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_OPAQUE: kprintf(" Opaque/struct\n"); break; 577 default: kprintf("\n"); 578 } 579 580 } 581 } 582 583 static int 584 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS) 585 { 586 int error; 587 588 error = priv_check(req->td, PRIV_SYSCTL_DEBUG); 589 if (error) 590 return (error); 591 sysctl_sysctl_debug_dump_node(&sysctl__children, 0); 592 593 return (ENOENT); 594 } 595 596 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING | CTLFLAG_RD, 597 0, 0, sysctl_sysctl_debug, "-", ""); 598 599 static int 600 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS) 601 { 602 int *name = (int *) arg1; 603 u_int namelen = arg2; 604 int error = 0; 605 struct sysctl_oid *oid; 606 struct sysctl_oid_list *lsp = &sysctl__children, *lsp2; 607 char buf[10]; 608 609 while (namelen) { 610 if (!lsp) { 611 ksnprintf(buf, sizeof(buf), "%d", *name); 612 if (req->oldidx) 613 error = SYSCTL_OUT(req, ".", 1); 614 if (!error) 615 error = SYSCTL_OUT(req, buf, strlen(buf)); 616 if (error) 617 goto out; 618 namelen--; 619 name++; 620 continue; 621 } 622 lsp2 = NULL; 623 SLIST_FOREACH(oid, lsp, oid_link) { 624 if (oid->oid_number != *name) 625 continue; 626 627 if (req->oldidx) 628 error = SYSCTL_OUT(req, ".", 1); 629 if (!error) 630 error = SYSCTL_OUT(req, oid->oid_name, 631 strlen(oid->oid_name)); 632 if (error) 633 goto out; 634 635 namelen--; 636 name++; 637 638 if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE) 639 break; 640 641 if (oid->oid_handler) 642 break; 643 644 lsp2 = SYSCTL_CHILDREN(oid); 645 break; 646 } 647 lsp = lsp2; 648 } 649 error = SYSCTL_OUT(req, "", 1); 650 out: 651 return (error); 652 } 653 654 SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_NOLOCK, 655 sysctl_sysctl_name, ""); 656 657 static int 658 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen, 659 int *next, int *len, int level, struct sysctl_oid **oidpp) 660 { 661 struct sysctl_oid *oidp; 662 663 *len = level; 664 SLIST_FOREACH(oidp, lsp, oid_link) { 665 *next = oidp->oid_number; 666 *oidpp = oidp; 667 668 if (oidp->oid_kind & CTLFLAG_SKIP) 669 continue; 670 671 if (!namelen) { 672 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 673 return (0); 674 if (oidp->oid_handler) 675 /* We really should call the handler here...*/ 676 return (0); 677 lsp = SYSCTL_CHILDREN(oidp); 678 if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1, 679 len, level+1, oidpp)) 680 return (0); 681 goto emptynode; 682 } 683 684 if (oidp->oid_number < *name) 685 continue; 686 687 if (oidp->oid_number > *name) { 688 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 689 return (0); 690 if (oidp->oid_handler) 691 return (0); 692 lsp = SYSCTL_CHILDREN(oidp); 693 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, 694 next+1, len, level+1, oidpp)) 695 return (0); 696 goto next; 697 } 698 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 699 continue; 700 701 if (oidp->oid_handler) 702 continue; 703 704 lsp = SYSCTL_CHILDREN(oidp); 705 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, 706 len, level+1, oidpp)) 707 return (0); 708 next: 709 namelen = 1; 710 emptynode: 711 *len = level; 712 } 713 return (1); 714 } 715 716 static int 717 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS) 718 { 719 int *name = (int *) arg1; 720 u_int namelen = arg2; 721 int i, j, error; 722 struct sysctl_oid *oid; 723 struct sysctl_oid_list *lsp = &sysctl__children; 724 int newoid[CTL_MAXNAME]; 725 726 i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid); 727 if (i) 728 return ENOENT; 729 error = SYSCTL_OUT(req, newoid, j * sizeof (int)); 730 731 return (error); 732 } 733 734 SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_NOLOCK, 735 sysctl_sysctl_next, ""); 736 737 static int 738 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp) 739 { 740 struct sysctl_oid *oidp; 741 struct sysctl_oid_list *lsp = &sysctl__children; 742 char *p; 743 744 SYSCTL_ASSERT_LOCKED(); 745 746 for (*len = 0; *len < CTL_MAXNAME;) { 747 p = strsep(&name, "."); 748 749 oidp = SLIST_FIRST(lsp); 750 for (;; oidp = SLIST_NEXT(oidp, oid_link)) { 751 if (oidp == NULL) 752 return (ENOENT); 753 if (strcmp(p, oidp->oid_name) == 0) 754 break; 755 } 756 *oid++ = oidp->oid_number; 757 (*len)++; 758 759 if (name == NULL || *name == '\0') { 760 if (oidpp) 761 *oidpp = oidp; 762 return (0); 763 } 764 765 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 766 break; 767 768 if (oidp->oid_handler) 769 break; 770 771 lsp = SYSCTL_CHILDREN(oidp); 772 } 773 return (ENOENT); 774 } 775 776 static int 777 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS) 778 { 779 char *p; 780 int error, oid[CTL_MAXNAME], len; 781 struct sysctl_oid *op = NULL; 782 783 if (!req->newlen) 784 return ENOENT; 785 if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */ 786 return (ENAMETOOLONG); 787 788 p = kmalloc(req->newlen+1, M_SYSCTL, M_WAITOK); 789 790 error = SYSCTL_IN(req, p, req->newlen); 791 if (error) { 792 kfree(p, M_SYSCTL); 793 return (error); 794 } 795 796 p [req->newlen] = '\0'; 797 798 error = name2oid(p, oid, &len, &op); 799 800 kfree(p, M_SYSCTL); 801 802 if (error) 803 return (error); 804 805 error = SYSCTL_OUT(req, oid, len * sizeof *oid); 806 return (error); 807 } 808 809 SYSCTL_PROC(_sysctl, 3, name2oid, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK, 810 0, 0, sysctl_sysctl_name2oid, "I", ""); 811 812 static int 813 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS) 814 { 815 struct sysctl_oid *oid; 816 int error; 817 818 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 819 if (error) 820 return (error); 821 822 if (!oid->oid_fmt) 823 return (ENOENT); 824 error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind)); 825 if (error) 826 return (error); 827 error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1); 828 return (error); 829 } 830 831 832 SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD | CTLFLAG_NOLOCK, 833 sysctl_sysctl_oidfmt, ""); 834 835 static int 836 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS) 837 { 838 struct sysctl_oid *oid; 839 int error; 840 841 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 842 if (error) 843 return (error); 844 845 if (!oid->oid_descr) 846 return (ENOENT); 847 error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1); 848 return (error); 849 } 850 851 SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD | CTLFLAG_NOLOCK, 852 sysctl_sysctl_oiddescr, ""); 853 854 /* 855 * Default "handler" functions. 856 */ 857 858 /* 859 * Handle an int, signed or unsigned. 860 * Two cases: 861 * a variable: point arg1 at it. 862 * a constant: pass it in arg2. 863 */ 864 865 int 866 sysctl_handle_int(SYSCTL_HANDLER_ARGS) 867 { 868 int error = 0; 869 870 if (arg1) 871 error = SYSCTL_OUT(req, arg1, sizeof(int)); 872 else 873 error = SYSCTL_OUT(req, &arg2, sizeof(int)); 874 875 if (error || !req->newptr) 876 return (error); 877 878 if (!arg1) 879 error = EPERM; 880 else 881 error = SYSCTL_IN(req, arg1, sizeof(int)); 882 return (error); 883 } 884 885 /* 886 * Handle a long, signed or unsigned. arg1 points to it. 887 */ 888 889 int 890 sysctl_handle_long(SYSCTL_HANDLER_ARGS) 891 { 892 int error = 0; 893 894 if (!arg1) 895 return (EINVAL); 896 error = SYSCTL_OUT(req, arg1, sizeof(long)); 897 898 if (error || !req->newptr) 899 return (error); 900 901 error = SYSCTL_IN(req, arg1, sizeof(long)); 902 return (error); 903 } 904 905 /* 906 * Handle a quad, signed or unsigned. arg1 points to it. 907 */ 908 909 int 910 sysctl_handle_quad(SYSCTL_HANDLER_ARGS) 911 { 912 int error = 0; 913 914 if (!arg1) 915 return (EINVAL); 916 error = SYSCTL_OUT(req, arg1, sizeof(quad_t)); 917 918 if (error || !req->newptr) 919 return (error); 920 921 error = SYSCTL_IN(req, arg1, sizeof(quad_t)); 922 return (error); 923 } 924 925 /* 926 * Handle our generic '\0' terminated 'C' string. 927 * Two cases: 928 * a variable string: point arg1 at it, arg2 is max length. 929 * a constant string: point arg1 at it, arg2 is zero. 930 */ 931 932 int 933 sysctl_handle_string(SYSCTL_HANDLER_ARGS) 934 { 935 int error=0; 936 937 error = SYSCTL_OUT(req, arg1, strlen((char *)arg1)+1); 938 939 if (error || !req->newptr) 940 return (error); 941 942 if ((req->newlen - req->newidx) >= arg2) { 943 error = EINVAL; 944 } else { 945 arg2 = (req->newlen - req->newidx); 946 error = SYSCTL_IN(req, arg1, arg2); 947 ((char *)arg1)[arg2] = '\0'; 948 } 949 950 return (error); 951 } 952 953 /* 954 * Handle any kind of opaque data. 955 * arg1 points to it, arg2 is the size. 956 */ 957 958 int 959 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS) 960 { 961 int error; 962 963 error = SYSCTL_OUT(req, arg1, arg2); 964 965 if (error || !req->newptr) 966 return (error); 967 968 error = SYSCTL_IN(req, arg1, arg2); 969 970 return (error); 971 } 972 973 /* 974 * Transfer functions to/from kernel space. 975 * XXX: rather untested at this point 976 */ 977 static int 978 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l) 979 { 980 size_t i = 0; 981 982 if (req->oldptr) { 983 i = l; 984 if (i > req->oldlen - req->oldidx) 985 i = req->oldlen - req->oldidx; 986 if (i > 0) 987 bcopy(p, (char *)req->oldptr + req->oldidx, i); 988 } 989 req->oldidx += l; 990 if (req->oldptr && i != l) 991 return (ENOMEM); 992 return (0); 993 } 994 995 static int 996 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l) 997 { 998 999 if (!req->newptr) 1000 return 0; 1001 if (req->newlen - req->newidx < l) 1002 return (EINVAL); 1003 bcopy((char *)req->newptr + req->newidx, p, l); 1004 req->newidx += l; 1005 return (0); 1006 } 1007 1008 int 1009 kernel_sysctl(int *name, u_int namelen, 1010 void *old, size_t *oldlenp, 1011 void *new, size_t newlen, size_t *retval) 1012 { 1013 int error = 0; 1014 struct sysctl_req req; 1015 1016 bzero(&req, sizeof req); 1017 1018 req.td = curthread; 1019 1020 if (oldlenp) { 1021 req.oldlen = *oldlenp; 1022 } 1023 req.validlen = req.oldlen; 1024 1025 if (old) { 1026 req.oldptr= old; 1027 } 1028 1029 if (new != NULL) { 1030 req.newlen = newlen; 1031 req.newptr = new; 1032 } 1033 1034 req.oldfunc = sysctl_old_kernel; 1035 req.newfunc = sysctl_new_kernel; 1036 #if 0 1037 req.lock = REQ_UNWIRED; 1038 #endif 1039 1040 SYSCTL_SLOCK(); 1041 error = sysctl_root(0, name, namelen, &req); 1042 SYSCTL_SUNLOCK(); 1043 1044 #if 0 1045 if (req.lock == REQ_WIRED && req.validlen > 0) 1046 vsunlock(req.oldptr, req.validlen); 1047 #endif 1048 1049 if (error && error != ENOMEM) 1050 return (error); 1051 1052 if (retval) { 1053 if (req.oldptr && req.oldidx > req.validlen) 1054 *retval = req.validlen; 1055 else 1056 *retval = req.oldidx; 1057 } 1058 return (error); 1059 } 1060 1061 int 1062 kernel_sysctlbyname(char *name, 1063 void *old, size_t *oldlenp, 1064 void *new, size_t newlen, size_t *retval) 1065 { 1066 int oid[CTL_MAXNAME]; 1067 size_t oidlen, plen; 1068 int error; 1069 1070 oid[0] = 0; /* sysctl internal magic */ 1071 oid[1] = 3; /* name2oid */ 1072 oidlen = sizeof(oid); 1073 1074 error = kernel_sysctl(oid, 2, oid, &oidlen, name, strlen(name), &plen); 1075 if (error) 1076 return (error); 1077 1078 error = kernel_sysctl(oid, plen / sizeof(int), old, oldlenp, 1079 new, newlen, retval); 1080 return (error); 1081 } 1082 1083 /* 1084 * Transfer function to/from user space. 1085 */ 1086 static int 1087 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l) 1088 { 1089 int error = 0; 1090 size_t i = 0; 1091 1092 #if 0 1093 if (req->lock == 1 && req->oldptr) { 1094 vslock(req->oldptr, req->oldlen); 1095 req->lock = 2; 1096 } 1097 #endif 1098 if (req->oldptr) { 1099 i = l; 1100 if (i > req->oldlen - req->oldidx) 1101 i = req->oldlen - req->oldidx; 1102 if (i > 0) 1103 error = copyout(p, (char *)req->oldptr + req->oldidx, 1104 i); 1105 } 1106 req->oldidx += l; 1107 if (error) 1108 return (error); 1109 if (req->oldptr && i < l) 1110 return (ENOMEM); 1111 return (0); 1112 } 1113 1114 static int 1115 sysctl_new_user(struct sysctl_req *req, void *p, size_t l) 1116 { 1117 int error; 1118 1119 if (!req->newptr) 1120 return 0; 1121 if (req->newlen - req->newidx < l) 1122 return (EINVAL); 1123 error = copyin((char *)req->newptr + req->newidx, p, l); 1124 req->newidx += l; 1125 return (error); 1126 } 1127 1128 int 1129 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid, 1130 int *nindx, struct sysctl_req *req) 1131 { 1132 struct sysctl_oid_list *lsp; 1133 struct sysctl_oid *oid; 1134 int indx; 1135 1136 lsp = &sysctl__children; 1137 indx = 0; 1138 while (indx < CTL_MAXNAME) { 1139 SLIST_FOREACH(oid, lsp, oid_link) { 1140 if (oid->oid_number == name[indx]) 1141 break; 1142 } 1143 if (oid == NULL) 1144 return (ENOENT); 1145 1146 indx++; 1147 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1148 if (oid->oid_handler != NULL || indx == namelen) { 1149 *noid = oid; 1150 if (nindx != NULL) 1151 *nindx = indx; 1152 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1153 ("%s found DYING node %p", __func__, oid)); 1154 return (0); 1155 } 1156 lsp = SYSCTL_CHILDREN(oid); 1157 } else if (indx == namelen) { 1158 *noid = oid; 1159 if (nindx != NULL) 1160 *nindx = indx; 1161 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1162 ("%s found DYING node %p", __func__, oid)); 1163 return (0); 1164 } else { 1165 return (ENOTDIR); 1166 } 1167 } 1168 return (ENOENT); 1169 } 1170 1171 /* 1172 * Traverse our tree, and find the right node, execute whatever it points 1173 * to, and return the resulting error code. 1174 */ 1175 static int 1176 sysctl_root(SYSCTL_HANDLER_ARGS) 1177 { 1178 struct thread *td = req->td; 1179 struct proc *p = td ? td->td_proc : NULL; 1180 struct sysctl_oid *oid; 1181 int error, indx; 1182 int lktype; 1183 1184 error = sysctl_find_oid(arg1, arg2, &oid, &indx, req); 1185 if (error) 1186 return (error); 1187 1188 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1189 /* 1190 * You can't call a sysctl when it's a node, but has 1191 * no handler. Inform the user that it's a node. 1192 * The indx may or may not be the same as namelen. 1193 */ 1194 if (oid->oid_handler == NULL) 1195 return (EISDIR); 1196 } 1197 1198 /* If writing isn't allowed */ 1199 if (req->newptr && (!(oid->oid_kind & CTLFLAG_WR) || 1200 ((oid->oid_kind & CTLFLAG_SECURE) && securelevel > 0))) 1201 return (EPERM); 1202 1203 /* Most likely only root can write */ 1204 if (!(oid->oid_kind & CTLFLAG_ANYBODY) && req->newptr && p && 1205 (error = priv_check_cred(td->td_ucred, 1206 (oid->oid_kind & CTLFLAG_PRISON) ? PRIV_SYSCTL_WRITEJAIL : 1207 PRIV_SYSCTL_WRITE, 0))) 1208 return (error); 1209 1210 if (oid->oid_handler == NULL) 1211 return EINVAL; 1212 1213 /* 1214 * Default oid locking is exclusive when modifying (newptr), 1215 * shared otherwise, unless overridden with a control flag. 1216 */ 1217 if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) { 1218 lktype = (req->newptr != NULL) ? LK_EXCLUSIVE : LK_SHARED; 1219 if (oid->oid_kind & CTLFLAG_SHLOCK) 1220 lktype = LK_SHARED; 1221 if (oid->oid_kind & CTLFLAG_EXLOCK) 1222 lktype = LK_EXCLUSIVE; 1223 #if 1 1224 lockmgr(&oid->oid_lock, lktype); 1225 #else 1226 /* DEBUGGING */ 1227 if (lockmgr(&oid->oid_lock, lktype | LK_SLEEPFAIL)) { 1228 kprintf("%s\n", oid->oid_name); 1229 lockmgr(&oid->oid_lock, lktype); 1230 } 1231 #endif 1232 } 1233 1234 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) 1235 error = oid->oid_handler(oid, (int *)arg1 + indx, arg2 - indx, 1236 req); 1237 else 1238 error = oid->oid_handler(oid, oid->oid_arg1, oid->oid_arg2, 1239 req); 1240 1241 if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) 1242 lockmgr(&oid->oid_lock, LK_RELEASE); 1243 return (error); 1244 } 1245 1246 int 1247 sys___sysctl(struct sysctl_args *uap) 1248 { 1249 int error, i, name[CTL_MAXNAME]; 1250 size_t j; 1251 1252 if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) 1253 return (EINVAL); 1254 1255 error = copyin(uap->name, &name, uap->namelen * sizeof(int)); 1256 if (error) 1257 return (error); 1258 1259 error = userland_sysctl(name, uap->namelen, 1260 uap->old, uap->oldlenp, 0, 1261 uap->new, uap->newlen, &j); 1262 if (error && error != ENOMEM) 1263 return (error); 1264 if (uap->oldlenp) { 1265 i = copyout(&j, uap->oldlenp, sizeof(j)); 1266 if (i) 1267 return (i); 1268 } 1269 return (error); 1270 } 1271 1272 /* 1273 * This is used from various compatibility syscalls too. That's why name 1274 * must be in kernel space. 1275 */ 1276 int 1277 userland_sysctl(int *name, u_int namelen, 1278 void *old, size_t *oldlenp, int inkernel, 1279 void *new, size_t newlen, size_t *retval) 1280 { 1281 int error = 0; 1282 struct sysctl_req req; 1283 1284 bzero(&req, sizeof req); 1285 1286 req.td = curthread; 1287 req.flags = 0; 1288 1289 if (oldlenp) { 1290 if (inkernel) { 1291 req.oldlen = *oldlenp; 1292 } else { 1293 error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp)); 1294 if (error) 1295 return (error); 1296 } 1297 } 1298 req.validlen = req.oldlen; 1299 1300 /* 1301 * NOTE: User supplied buffers are not guaranteed to be good, 1302 * the sysctl copyins and copyouts can fail. 1303 */ 1304 if (old) 1305 req.oldptr= old; 1306 1307 if (new != NULL) { 1308 req.newlen = newlen; 1309 req.newptr = new; 1310 } 1311 1312 req.oldfunc = sysctl_old_user; 1313 req.newfunc = sysctl_new_user; 1314 #if 0 1315 req.lock = REQ_UNWIRED; 1316 #endif 1317 1318 #ifdef KTRACE 1319 if (KTRPOINT(curthread, KTR_SYSCTL)) 1320 ktrsysctl(name, namelen); 1321 #endif 1322 1323 for (;;) { 1324 req.oldidx = 0; 1325 req.newidx = 0; 1326 SYSCTL_SLOCK(); 1327 error = sysctl_root(0, name, namelen, &req); 1328 SYSCTL_SUNLOCK(); 1329 if (error != EAGAIN) 1330 break; 1331 lwkt_yield(); 1332 } 1333 1334 #if 0 1335 if (req.lock == REQ_WIRED && req.validlen > 0) 1336 vsunlock(req.oldptr, req.validlen); 1337 #endif 1338 if (error && error != ENOMEM) 1339 return (error); 1340 1341 if (retval) { 1342 if (req.oldptr && req.oldidx > req.validlen) 1343 *retval = req.validlen; 1344 else 1345 *retval = req.oldidx; 1346 } 1347 return (error); 1348 } 1349 1350 int 1351 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high) 1352 { 1353 int error, value; 1354 1355 value = *(int *)arg1; 1356 error = sysctl_handle_int(oidp, &value, 0, req); 1357 if (error || !req->newptr) 1358 return (error); 1359 if (value < low || value > high) 1360 return (EINVAL); 1361 *(int *)arg1 = value; 1362 return (0); 1363 } 1364 1365 /* 1366 * Drain into a sysctl struct. The user buffer should be wired if a page 1367 * fault would cause issue. 1368 */ 1369 static int 1370 sbuf_sysctl_drain(void *arg, const char *data, int len) 1371 { 1372 struct sysctl_req *req = arg; 1373 int error; 1374 1375 error = SYSCTL_OUT(req, data, len); 1376 KASSERT(error >= 0, ("Got unexpected negative value %d", error)); 1377 return (error == 0 ? len : -error); 1378 } 1379 1380 struct sbuf * 1381 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length, 1382 struct sysctl_req *req) 1383 { 1384 1385 s = sbuf_new(s, buf, length, SBUF_FIXEDLEN); 1386 sbuf_set_drain(s, sbuf_sysctl_drain, req); 1387 return (s); 1388 } 1389 1390 /* 1391 * The exclusive sysctl lock only protects its topology, and is 1392 * very expensive, but allows us to use a pcpu shared lock for 1393 * critical path accesses. 1394 */ 1395 void 1396 _sysctl_xlock(void) 1397 { 1398 globaldata_t gd; 1399 int i; 1400 1401 for (i = 0; i < ncpus; ++i) { 1402 gd = globaldata_find(i); 1403 lockmgr(&gd->gd_sysctllock, LK_EXCLUSIVE); 1404 } 1405 } 1406 1407 void 1408 _sysctl_xunlock(void) 1409 { 1410 globaldata_t gd; 1411 int i; 1412 1413 for (i = 0; i < ncpus; ++i) { 1414 gd = globaldata_find(i); 1415 lockmgr(&gd->gd_sysctllock, LK_RELEASE); 1416 } 1417 } 1418