kern_sysctl.c revision 335699
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 * 4. 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 */ 37 38#include <sys/cdefs.h> 39__FBSDID("$FreeBSD: stable/11/sys/kern/kern_sysctl.c 335699 2018-06-27 07:24:07Z hselasky $"); 40 41#include "opt_capsicum.h" 42#include "opt_compat.h" 43#include "opt_ktrace.h" 44 45#include <sys/param.h> 46#include <sys/fail.h> 47#include <sys/systm.h> 48#include <sys/capsicum.h> 49#include <sys/kernel.h> 50#include <sys/sysctl.h> 51#include <sys/malloc.h> 52#include <sys/priv.h> 53#include <sys/proc.h> 54#include <sys/jail.h> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/rmlock.h> 58#include <sys/sbuf.h> 59#include <sys/sx.h> 60#include <sys/sysproto.h> 61#include <sys/uio.h> 62#ifdef KTRACE 63#include <sys/ktrace.h> 64#endif 65 66#include <net/vnet.h> 67 68#include <security/mac/mac_framework.h> 69 70#include <vm/vm.h> 71#include <vm/vm_extern.h> 72 73static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic"); 74static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids"); 75static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer"); 76 77/* 78 * The sysctllock protects the MIB tree. It also protects sysctl 79 * contexts used with dynamic sysctls. The sysctl_register_oid() and 80 * sysctl_unregister_oid() routines require the sysctllock to already 81 * be held, so the sysctl_wlock() and sysctl_wunlock() routines are 82 * provided for the few places in the kernel which need to use that 83 * API rather than using the dynamic API. Use of the dynamic API is 84 * strongly encouraged for most code. 85 * 86 * The sysctlmemlock is used to limit the amount of user memory wired for 87 * sysctl requests. This is implemented by serializing any userland 88 * sysctl requests larger than a single page via an exclusive lock. 89 */ 90static struct rmlock sysctllock; 91static struct sx __exclusive_cache_line sysctlmemlock; 92 93#define SYSCTL_WLOCK() rm_wlock(&sysctllock) 94#define SYSCTL_WUNLOCK() rm_wunlock(&sysctllock) 95#define SYSCTL_RLOCK(tracker) rm_rlock(&sysctllock, (tracker)) 96#define SYSCTL_RUNLOCK(tracker) rm_runlock(&sysctllock, (tracker)) 97#define SYSCTL_WLOCKED() rm_wowned(&sysctllock) 98#define SYSCTL_ASSERT_LOCKED() rm_assert(&sysctllock, RA_LOCKED) 99#define SYSCTL_ASSERT_WLOCKED() rm_assert(&sysctllock, RA_WLOCKED) 100#define SYSCTL_ASSERT_RLOCKED() rm_assert(&sysctllock, RA_RLOCKED) 101#define SYSCTL_INIT() rm_init_flags(&sysctllock, "sysctl lock", \ 102 RM_SLEEPABLE) 103#define SYSCTL_SLEEP(ch, wmesg, timo) \ 104 rm_sleep(ch, &sysctllock, 0, wmesg, timo) 105 106static int sysctl_root(SYSCTL_HANDLER_ARGS); 107 108/* Root list */ 109struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children); 110 111static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, 112 int recurse); 113static int sysctl_old_kernel(struct sysctl_req *, const void *, size_t); 114static int sysctl_new_kernel(struct sysctl_req *, void *, size_t); 115 116static struct sysctl_oid * 117sysctl_find_oidname(const char *name, struct sysctl_oid_list *list) 118{ 119 struct sysctl_oid *oidp; 120 121 SYSCTL_ASSERT_LOCKED(); 122 SLIST_FOREACH(oidp, list, oid_link) { 123 if (strcmp(oidp->oid_name, name) == 0) { 124 return (oidp); 125 } 126 } 127 return (NULL); 128} 129 130/* 131 * Initialization of the MIB tree. 132 * 133 * Order by number in each list. 134 */ 135void 136sysctl_wlock(void) 137{ 138 139 SYSCTL_WLOCK(); 140} 141 142void 143sysctl_wunlock(void) 144{ 145 146 SYSCTL_WUNLOCK(); 147} 148 149static int 150sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2, 151 struct sysctl_req *req, struct rm_priotracker *tracker) 152{ 153 int error; 154 155 if (oid->oid_kind & CTLFLAG_DYN) 156 atomic_add_int(&oid->oid_running, 1); 157 158 if (tracker != NULL) 159 SYSCTL_RUNLOCK(tracker); 160 else 161 SYSCTL_WUNLOCK(); 162 163 if (!(oid->oid_kind & CTLFLAG_MPSAFE)) 164 mtx_lock(&Giant); 165 error = oid->oid_handler(oid, arg1, arg2, req); 166 if (!(oid->oid_kind & CTLFLAG_MPSAFE)) 167 mtx_unlock(&Giant); 168 169 KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error); 170 171 if (tracker != NULL) 172 SYSCTL_RLOCK(tracker); 173 else 174 SYSCTL_WLOCK(); 175 176 if (oid->oid_kind & CTLFLAG_DYN) { 177 if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 && 178 (oid->oid_kind & CTLFLAG_DYING) != 0) 179 wakeup(&oid->oid_running); 180 } 181 182 return (error); 183} 184 185static void 186sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp) 187{ 188 struct sysctl_req req; 189 struct sysctl_oid *curr; 190 char *penv = NULL; 191 char path[96]; 192 ssize_t rem = sizeof(path); 193 ssize_t len; 194 uint8_t data[512] __aligned(sizeof(uint64_t)); 195 int size; 196 int error; 197 198 path[--rem] = 0; 199 200 for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) { 201 len = strlen(curr->oid_name); 202 rem -= len; 203 if (curr != oidp) 204 rem -= 1; 205 if (rem < 0) { 206 printf("OID path exceeds %d bytes\n", (int)sizeof(path)); 207 return; 208 } 209 memcpy(path + rem, curr->oid_name, len); 210 if (curr != oidp) 211 path[rem + len] = '.'; 212 } 213 214 memset(&req, 0, sizeof(req)); 215 216 req.td = curthread; 217 req.oldfunc = sysctl_old_kernel; 218 req.newfunc = sysctl_new_kernel; 219 req.lock = REQ_UNWIRED; 220 221 switch (oidp->oid_kind & CTLTYPE) { 222 case CTLTYPE_INT: 223 if (getenv_array(path + rem, data, sizeof(data), &size, 224 sizeof(int), GETENV_SIGNED) == 0) 225 return; 226 req.newlen = size; 227 req.newptr = data; 228 break; 229 case CTLTYPE_UINT: 230 if (getenv_array(path + rem, data, sizeof(data), &size, 231 sizeof(int), GETENV_UNSIGNED) == 0) 232 return; 233 req.newlen = size; 234 req.newptr = data; 235 break; 236 case CTLTYPE_LONG: 237 if (getenv_array(path + rem, data, sizeof(data), &size, 238 sizeof(long), GETENV_SIGNED) == 0) 239 return; 240 req.newlen = size; 241 req.newptr = data; 242 break; 243 case CTLTYPE_ULONG: 244 if (getenv_array(path + rem, data, sizeof(data), &size, 245 sizeof(long), GETENV_UNSIGNED) == 0) 246 return; 247 req.newlen = size; 248 req.newptr = data; 249 break; 250 case CTLTYPE_S8: 251 if (getenv_array(path + rem, data, sizeof(data), &size, 252 sizeof(int8_t), GETENV_SIGNED) == 0) 253 return; 254 req.newlen = size; 255 req.newptr = data; 256 break; 257 case CTLTYPE_S16: 258 if (getenv_array(path + rem, data, sizeof(data), &size, 259 sizeof(int16_t), GETENV_SIGNED) == 0) 260 return; 261 req.newlen = size; 262 req.newptr = data; 263 break; 264 case CTLTYPE_S32: 265 if (getenv_array(path + rem, data, sizeof(data), &size, 266 sizeof(int32_t), GETENV_SIGNED) == 0) 267 return; 268 req.newlen = size; 269 req.newptr = data; 270 break; 271 case CTLTYPE_S64: 272 if (getenv_array(path + rem, data, sizeof(data), &size, 273 sizeof(int64_t), GETENV_SIGNED) == 0) 274 return; 275 req.newlen = size; 276 req.newptr = data; 277 break; 278 case CTLTYPE_U8: 279 if (getenv_array(path + rem, data, sizeof(data), &size, 280 sizeof(uint8_t), GETENV_UNSIGNED) == 0) 281 return; 282 req.newlen = size; 283 req.newptr = data; 284 break; 285 case CTLTYPE_U16: 286 if (getenv_array(path + rem, data, sizeof(data), &size, 287 sizeof(uint16_t), GETENV_UNSIGNED) == 0) 288 return; 289 req.newlen = size; 290 req.newptr = data; 291 break; 292 case CTLTYPE_U32: 293 if (getenv_array(path + rem, data, sizeof(data), &size, 294 sizeof(uint32_t), GETENV_UNSIGNED) == 0) 295 return; 296 req.newlen = size; 297 req.newptr = data; 298 break; 299 case CTLTYPE_U64: 300 if (getenv_array(path + rem, data, sizeof(data), &size, 301 sizeof(uint64_t), GETENV_UNSIGNED) == 0) 302 return; 303 req.newlen = size; 304 req.newptr = data; 305 break; 306 case CTLTYPE_STRING: 307 penv = kern_getenv(path + rem); 308 if (penv == NULL) 309 return; 310 req.newlen = strlen(penv); 311 req.newptr = penv; 312 break; 313 default: 314 return; 315 } 316 error = sysctl_root_handler_locked(oidp, oidp->oid_arg1, 317 oidp->oid_arg2, &req, NULL); 318 if (error != 0) 319 printf("Setting sysctl %s failed: %d\n", path + rem, error); 320 if (penv != NULL) 321 freeenv(penv); 322} 323 324void 325sysctl_register_oid(struct sysctl_oid *oidp) 326{ 327 struct sysctl_oid_list *parent = oidp->oid_parent; 328 struct sysctl_oid *p; 329 struct sysctl_oid *q; 330 int oid_number; 331 int timeout = 2; 332 333 /* 334 * First check if another oid with the same name already 335 * exists in the parent's list. 336 */ 337 SYSCTL_ASSERT_WLOCKED(); 338 p = sysctl_find_oidname(oidp->oid_name, parent); 339 if (p != NULL) { 340 if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 341 p->oid_refcnt++; 342 return; 343 } else { 344 printf("can't re-use a leaf (%s)!\n", p->oid_name); 345 return; 346 } 347 } 348 /* get current OID number */ 349 oid_number = oidp->oid_number; 350 351#if (OID_AUTO >= 0) 352#error "OID_AUTO is expected to be a negative value" 353#endif 354 /* 355 * Any negative OID number qualifies as OID_AUTO. Valid OID 356 * numbers should always be positive. 357 * 358 * NOTE: DO NOT change the starting value here, change it in 359 * <sys/sysctl.h>, and make sure it is at least 256 to 360 * accommodate e.g. net.inet.raw as a static sysctl node. 361 */ 362 if (oid_number < 0) { 363 static int newoid; 364 365 /* 366 * By decrementing the next OID number we spend less 367 * time inserting the OIDs into a sorted list. 368 */ 369 if (--newoid < CTL_AUTO_START) 370 newoid = 0x7fffffff; 371 372 oid_number = newoid; 373 } 374 375 /* 376 * Insert the OID into the parent's list sorted by OID number. 377 */ 378retry: 379 q = NULL; 380 SLIST_FOREACH(p, parent, oid_link) { 381 /* check if the current OID number is in use */ 382 if (oid_number == p->oid_number) { 383 /* get the next valid OID number */ 384 if (oid_number < CTL_AUTO_START || 385 oid_number == 0x7fffffff) { 386 /* wraparound - restart */ 387 oid_number = CTL_AUTO_START; 388 /* don't loop forever */ 389 if (!timeout--) 390 panic("sysctl: Out of OID numbers\n"); 391 goto retry; 392 } else { 393 oid_number++; 394 } 395 } else if (oid_number < p->oid_number) 396 break; 397 q = p; 398 } 399 /* check for non-auto OID number collision */ 400 if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START && 401 oid_number >= CTL_AUTO_START) { 402 printf("sysctl: OID number(%d) is already in use for '%s'\n", 403 oidp->oid_number, oidp->oid_name); 404 } 405 /* update the OID number, if any */ 406 oidp->oid_number = oid_number; 407 if (q != NULL) 408 SLIST_INSERT_AFTER(q, oidp, oid_link); 409 else 410 SLIST_INSERT_HEAD(parent, oidp, oid_link); 411 412 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE && 413#ifdef VIMAGE 414 (oidp->oid_kind & CTLFLAG_VNET) == 0 && 415#endif 416 (oidp->oid_kind & CTLFLAG_TUN) != 0 && 417 (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) { 418 /* only fetch value once */ 419 oidp->oid_kind |= CTLFLAG_NOFETCH; 420 /* try to fetch value from kernel environment */ 421 sysctl_load_tunable_by_oid_locked(oidp); 422 } 423} 424 425void 426sysctl_register_disabled_oid(struct sysctl_oid *oidp) 427{ 428 429 /* 430 * Mark the leaf as dormant if it's not to be immediately enabled. 431 * We do not disable nodes as they can be shared between modules 432 * and it is always safe to access a node. 433 */ 434 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0, 435 ("internal flag is set in oid_kind")); 436 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 437 oidp->oid_kind |= CTLFLAG_DORMANT; 438 sysctl_register_oid(oidp); 439} 440 441void 442sysctl_enable_oid(struct sysctl_oid *oidp) 443{ 444 445 SYSCTL_ASSERT_WLOCKED(); 446 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 447 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0, 448 ("sysctl node is marked as dormant")); 449 return; 450 } 451 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0, 452 ("enabling already enabled sysctl oid")); 453 oidp->oid_kind &= ~CTLFLAG_DORMANT; 454} 455 456void 457sysctl_unregister_oid(struct sysctl_oid *oidp) 458{ 459 struct sysctl_oid *p; 460 int error; 461 462 SYSCTL_ASSERT_WLOCKED(); 463 error = ENOENT; 464 if (oidp->oid_number == OID_AUTO) { 465 error = EINVAL; 466 } else { 467 SLIST_FOREACH(p, oidp->oid_parent, oid_link) { 468 if (p == oidp) { 469 SLIST_REMOVE(oidp->oid_parent, oidp, 470 sysctl_oid, oid_link); 471 error = 0; 472 break; 473 } 474 } 475 } 476 477 /* 478 * This can happen when a module fails to register and is 479 * being unloaded afterwards. It should not be a panic() 480 * for normal use. 481 */ 482 if (error) 483 printf("%s: failed to unregister sysctl\n", __func__); 484} 485 486/* Initialize a new context to keep track of dynamically added sysctls. */ 487int 488sysctl_ctx_init(struct sysctl_ctx_list *c) 489{ 490 491 if (c == NULL) { 492 return (EINVAL); 493 } 494 495 /* 496 * No locking here, the caller is responsible for not adding 497 * new nodes to a context until after this function has 498 * returned. 499 */ 500 TAILQ_INIT(c); 501 return (0); 502} 503 504/* Free the context, and destroy all dynamic oids registered in this context */ 505int 506sysctl_ctx_free(struct sysctl_ctx_list *clist) 507{ 508 struct sysctl_ctx_entry *e, *e1; 509 int error; 510 511 error = 0; 512 /* 513 * First perform a "dry run" to check if it's ok to remove oids. 514 * XXX FIXME 515 * XXX This algorithm is a hack. But I don't know any 516 * XXX better solution for now... 517 */ 518 SYSCTL_WLOCK(); 519 TAILQ_FOREACH(e, clist, link) { 520 error = sysctl_remove_oid_locked(e->entry, 0, 0); 521 if (error) 522 break; 523 } 524 /* 525 * Restore deregistered entries, either from the end, 526 * or from the place where error occurred. 527 * e contains the entry that was not unregistered 528 */ 529 if (error) 530 e1 = TAILQ_PREV(e, sysctl_ctx_list, link); 531 else 532 e1 = TAILQ_LAST(clist, sysctl_ctx_list); 533 while (e1 != NULL) { 534 sysctl_register_oid(e1->entry); 535 e1 = TAILQ_PREV(e1, sysctl_ctx_list, link); 536 } 537 if (error) { 538 SYSCTL_WUNLOCK(); 539 return(EBUSY); 540 } 541 /* Now really delete the entries */ 542 e = TAILQ_FIRST(clist); 543 while (e != NULL) { 544 e1 = TAILQ_NEXT(e, link); 545 error = sysctl_remove_oid_locked(e->entry, 1, 0); 546 if (error) 547 panic("sysctl_remove_oid: corrupt tree, entry: %s", 548 e->entry->oid_name); 549 free(e, M_SYSCTLOID); 550 e = e1; 551 } 552 SYSCTL_WUNLOCK(); 553 return (error); 554} 555 556/* Add an entry to the context */ 557struct sysctl_ctx_entry * 558sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 559{ 560 struct sysctl_ctx_entry *e; 561 562 SYSCTL_ASSERT_WLOCKED(); 563 if (clist == NULL || oidp == NULL) 564 return(NULL); 565 e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK); 566 e->entry = oidp; 567 TAILQ_INSERT_HEAD(clist, e, link); 568 return (e); 569} 570 571/* Find an entry in the context */ 572struct sysctl_ctx_entry * 573sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 574{ 575 struct sysctl_ctx_entry *e; 576 577 SYSCTL_ASSERT_WLOCKED(); 578 if (clist == NULL || oidp == NULL) 579 return(NULL); 580 TAILQ_FOREACH(e, clist, link) { 581 if(e->entry == oidp) 582 return(e); 583 } 584 return (e); 585} 586 587/* 588 * Delete an entry from the context. 589 * NOTE: this function doesn't free oidp! You have to remove it 590 * with sysctl_remove_oid(). 591 */ 592int 593sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 594{ 595 struct sysctl_ctx_entry *e; 596 597 if (clist == NULL || oidp == NULL) 598 return (EINVAL); 599 SYSCTL_WLOCK(); 600 e = sysctl_ctx_entry_find(clist, oidp); 601 if (e != NULL) { 602 TAILQ_REMOVE(clist, e, link); 603 SYSCTL_WUNLOCK(); 604 free(e, M_SYSCTLOID); 605 return (0); 606 } else { 607 SYSCTL_WUNLOCK(); 608 return (ENOENT); 609 } 610} 611 612/* 613 * Remove dynamically created sysctl trees. 614 * oidp - top of the tree to be removed 615 * del - if 0 - just deregister, otherwise free up entries as well 616 * recurse - if != 0 traverse the subtree to be deleted 617 */ 618int 619sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse) 620{ 621 int error; 622 623 SYSCTL_WLOCK(); 624 error = sysctl_remove_oid_locked(oidp, del, recurse); 625 SYSCTL_WUNLOCK(); 626 return (error); 627} 628 629int 630sysctl_remove_name(struct sysctl_oid *parent, const char *name, 631 int del, int recurse) 632{ 633 struct sysctl_oid *p, *tmp; 634 int error; 635 636 error = ENOENT; 637 SYSCTL_WLOCK(); 638 SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) { 639 if (strcmp(p->oid_name, name) == 0) { 640 error = sysctl_remove_oid_locked(p, del, recurse); 641 break; 642 } 643 } 644 SYSCTL_WUNLOCK(); 645 646 return (error); 647} 648 649 650static int 651sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse) 652{ 653 struct sysctl_oid *p, *tmp; 654 int error; 655 656 SYSCTL_ASSERT_WLOCKED(); 657 if (oidp == NULL) 658 return(EINVAL); 659 if ((oidp->oid_kind & CTLFLAG_DYN) == 0) { 660 printf("Warning: can't remove non-dynamic nodes (%s)!\n", 661 oidp->oid_name); 662 return (EINVAL); 663 } 664 /* 665 * WARNING: normal method to do this should be through 666 * sysctl_ctx_free(). Use recursing as the last resort 667 * method to purge your sysctl tree of leftovers... 668 * However, if some other code still references these nodes, 669 * it will panic. 670 */ 671 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 672 if (oidp->oid_refcnt == 1) { 673 SLIST_FOREACH_SAFE(p, 674 SYSCTL_CHILDREN(oidp), oid_link, tmp) { 675 if (!recurse) { 676 printf("Warning: failed attempt to " 677 "remove oid %s with child %s\n", 678 oidp->oid_name, p->oid_name); 679 return (ENOTEMPTY); 680 } 681 error = sysctl_remove_oid_locked(p, del, 682 recurse); 683 if (error) 684 return (error); 685 } 686 } 687 } 688 if (oidp->oid_refcnt > 1 ) { 689 oidp->oid_refcnt--; 690 } else { 691 if (oidp->oid_refcnt == 0) { 692 printf("Warning: bad oid_refcnt=%u (%s)!\n", 693 oidp->oid_refcnt, oidp->oid_name); 694 return (EINVAL); 695 } 696 sysctl_unregister_oid(oidp); 697 if (del) { 698 /* 699 * Wait for all threads running the handler to drain. 700 * This preserves the previous behavior when the 701 * sysctl lock was held across a handler invocation, 702 * and is necessary for module unload correctness. 703 */ 704 while (oidp->oid_running > 0) { 705 oidp->oid_kind |= CTLFLAG_DYING; 706 SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0); 707 } 708 if (oidp->oid_descr) 709 free(__DECONST(char *, oidp->oid_descr), 710 M_SYSCTLOID); 711 free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID); 712 free(oidp, M_SYSCTLOID); 713 } 714 } 715 return (0); 716} 717/* 718 * Create new sysctls at run time. 719 * clist may point to a valid context initialized with sysctl_ctx_init(). 720 */ 721struct sysctl_oid * 722sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent, 723 int number, const char *name, int kind, void *arg1, intmax_t arg2, 724 int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr) 725{ 726 struct sysctl_oid *oidp; 727 728 /* You have to hook up somewhere.. */ 729 if (parent == NULL) 730 return(NULL); 731 /* Check if the node already exists, otherwise create it */ 732 SYSCTL_WLOCK(); 733 oidp = sysctl_find_oidname(name, parent); 734 if (oidp != NULL) { 735 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 736 oidp->oid_refcnt++; 737 /* Update the context */ 738 if (clist != NULL) 739 sysctl_ctx_entry_add(clist, oidp); 740 SYSCTL_WUNLOCK(); 741 return (oidp); 742 } else { 743 SYSCTL_WUNLOCK(); 744 printf("can't re-use a leaf (%s)!\n", name); 745 return (NULL); 746 } 747 } 748 oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO); 749 oidp->oid_parent = parent; 750 SLIST_INIT(&oidp->oid_children); 751 oidp->oid_number = number; 752 oidp->oid_refcnt = 1; 753 oidp->oid_name = strdup(name, M_SYSCTLOID); 754 oidp->oid_handler = handler; 755 oidp->oid_kind = CTLFLAG_DYN | kind; 756 oidp->oid_arg1 = arg1; 757 oidp->oid_arg2 = arg2; 758 oidp->oid_fmt = fmt; 759 if (descr != NULL) 760 oidp->oid_descr = strdup(descr, M_SYSCTLOID); 761 /* Update the context, if used */ 762 if (clist != NULL) 763 sysctl_ctx_entry_add(clist, oidp); 764 /* Register this oid */ 765 sysctl_register_oid(oidp); 766 SYSCTL_WUNLOCK(); 767 return (oidp); 768} 769 770/* 771 * Rename an existing oid. 772 */ 773void 774sysctl_rename_oid(struct sysctl_oid *oidp, const char *name) 775{ 776 char *newname; 777 char *oldname; 778 779 newname = strdup(name, M_SYSCTLOID); 780 SYSCTL_WLOCK(); 781 oldname = __DECONST(char *, oidp->oid_name); 782 oidp->oid_name = newname; 783 SYSCTL_WUNLOCK(); 784 free(oldname, M_SYSCTLOID); 785} 786 787/* 788 * Reparent an existing oid. 789 */ 790int 791sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent) 792{ 793 struct sysctl_oid *oidp; 794 795 SYSCTL_WLOCK(); 796 if (oid->oid_parent == parent) { 797 SYSCTL_WUNLOCK(); 798 return (0); 799 } 800 oidp = sysctl_find_oidname(oid->oid_name, parent); 801 if (oidp != NULL) { 802 SYSCTL_WUNLOCK(); 803 return (EEXIST); 804 } 805 sysctl_unregister_oid(oid); 806 oid->oid_parent = parent; 807 oid->oid_number = OID_AUTO; 808 sysctl_register_oid(oid); 809 SYSCTL_WUNLOCK(); 810 return (0); 811} 812 813/* 814 * Register the kernel's oids on startup. 815 */ 816SET_DECLARE(sysctl_set, struct sysctl_oid); 817 818static void 819sysctl_register_all(void *arg) 820{ 821 struct sysctl_oid **oidp; 822 823 sx_init(&sysctlmemlock, "sysctl mem"); 824 SYSCTL_INIT(); 825 SYSCTL_WLOCK(); 826 SET_FOREACH(oidp, sysctl_set) 827 sysctl_register_oid(*oidp); 828 SYSCTL_WUNLOCK(); 829} 830SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0); 831 832/* 833 * "Staff-functions" 834 * 835 * These functions implement a presently undocumented interface 836 * used by the sysctl program to walk the tree, and get the type 837 * so it can print the value. 838 * This interface is under work and consideration, and should probably 839 * be killed with a big axe by the first person who can find the time. 840 * (be aware though, that the proper interface isn't as obvious as it 841 * may seem, there are various conflicting requirements. 842 * 843 * {0,0} printf the entire MIB-tree. 844 * {0,1,...} return the name of the "..." OID. 845 * {0,2,...} return the next OID. 846 * {0,3} return the OID of the name in "new" 847 * {0,4,...} return the kind & format info for the "..." OID. 848 * {0,5,...} return the description the "..." OID. 849 */ 850 851#ifdef SYSCTL_DEBUG 852static void 853sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i) 854{ 855 int k; 856 struct sysctl_oid *oidp; 857 858 SYSCTL_ASSERT_LOCKED(); 859 SLIST_FOREACH(oidp, l, oid_link) { 860 861 for (k=0; k<i; k++) 862 printf(" "); 863 864 printf("%d %s ", oidp->oid_number, oidp->oid_name); 865 866 printf("%c%c", 867 oidp->oid_kind & CTLFLAG_RD ? 'R':' ', 868 oidp->oid_kind & CTLFLAG_WR ? 'W':' '); 869 870 if (oidp->oid_handler) 871 printf(" *Handler"); 872 873 switch (oidp->oid_kind & CTLTYPE) { 874 case CTLTYPE_NODE: 875 printf(" Node\n"); 876 if (!oidp->oid_handler) { 877 sysctl_sysctl_debug_dump_node( 878 SYSCTL_CHILDREN(oidp), i + 2); 879 } 880 break; 881 case CTLTYPE_INT: printf(" Int\n"); break; 882 case CTLTYPE_UINT: printf(" u_int\n"); break; 883 case CTLTYPE_LONG: printf(" Long\n"); break; 884 case CTLTYPE_ULONG: printf(" u_long\n"); break; 885 case CTLTYPE_STRING: printf(" String\n"); break; 886 case CTLTYPE_S8: printf(" int8_t\n"); break; 887 case CTLTYPE_S16: printf(" int16_t\n"); break; 888 case CTLTYPE_S32: printf(" int32_t\n"); break; 889 case CTLTYPE_S64: printf(" int64_t\n"); break; 890 case CTLTYPE_U8: printf(" uint8_t\n"); break; 891 case CTLTYPE_U16: printf(" uint16_t\n"); break; 892 case CTLTYPE_U32: printf(" uint32_t\n"); break; 893 case CTLTYPE_U64: printf(" uint64_t\n"); break; 894 case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break; 895 default: printf("\n"); 896 } 897 898 } 899} 900 901static int 902sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS) 903{ 904 struct rm_priotracker tracker; 905 int error; 906 907 error = priv_check(req->td, PRIV_SYSCTL_DEBUG); 908 if (error) 909 return (error); 910 SYSCTL_RLOCK(&tracker); 911 sysctl_sysctl_debug_dump_node(&sysctl__children, 0); 912 SYSCTL_RUNLOCK(&tracker); 913 return (ENOENT); 914} 915 916SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE, 917 0, 0, sysctl_sysctl_debug, "-", ""); 918#endif 919 920static int 921sysctl_sysctl_name(SYSCTL_HANDLER_ARGS) 922{ 923 int *name = (int *) arg1; 924 u_int namelen = arg2; 925 int error = 0; 926 struct sysctl_oid *oid; 927 struct sysctl_oid_list *lsp = &sysctl__children, *lsp2; 928 struct rm_priotracker tracker; 929 char buf[10]; 930 931 SYSCTL_RLOCK(&tracker); 932 while (namelen) { 933 if (!lsp) { 934 snprintf(buf,sizeof(buf),"%d",*name); 935 if (req->oldidx) 936 error = SYSCTL_OUT(req, ".", 1); 937 if (!error) 938 error = SYSCTL_OUT(req, buf, strlen(buf)); 939 if (error) 940 goto out; 941 namelen--; 942 name++; 943 continue; 944 } 945 lsp2 = NULL; 946 SLIST_FOREACH(oid, lsp, oid_link) { 947 if (oid->oid_number != *name) 948 continue; 949 950 if (req->oldidx) 951 error = SYSCTL_OUT(req, ".", 1); 952 if (!error) 953 error = SYSCTL_OUT(req, oid->oid_name, 954 strlen(oid->oid_name)); 955 if (error) 956 goto out; 957 958 namelen--; 959 name++; 960 961 if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE) 962 break; 963 964 if (oid->oid_handler) 965 break; 966 967 lsp2 = SYSCTL_CHILDREN(oid); 968 break; 969 } 970 lsp = lsp2; 971 } 972 error = SYSCTL_OUT(req, "", 1); 973 out: 974 SYSCTL_RUNLOCK(&tracker); 975 return (error); 976} 977 978/* 979 * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in 980 * capability mode. 981 */ 982static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, 983 sysctl_sysctl_name, ""); 984 985static int 986sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen, 987 int *next, int *len, int level, struct sysctl_oid **oidpp) 988{ 989 struct sysctl_oid *oidp; 990 991 SYSCTL_ASSERT_LOCKED(); 992 *len = level; 993 SLIST_FOREACH(oidp, lsp, oid_link) { 994 *next = oidp->oid_number; 995 *oidpp = oidp; 996 997 if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0) 998 continue; 999 1000 if (!namelen) { 1001 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 1002 return (0); 1003 if (oidp->oid_handler) 1004 /* We really should call the handler here...*/ 1005 return (0); 1006 lsp = SYSCTL_CHILDREN(oidp); 1007 if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1, 1008 len, level+1, oidpp)) 1009 return (0); 1010 goto emptynode; 1011 } 1012 1013 if (oidp->oid_number < *name) 1014 continue; 1015 1016 if (oidp->oid_number > *name) { 1017 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 1018 return (0); 1019 if (oidp->oid_handler) 1020 return (0); 1021 lsp = SYSCTL_CHILDREN(oidp); 1022 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, 1023 next+1, len, level+1, oidpp)) 1024 return (0); 1025 goto next; 1026 } 1027 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 1028 continue; 1029 1030 if (oidp->oid_handler) 1031 continue; 1032 1033 lsp = SYSCTL_CHILDREN(oidp); 1034 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, 1035 len, level+1, oidpp)) 1036 return (0); 1037 next: 1038 namelen = 1; 1039 emptynode: 1040 *len = level; 1041 } 1042 return (1); 1043} 1044 1045static int 1046sysctl_sysctl_next(SYSCTL_HANDLER_ARGS) 1047{ 1048 int *name = (int *) arg1; 1049 u_int namelen = arg2; 1050 int i, j, error; 1051 struct sysctl_oid *oid; 1052 struct sysctl_oid_list *lsp = &sysctl__children; 1053 struct rm_priotracker tracker; 1054 int newoid[CTL_MAXNAME]; 1055 1056 SYSCTL_RLOCK(&tracker); 1057 i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid); 1058 SYSCTL_RUNLOCK(&tracker); 1059 if (i) 1060 return (ENOENT); 1061 error = SYSCTL_OUT(req, newoid, j * sizeof (int)); 1062 return (error); 1063} 1064 1065/* 1066 * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in 1067 * capability mode. 1068 */ 1069static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, 1070 sysctl_sysctl_next, ""); 1071 1072static int 1073name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp) 1074{ 1075 struct sysctl_oid *oidp; 1076 struct sysctl_oid_list *lsp = &sysctl__children; 1077 char *p; 1078 1079 SYSCTL_ASSERT_LOCKED(); 1080 1081 for (*len = 0; *len < CTL_MAXNAME;) { 1082 p = strsep(&name, "."); 1083 1084 oidp = SLIST_FIRST(lsp); 1085 for (;; oidp = SLIST_NEXT(oidp, oid_link)) { 1086 if (oidp == NULL) 1087 return (ENOENT); 1088 if (strcmp(p, oidp->oid_name) == 0) 1089 break; 1090 } 1091 *oid++ = oidp->oid_number; 1092 (*len)++; 1093 1094 if (name == NULL || *name == '\0') { 1095 if (oidpp) 1096 *oidpp = oidp; 1097 return (0); 1098 } 1099 1100 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 1101 break; 1102 1103 if (oidp->oid_handler) 1104 break; 1105 1106 lsp = SYSCTL_CHILDREN(oidp); 1107 } 1108 return (ENOENT); 1109} 1110 1111static int 1112sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS) 1113{ 1114 char *p; 1115 int error, oid[CTL_MAXNAME], len = 0; 1116 struct sysctl_oid *op = NULL; 1117 struct rm_priotracker tracker; 1118 char buf[32]; 1119 1120 if (!req->newlen) 1121 return (ENOENT); 1122 if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */ 1123 return (ENAMETOOLONG); 1124 1125 p = buf; 1126 if (req->newlen >= sizeof(buf)) 1127 p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK); 1128 1129 error = SYSCTL_IN(req, p, req->newlen); 1130 if (error) { 1131 if (p != buf) 1132 free(p, M_SYSCTL); 1133 return (error); 1134 } 1135 1136 p [req->newlen] = '\0'; 1137 1138 SYSCTL_RLOCK(&tracker); 1139 error = name2oid(p, oid, &len, &op); 1140 SYSCTL_RUNLOCK(&tracker); 1141 1142 if (p != buf) 1143 free(p, M_SYSCTL); 1144 1145 if (error) 1146 return (error); 1147 1148 error = SYSCTL_OUT(req, oid, len * sizeof *oid); 1149 return (error); 1150} 1151 1152/* 1153 * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in 1154 * capability mode. 1155 */ 1156SYSCTL_PROC(_sysctl, 3, name2oid, 1157 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE 1158 | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", ""); 1159 1160static int 1161sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS) 1162{ 1163 struct sysctl_oid *oid; 1164 struct rm_priotracker tracker; 1165 int error; 1166 1167 SYSCTL_RLOCK(&tracker); 1168 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 1169 if (error) 1170 goto out; 1171 1172 if (oid->oid_fmt == NULL) { 1173 error = ENOENT; 1174 goto out; 1175 } 1176 error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind)); 1177 if (error) 1178 goto out; 1179 error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1); 1180 out: 1181 SYSCTL_RUNLOCK(&tracker); 1182 return (error); 1183} 1184 1185 1186static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, 1187 sysctl_sysctl_oidfmt, ""); 1188 1189static int 1190sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS) 1191{ 1192 struct sysctl_oid *oid; 1193 struct rm_priotracker tracker; 1194 int error; 1195 1196 SYSCTL_RLOCK(&tracker); 1197 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 1198 if (error) 1199 goto out; 1200 1201 if (oid->oid_descr == NULL) { 1202 error = ENOENT; 1203 goto out; 1204 } 1205 error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1); 1206 out: 1207 SYSCTL_RUNLOCK(&tracker); 1208 return (error); 1209} 1210 1211static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, 1212 sysctl_sysctl_oiddescr, ""); 1213 1214/* 1215 * Default "handler" functions. 1216 */ 1217 1218/* 1219 * Handle a bool. 1220 * Two cases: 1221 * a variable: point arg1 at it. 1222 * a constant: pass it in arg2. 1223 */ 1224 1225int 1226sysctl_handle_bool(SYSCTL_HANDLER_ARGS) 1227{ 1228 uint8_t temp; 1229 int error; 1230 1231 /* 1232 * Attempt to get a coherent snapshot by making a copy of the data. 1233 */ 1234 if (arg1) 1235 temp = *(bool *)arg1 ? 1 : 0; 1236 else 1237 temp = arg2 ? 1 : 0; 1238 1239 error = SYSCTL_OUT(req, &temp, sizeof(temp)); 1240 if (error || !req->newptr) 1241 return (error); 1242 1243 if (!arg1) 1244 error = EPERM; 1245 else { 1246 error = SYSCTL_IN(req, &temp, sizeof(temp)); 1247 if (!error) 1248 *(bool *)arg1 = temp ? 1 : 0; 1249 } 1250 return (error); 1251} 1252 1253/* 1254 * Handle an int8_t, signed or unsigned. 1255 * Two cases: 1256 * a variable: point arg1 at it. 1257 * a constant: pass it in arg2. 1258 */ 1259 1260int 1261sysctl_handle_8(SYSCTL_HANDLER_ARGS) 1262{ 1263 int8_t tmpout; 1264 int error = 0; 1265 1266 /* 1267 * Attempt to get a coherent snapshot by making a copy of the data. 1268 */ 1269 if (arg1) 1270 tmpout = *(int8_t *)arg1; 1271 else 1272 tmpout = arg2; 1273 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1274 1275 if (error || !req->newptr) 1276 return (error); 1277 1278 if (!arg1) 1279 error = EPERM; 1280 else 1281 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1282 return (error); 1283} 1284 1285/* 1286 * Handle an int16_t, signed or unsigned. 1287 * Two cases: 1288 * a variable: point arg1 at it. 1289 * a constant: pass it in arg2. 1290 */ 1291 1292int 1293sysctl_handle_16(SYSCTL_HANDLER_ARGS) 1294{ 1295 int16_t tmpout; 1296 int error = 0; 1297 1298 /* 1299 * Attempt to get a coherent snapshot by making a copy of the data. 1300 */ 1301 if (arg1) 1302 tmpout = *(int16_t *)arg1; 1303 else 1304 tmpout = arg2; 1305 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1306 1307 if (error || !req->newptr) 1308 return (error); 1309 1310 if (!arg1) 1311 error = EPERM; 1312 else 1313 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1314 return (error); 1315} 1316 1317/* 1318 * Handle an int32_t, signed or unsigned. 1319 * Two cases: 1320 * a variable: point arg1 at it. 1321 * a constant: pass it in arg2. 1322 */ 1323 1324int 1325sysctl_handle_32(SYSCTL_HANDLER_ARGS) 1326{ 1327 int32_t tmpout; 1328 int error = 0; 1329 1330 /* 1331 * Attempt to get a coherent snapshot by making a copy of the data. 1332 */ 1333 if (arg1) 1334 tmpout = *(int32_t *)arg1; 1335 else 1336 tmpout = arg2; 1337 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1338 1339 if (error || !req->newptr) 1340 return (error); 1341 1342 if (!arg1) 1343 error = EPERM; 1344 else 1345 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1346 return (error); 1347} 1348 1349/* 1350 * Handle an int, signed or unsigned. 1351 * Two cases: 1352 * a variable: point arg1 at it. 1353 * a constant: pass it in arg2. 1354 */ 1355 1356int 1357sysctl_handle_int(SYSCTL_HANDLER_ARGS) 1358{ 1359 int tmpout, error = 0; 1360 1361 /* 1362 * Attempt to get a coherent snapshot by making a copy of the data. 1363 */ 1364 if (arg1) 1365 tmpout = *(int *)arg1; 1366 else 1367 tmpout = arg2; 1368 error = SYSCTL_OUT(req, &tmpout, sizeof(int)); 1369 1370 if (error || !req->newptr) 1371 return (error); 1372 1373 if (!arg1) 1374 error = EPERM; 1375 else 1376 error = SYSCTL_IN(req, arg1, sizeof(int)); 1377 return (error); 1378} 1379 1380/* 1381 * Based on on sysctl_handle_int() convert milliseconds into ticks. 1382 * Note: this is used by TCP. 1383 */ 1384 1385int 1386sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS) 1387{ 1388 int error, s, tt; 1389 1390 tt = *(int *)arg1; 1391 s = (int)((int64_t)tt * 1000 / hz); 1392 1393 error = sysctl_handle_int(oidp, &s, 0, req); 1394 if (error || !req->newptr) 1395 return (error); 1396 1397 tt = (int)((int64_t)s * hz / 1000); 1398 if (tt < 1) 1399 return (EINVAL); 1400 1401 *(int *)arg1 = tt; 1402 return (0); 1403} 1404 1405 1406/* 1407 * Handle a long, signed or unsigned. 1408 * Two cases: 1409 * a variable: point arg1 at it. 1410 * a constant: pass it in arg2. 1411 */ 1412 1413int 1414sysctl_handle_long(SYSCTL_HANDLER_ARGS) 1415{ 1416 int error = 0; 1417 long tmplong; 1418#ifdef SCTL_MASK32 1419 int tmpint; 1420#endif 1421 1422 /* 1423 * Attempt to get a coherent snapshot by making a copy of the data. 1424 */ 1425 if (arg1) 1426 tmplong = *(long *)arg1; 1427 else 1428 tmplong = arg2; 1429#ifdef SCTL_MASK32 1430 if (req->flags & SCTL_MASK32) { 1431 tmpint = tmplong; 1432 error = SYSCTL_OUT(req, &tmpint, sizeof(int)); 1433 } else 1434#endif 1435 error = SYSCTL_OUT(req, &tmplong, sizeof(long)); 1436 1437 if (error || !req->newptr) 1438 return (error); 1439 1440 if (!arg1) 1441 error = EPERM; 1442#ifdef SCTL_MASK32 1443 else if (req->flags & SCTL_MASK32) { 1444 error = SYSCTL_IN(req, &tmpint, sizeof(int)); 1445 *(long *)arg1 = (long)tmpint; 1446 } 1447#endif 1448 else 1449 error = SYSCTL_IN(req, arg1, sizeof(long)); 1450 return (error); 1451} 1452 1453/* 1454 * Handle a 64 bit int, signed or unsigned. 1455 * Two cases: 1456 * a variable: point arg1 at it. 1457 * a constant: pass it in arg2. 1458 */ 1459int 1460sysctl_handle_64(SYSCTL_HANDLER_ARGS) 1461{ 1462 int error = 0; 1463 uint64_t tmpout; 1464 1465 /* 1466 * Attempt to get a coherent snapshot by making a copy of the data. 1467 */ 1468 if (arg1) 1469 tmpout = *(uint64_t *)arg1; 1470 else 1471 tmpout = arg2; 1472 error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t)); 1473 1474 if (error || !req->newptr) 1475 return (error); 1476 1477 if (!arg1) 1478 error = EPERM; 1479 else 1480 error = SYSCTL_IN(req, arg1, sizeof(uint64_t)); 1481 return (error); 1482} 1483 1484/* 1485 * Handle our generic '\0' terminated 'C' string. 1486 * Two cases: 1487 * a variable string: point arg1 at it, arg2 is max length. 1488 * a constant string: point arg1 at it, arg2 is zero. 1489 */ 1490 1491int 1492sysctl_handle_string(SYSCTL_HANDLER_ARGS) 1493{ 1494 size_t outlen; 1495 int error = 0, ro_string = 0; 1496 1497 /* 1498 * A zero-length buffer indicates a fixed size read-only 1499 * string: 1500 */ 1501 if (arg2 == 0) { 1502 arg2 = strlen((char *)arg1) + 1; 1503 ro_string = 1; 1504 } 1505 1506 if (req->oldptr != NULL) { 1507 char *tmparg; 1508 1509 if (ro_string) { 1510 tmparg = arg1; 1511 } else { 1512 /* try to make a coherent snapshot of the string */ 1513 tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK); 1514 memcpy(tmparg, arg1, arg2); 1515 } 1516 1517 outlen = strnlen(tmparg, arg2 - 1) + 1; 1518 error = SYSCTL_OUT(req, tmparg, outlen); 1519 1520 if (!ro_string) 1521 free(tmparg, M_SYSCTLTMP); 1522 } else { 1523 outlen = strnlen((char *)arg1, arg2 - 1) + 1; 1524 error = SYSCTL_OUT(req, NULL, outlen); 1525 } 1526 if (error || !req->newptr) 1527 return (error); 1528 1529 if ((req->newlen - req->newidx) >= arg2) { 1530 error = EINVAL; 1531 } else { 1532 arg2 = (req->newlen - req->newidx); 1533 error = SYSCTL_IN(req, arg1, arg2); 1534 ((char *)arg1)[arg2] = '\0'; 1535 } 1536 return (error); 1537} 1538 1539/* 1540 * Handle any kind of opaque data. 1541 * arg1 points to it, arg2 is the size. 1542 */ 1543 1544int 1545sysctl_handle_opaque(SYSCTL_HANDLER_ARGS) 1546{ 1547 int error, tries; 1548 u_int generation; 1549 struct sysctl_req req2; 1550 1551 /* 1552 * Attempt to get a coherent snapshot, by using the thread 1553 * pre-emption counter updated from within mi_switch() to 1554 * determine if we were pre-empted during a bcopy() or 1555 * copyout(). Make 3 attempts at doing this before giving up. 1556 * If we encounter an error, stop immediately. 1557 */ 1558 tries = 0; 1559 req2 = *req; 1560retry: 1561 generation = curthread->td_generation; 1562 error = SYSCTL_OUT(req, arg1, arg2); 1563 if (error) 1564 return (error); 1565 tries++; 1566 if (generation != curthread->td_generation && tries < 3) { 1567 *req = req2; 1568 goto retry; 1569 } 1570 1571 error = SYSCTL_IN(req, arg1, arg2); 1572 1573 return (error); 1574} 1575 1576/* 1577 * Transfer functions to/from kernel space. 1578 * XXX: rather untested at this point 1579 */ 1580static int 1581sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l) 1582{ 1583 size_t i = 0; 1584 1585 if (req->oldptr) { 1586 i = l; 1587 if (req->oldlen <= req->oldidx) 1588 i = 0; 1589 else 1590 if (i > req->oldlen - req->oldidx) 1591 i = req->oldlen - req->oldidx; 1592 if (i > 0) 1593 bcopy(p, (char *)req->oldptr + req->oldidx, i); 1594 } 1595 req->oldidx += l; 1596 if (req->oldptr && i != l) 1597 return (ENOMEM); 1598 return (0); 1599} 1600 1601static int 1602sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l) 1603{ 1604 if (!req->newptr) 1605 return (0); 1606 if (req->newlen - req->newidx < l) 1607 return (EINVAL); 1608 bcopy((char *)req->newptr + req->newidx, p, l); 1609 req->newidx += l; 1610 return (0); 1611} 1612 1613int 1614kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old, 1615 size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags) 1616{ 1617 int error = 0; 1618 struct sysctl_req req; 1619 1620 bzero(&req, sizeof req); 1621 1622 req.td = td; 1623 req.flags = flags; 1624 1625 if (oldlenp) { 1626 req.oldlen = *oldlenp; 1627 } 1628 req.validlen = req.oldlen; 1629 1630 if (old) { 1631 req.oldptr= old; 1632 } 1633 1634 if (new != NULL) { 1635 req.newlen = newlen; 1636 req.newptr = new; 1637 } 1638 1639 req.oldfunc = sysctl_old_kernel; 1640 req.newfunc = sysctl_new_kernel; 1641 req.lock = REQ_UNWIRED; 1642 1643 error = sysctl_root(0, name, namelen, &req); 1644 1645 if (req.lock == REQ_WIRED && req.validlen > 0) 1646 vsunlock(req.oldptr, req.validlen); 1647 1648 if (error && error != ENOMEM) 1649 return (error); 1650 1651 if (retval) { 1652 if (req.oldptr && req.oldidx > req.validlen) 1653 *retval = req.validlen; 1654 else 1655 *retval = req.oldidx; 1656 } 1657 return (error); 1658} 1659 1660int 1661kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp, 1662 void *new, size_t newlen, size_t *retval, int flags) 1663{ 1664 int oid[CTL_MAXNAME]; 1665 size_t oidlen, plen; 1666 int error; 1667 1668 oid[0] = 0; /* sysctl internal magic */ 1669 oid[1] = 3; /* name2oid */ 1670 oidlen = sizeof(oid); 1671 1672 error = kernel_sysctl(td, oid, 2, oid, &oidlen, 1673 (void *)name, strlen(name), &plen, flags); 1674 if (error) 1675 return (error); 1676 1677 error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp, 1678 new, newlen, retval, flags); 1679 return (error); 1680} 1681 1682/* 1683 * Transfer function to/from user space. 1684 */ 1685static int 1686sysctl_old_user(struct sysctl_req *req, const void *p, size_t l) 1687{ 1688 size_t i, len, origidx; 1689 int error; 1690 1691 origidx = req->oldidx; 1692 req->oldidx += l; 1693 if (req->oldptr == NULL) 1694 return (0); 1695 /* 1696 * If we have not wired the user supplied buffer and we are currently 1697 * holding locks, drop a witness warning, as it's possible that 1698 * write operations to the user page can sleep. 1699 */ 1700 if (req->lock != REQ_WIRED) 1701 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 1702 "sysctl_old_user()"); 1703 i = l; 1704 len = req->validlen; 1705 if (len <= origidx) 1706 i = 0; 1707 else { 1708 if (i > len - origidx) 1709 i = len - origidx; 1710 if (req->lock == REQ_WIRED) { 1711 error = copyout_nofault(p, (char *)req->oldptr + 1712 origidx, i); 1713 } else 1714 error = copyout(p, (char *)req->oldptr + origidx, i); 1715 if (error != 0) 1716 return (error); 1717 } 1718 if (i < l) 1719 return (ENOMEM); 1720 return (0); 1721} 1722 1723static int 1724sysctl_new_user(struct sysctl_req *req, void *p, size_t l) 1725{ 1726 int error; 1727 1728 if (!req->newptr) 1729 return (0); 1730 if (req->newlen - req->newidx < l) 1731 return (EINVAL); 1732 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 1733 "sysctl_new_user()"); 1734 error = copyin((char *)req->newptr + req->newidx, p, l); 1735 req->newidx += l; 1736 return (error); 1737} 1738 1739/* 1740 * Wire the user space destination buffer. If set to a value greater than 1741 * zero, the len parameter limits the maximum amount of wired memory. 1742 */ 1743int 1744sysctl_wire_old_buffer(struct sysctl_req *req, size_t len) 1745{ 1746 int ret; 1747 size_t wiredlen; 1748 1749 wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen; 1750 ret = 0; 1751 if (req->lock != REQ_WIRED && req->oldptr && 1752 req->oldfunc == sysctl_old_user) { 1753 if (wiredlen != 0) { 1754 ret = vslock(req->oldptr, wiredlen); 1755 if (ret != 0) { 1756 if (ret != ENOMEM) 1757 return (ret); 1758 wiredlen = 0; 1759 } 1760 } 1761 req->lock = REQ_WIRED; 1762 req->validlen = wiredlen; 1763 } 1764 return (0); 1765} 1766 1767int 1768sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid, 1769 int *nindx, struct sysctl_req *req) 1770{ 1771 struct sysctl_oid_list *lsp; 1772 struct sysctl_oid *oid; 1773 int indx; 1774 1775 SYSCTL_ASSERT_LOCKED(); 1776 lsp = &sysctl__children; 1777 indx = 0; 1778 while (indx < CTL_MAXNAME) { 1779 SLIST_FOREACH(oid, lsp, oid_link) { 1780 if (oid->oid_number == name[indx]) 1781 break; 1782 } 1783 if (oid == NULL) 1784 return (ENOENT); 1785 1786 indx++; 1787 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1788 if (oid->oid_handler != NULL || indx == namelen) { 1789 *noid = oid; 1790 if (nindx != NULL) 1791 *nindx = indx; 1792 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1793 ("%s found DYING node %p", __func__, oid)); 1794 return (0); 1795 } 1796 lsp = SYSCTL_CHILDREN(oid); 1797 } else if (indx == namelen) { 1798 if ((oid->oid_kind & CTLFLAG_DORMANT) != 0) 1799 return (ENOENT); 1800 *noid = oid; 1801 if (nindx != NULL) 1802 *nindx = indx; 1803 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1804 ("%s found DYING node %p", __func__, oid)); 1805 return (0); 1806 } else { 1807 return (ENOTDIR); 1808 } 1809 } 1810 return (ENOENT); 1811} 1812 1813/* 1814 * Traverse our tree, and find the right node, execute whatever it points 1815 * to, and return the resulting error code. 1816 */ 1817 1818static int 1819sysctl_root(SYSCTL_HANDLER_ARGS) 1820{ 1821 struct sysctl_oid *oid; 1822 struct rm_priotracker tracker; 1823 int error, indx, lvl; 1824 1825 SYSCTL_RLOCK(&tracker); 1826 1827 error = sysctl_find_oid(arg1, arg2, &oid, &indx, req); 1828 if (error) 1829 goto out; 1830 1831 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1832 /* 1833 * You can't call a sysctl when it's a node, but has 1834 * no handler. Inform the user that it's a node. 1835 * The indx may or may not be the same as namelen. 1836 */ 1837 if (oid->oid_handler == NULL) { 1838 error = EISDIR; 1839 goto out; 1840 } 1841 } 1842 1843 /* Is this sysctl writable? */ 1844 if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) { 1845 error = EPERM; 1846 goto out; 1847 } 1848 1849 KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL")); 1850 1851#ifdef CAPABILITY_MODE 1852 /* 1853 * If the process is in capability mode, then don't permit reading or 1854 * writing unless specifically granted for the node. 1855 */ 1856 if (IN_CAPABILITY_MODE(req->td)) { 1857 if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) || 1858 (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) { 1859 error = EPERM; 1860 goto out; 1861 } 1862 } 1863#endif 1864 1865 /* Is this sysctl sensitive to securelevels? */ 1866 if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) { 1867 lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE; 1868 error = securelevel_gt(req->td->td_ucred, lvl); 1869 if (error) 1870 goto out; 1871 } 1872 1873 /* Is this sysctl writable by only privileged users? */ 1874 if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) { 1875 int priv; 1876 1877 if (oid->oid_kind & CTLFLAG_PRISON) 1878 priv = PRIV_SYSCTL_WRITEJAIL; 1879#ifdef VIMAGE 1880 else if ((oid->oid_kind & CTLFLAG_VNET) && 1881 prison_owns_vnet(req->td->td_ucred)) 1882 priv = PRIV_SYSCTL_WRITEJAIL; 1883#endif 1884 else 1885 priv = PRIV_SYSCTL_WRITE; 1886 error = priv_check(req->td, priv); 1887 if (error) 1888 goto out; 1889 } 1890 1891 if (!oid->oid_handler) { 1892 error = EINVAL; 1893 goto out; 1894 } 1895 1896 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1897 arg1 = (int *)arg1 + indx; 1898 arg2 -= indx; 1899 } else { 1900 arg1 = oid->oid_arg1; 1901 arg2 = oid->oid_arg2; 1902 } 1903#ifdef MAC 1904 error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2, 1905 req); 1906 if (error != 0) 1907 goto out; 1908#endif 1909#ifdef VIMAGE 1910 if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL) 1911 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1); 1912#endif 1913 error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker); 1914 1915out: 1916 SYSCTL_RUNLOCK(&tracker); 1917 return (error); 1918} 1919 1920#ifndef _SYS_SYSPROTO_H_ 1921struct sysctl_args { 1922 int *name; 1923 u_int namelen; 1924 void *old; 1925 size_t *oldlenp; 1926 void *new; 1927 size_t newlen; 1928}; 1929#endif 1930int 1931sys___sysctl(struct thread *td, struct sysctl_args *uap) 1932{ 1933 int error, i, name[CTL_MAXNAME]; 1934 size_t j; 1935 1936 if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) 1937 return (EINVAL); 1938 1939 error = copyin(uap->name, &name, uap->namelen * sizeof(int)); 1940 if (error) 1941 return (error); 1942 1943 error = userland_sysctl(td, name, uap->namelen, 1944 uap->old, uap->oldlenp, 0, 1945 uap->new, uap->newlen, &j, 0); 1946 if (error && error != ENOMEM) 1947 return (error); 1948 if (uap->oldlenp) { 1949 i = copyout(&j, uap->oldlenp, sizeof(j)); 1950 if (i) 1951 return (i); 1952 } 1953 return (error); 1954} 1955 1956/* 1957 * This is used from various compatibility syscalls too. That's why name 1958 * must be in kernel space. 1959 */ 1960int 1961userland_sysctl(struct thread *td, int *name, u_int namelen, void *old, 1962 size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval, 1963 int flags) 1964{ 1965 int error = 0, memlocked; 1966 struct sysctl_req req; 1967 1968 bzero(&req, sizeof req); 1969 1970 req.td = td; 1971 req.flags = flags; 1972 1973 if (oldlenp) { 1974 if (inkernel) { 1975 req.oldlen = *oldlenp; 1976 } else { 1977 error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp)); 1978 if (error) 1979 return (error); 1980 } 1981 } 1982 req.validlen = req.oldlen; 1983 req.oldptr = old; 1984 1985 if (new != NULL) { 1986 req.newlen = newlen; 1987 req.newptr = new; 1988 } 1989 1990 req.oldfunc = sysctl_old_user; 1991 req.newfunc = sysctl_new_user; 1992 req.lock = REQ_UNWIRED; 1993 1994#ifdef KTRACE 1995 if (KTRPOINT(curthread, KTR_SYSCTL)) 1996 ktrsysctl(name, namelen); 1997#endif 1998 1999 if (req.oldptr && req.oldlen > PAGE_SIZE) { 2000 memlocked = 1; 2001 sx_xlock(&sysctlmemlock); 2002 } else 2003 memlocked = 0; 2004 CURVNET_SET(TD_TO_VNET(td)); 2005 2006 for (;;) { 2007 req.oldidx = 0; 2008 req.newidx = 0; 2009 error = sysctl_root(0, name, namelen, &req); 2010 if (error != EAGAIN) 2011 break; 2012 kern_yield(PRI_USER); 2013 } 2014 2015 CURVNET_RESTORE(); 2016 2017 if (req.lock == REQ_WIRED && req.validlen > 0) 2018 vsunlock(req.oldptr, req.validlen); 2019 if (memlocked) 2020 sx_xunlock(&sysctlmemlock); 2021 2022 if (error && error != ENOMEM) 2023 return (error); 2024 2025 if (retval) { 2026 if (req.oldptr && req.oldidx > req.validlen) 2027 *retval = req.validlen; 2028 else 2029 *retval = req.oldidx; 2030 } 2031 return (error); 2032} 2033 2034/* 2035 * Drain into a sysctl struct. The user buffer should be wired if a page 2036 * fault would cause issue. 2037 */ 2038static int 2039sbuf_sysctl_drain(void *arg, const char *data, int len) 2040{ 2041 struct sysctl_req *req = arg; 2042 int error; 2043 2044 error = SYSCTL_OUT(req, data, len); 2045 KASSERT(error >= 0, ("Got unexpected negative value %d", error)); 2046 return (error == 0 ? len : -error); 2047} 2048 2049struct sbuf * 2050sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length, 2051 struct sysctl_req *req) 2052{ 2053 2054 /* Supply a default buffer size if none given. */ 2055 if (buf == NULL && length == 0) 2056 length = 64; 2057 s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL); 2058 sbuf_set_drain(s, sbuf_sysctl_drain, req); 2059 return (s); 2060} 2061