 1134 #endif 1135 if (req->oldptr) { 1136 i = l; 1137 if (i > req->oldlen - req->oldidx) 1138 i = req->oldlen - req->oldidx; 1139 if (i > 0) 1140 error = copyout(p, (char *)req->oldptr + req->oldidx, 1141 i); 1142 } 1143 req->oldidx += l; 1144 if (error) 1145 return (error); 1146 if (req->oldptr && i < l) 1147 return (ENOMEM); 1148 return (0); 1149 } 1150 1151 static int 1152 sysctl_new_user(struct sysctl_req *req, void *p, size_t l) 1153 { 1154 int error; 1155 1156 if (!req->newptr) 1157 return 0; 1158 if (req->newlen - req->newidx < l) 1159 return (EINVAL); 1160 error = copyin((char *)req->newptr + req->newidx, p, l); 1161 req->newidx += l; 1162 return (error); 1163 } 1164 1165 int 1166 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid, 1167 int *nindx, struct sysctl_req *req) 1168 { 1169 struct sysctl_oid_list *lsp; 1170 struct sysctl_oid *oid; 1171 int indx; 1172 1173 lsp = &sysctl__children; 1174 indx = 0; 1175 while (indx < CTL_MAXNAME) { 1176 SLIST_FOREACH(oid, lsp, oid_link) { 1177 if (oid->oid_number == name[indx]) 1178 break; 1179 } 1180 if (oid == NULL) 1181 return (ENOENT); 1182 1183 indx++; 1184 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1185 if (oid->oid_handler != NULL || indx == namelen) { 1186 *noid = oid; 1187 if (nindx != NULL) 1188 *nindx = indx; 1189 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1190 ("%s found DYING node %p", __func__, oid)); 1191 return (0); 1192 } 1193 lsp = SYSCTL_CHILDREN(oid); 1194 } else if (indx == namelen) { 1195 *noid = oid; 1196 if (nindx != NULL) 1197 *nindx = indx; 1198 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1199 ("%s found DYING node %p", __func__, oid)); 1200 return (0); 1201 } else { 1202 return (ENOTDIR); 1203 } 1204 } 1205 return (ENOENT); 1206 } 1207 1208 /* 1209 * Traverse our tree, and find the right node, execute whatever it points 1210 * to, and return the resulting error code. 1211 */ 1212 static int 1213 sysctl_root(SYSCTL_HANDLER_ARGS) 1214 { 1215 struct thread *td = req->td; 1216 struct proc *p = td ? td->td_proc : NULL; 1217 struct sysctl_oid *oid; 1218 int error, indx; 1219 int lktype; 1220 1221 error = sysctl_find_oid(arg1, arg2, &oid, &indx, req); 1222 if (error) 1223 return (error); 1224 1225 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1226 /* 1227 * You can't call a sysctl when it's a node, but has 1228 * no handler. Inform the user that it's a node. 1229 * The indx may or may not be the same as namelen. 1230 */ 1231 if (oid->oid_handler == NULL) 1232 return (EISDIR); 1233 } 1234 1235 /* If writing isn't allowed */ 1236 if (req->newptr && (!(oid->oid_kind & CTLFLAG_WR) || 1237 ((oid->oid_kind & CTLFLAG_SECURE) && securelevel > 0))) 1238 return (EPERM); 1239 1240 /* Most likely only root can write */ 1241 if (!(oid->oid_kind & CTLFLAG_ANYBODY) && req->newptr && p && 1242 (error = priv_check_cred(td->td_ucred, 1243 (oid->oid_kind & CTLFLAG_PRISON) ? PRIV_SYSCTL_WRITEJAIL : 1244 PRIV_SYSCTL_WRITE, 0))) 1245 return (error); 1246 1247 if (oid->oid_handler == NULL) 1248 return EINVAL; 1249 1250 /* 1251 * Default oid locking is exclusive when modifying (newptr), 1252 * shared otherwise, unless overridden with a control flag. 1253 */ 1254 if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) { 1255 lktype = (req->newptr != NULL) ? LK_EXCLUSIVE : LK_SHARED; 1256 if (oid->oid_kind & CTLFLAG_SHLOCK) 1257 lktype = LK_SHARED; 1258 if (oid->oid_kind & CTLFLAG_EXLOCK) 1259 lktype = LK_EXCLUSIVE; 1260 #if 1 1261 lockmgr(&oid->oid_lock, lktype); 1262 #else 1263 /* DEBUGGING */ 1264 if (lockmgr(&oid->oid_lock, lktype | LK_SLEEPFAIL)) { 1265 kprintf("%s\n", oid->oid_name); 1266 lockmgr(&oid->oid_lock, lktype); 1267 } 1268 #endif 1269 } 1270 1271 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) 1272 error = oid->oid_handler(oid, (int *)arg1 + indx, arg2 - indx, 1273 req); 1274 else 1275 error = oid->oid_handler(oid, oid->oid_arg1, oid->oid_arg2, 1276 req); 1277 1278 if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) 1279 lockmgr(&oid->oid_lock, LK_RELEASE); 1280 return (error); 1281 } 1282 1283 int 1284 sys___sysctl(struct sysctl_args *uap) 1285 { 1286 int error, i, name[CTL_MAXNAME]; 1287 size_t j; 1288 1289 if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) 1290 return (EINVAL); 1291 1292 error = copyin(uap->name, &name, uap->namelen * sizeof(int)); 1293 if (error) 1294 return (error); 1295 1296 error = userland_sysctl(name, uap->namelen, 1297 uap->old, uap->oldlenp, 0, 1298 uap->new, uap->newlen, &j); 1299 if (error && error != ENOMEM) 1300 return (error); 1301 if (uap->oldlenp) { 1302 i = copyout(&j, uap->oldlenp, sizeof(j)); 1303 if (i) 1304 return (i); 1305 } 1306 return (error); 1307 } 1308 1309 /* 1310 * This is used from various compatibility syscalls too. That's why name 1311 * must be in kernel space. 1312 */ 1313 int 1314 userland_sysctl(int *name, u_int namelen, 1315 void *old, size_t *oldlenp, int inkernel, 1316 void *new, size_t newlen, size_t *retval) 1317 { 1318 int error = 0; 1319 struct sysctl_req req; 1320 1321 bzero(&req, sizeof req); 1322 1323 req.td = curthread; 1324 req.flags = 0; 1325 1326 if (oldlenp) { 1327 if (inkernel) { 1328 req.oldlen = *oldlenp; 1329 } else { 1330 error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp)); 1331 if (error) 1332 return (error); 1333 } 1334 } 1335 req.validlen = req.oldlen; 1336 1337 /* 1338 * NOTE: User supplied buffers are not guaranteed to be good, 1339 * the sysctl copyins and copyouts can fail. 1340 */ 1341 if (old) 1342 req.oldptr= old; 1343 1344 if (new != NULL) { 1345 req.newlen = newlen; 1346 req.newptr = new; 1347 } 1348 1349 req.oldfunc = sysctl_old_user; 1350 req.newfunc = sysctl_new_user; 1351 #if 0 1352 req.lock = REQ_UNWIRED; 1353 #endif 1354 1355 #ifdef KTRACE 1356 if (KTRPOINT(curthread, KTR_SYSCTL)) 1357 ktrsysctl(name, namelen); 1358 #endif 1359 1360 for (;;) { 1361 req.oldidx = 0; 1362 req.newidx = 0; 1363 SYSCTL_SLOCK(); 1364 error = sysctl_root(0, name, namelen, &req); 1365 SYSCTL_SUNLOCK(); 1366 if (error != EAGAIN) 1367 break; 1368 lwkt_yield(); 1369 } 1370 1371 #if 0 1372 if (req.lock == REQ_WIRED && req.validlen > 0) 1373 vsunlock(req.oldptr, req.validlen); 1374 #endif 1375 if (error && error != ENOMEM) 1376 return (error); 1377 1378 if (retval) { 1379 if (req.oldptr && req.oldidx > req.validlen) 1380 *retval = req.validlen; 1381 else 1382 *retval = req.oldidx; 1383 } 1384 return (error); 1385 } 1386 1387 int 1388 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high) 1389 { 1390 int error, value; 1391 1392 value = *(int *)arg1; 1393 error = sysctl_handle_int(oidp, &value, 0, req); 1394 if (error || !req->newptr) 1395 return (error); 1396 if (value < low || value > high) 1397 return (EINVAL); 1398 *(int *)arg1 = value; 1399 return (0); 1400 } 1401 1402 /* 1403 * Drain into a sysctl struct. The user buffer should be wired if a page 1404 * fault would cause issue. 1405 */ 1406 static int 1407 sbuf_sysctl_drain(void *arg, const char *data, int len) 1408 { 1409 struct sysctl_req *req = arg; 1410 int error; 1411 1412 error = SYSCTL_OUT(req, data, len); 1413 KASSERT(error >= 0, ("Got unexpected negative value %d", error)); 1414 return (error == 0 ? len : -error); 1415 } 1416 1417 struct sbuf * 1418 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length, 1419 struct sysctl_req *req) 1420 { 1421 1422 s = sbuf_new(s, buf, length, SBUF_FIXEDLEN); 1423 sbuf_set_drain(s, sbuf_sysctl_drain, req); 1424 return (s); 1425 } 1426 1427 /* 1428 * The exclusive sysctl lock only protects its topology, and is 1429 * very expensive, but allows us to use a pcpu shared lock for 1430 * critical path accesses. 1431 */ 1432 void 1433 _sysctl_xlock(void) 1434 { 1435 globaldata_t gd; 1436 int i; 1437 1438 for (i = 0; i < ncpus; ++i) { 1439 gd = globaldata_find(i); 1440 lockmgr(&gd->gd_sysctllock, LK_EXCLUSIVE); 1441 } 1442 } 1443 1444 void 1445 _sysctl_xunlock(void) 1446 { 1447 globaldata_t gd; 1448 int i; 1449 1450 for (i = 0; i < ncpus; ++i) { 1451 gd = globaldata_find(i); 1452 lockmgr(&gd->gd_sysctllock, LK_RELEASE); 1453 } 1454 } 1455