kern_sysctl.c revision 290475
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: head/sys/kern/kern_sysctl.c 290475 2015-11-07 01:43:01Z cem $"); 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 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, intptr_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 if (tracker != NULL) 170 SYSCTL_RLOCK(tracker); 171 else 172 SYSCTL_WLOCK(); 173 174 if (oid->oid_kind & CTLFLAG_DYN) { 175 if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 && 176 (oid->oid_kind & CTLFLAG_DYING) != 0) 177 wakeup(&oid->oid_running); 178 } 179 180 return (error); 181} 182 183static void 184sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp) 185{ 186 struct sysctl_req req; 187 struct sysctl_oid *curr; 188 char *penv = NULL; 189 char path[64]; 190 ssize_t rem = sizeof(path); 191 ssize_t len; 192 int val_int; 193 long val_long; 194 int64_t val_64; 195 quad_t val_quad; 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_int(path + rem, &val_int) == 0) 224 return; 225 req.newlen = sizeof(val_int); 226 req.newptr = &val_int; 227 break; 228 case CTLTYPE_UINT: 229 if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0) 230 return; 231 req.newlen = sizeof(val_int); 232 req.newptr = &val_int; 233 break; 234 case CTLTYPE_LONG: 235 if (getenv_long(path + rem, &val_long) == 0) 236 return; 237 req.newlen = sizeof(val_long); 238 req.newptr = &val_long; 239 break; 240 case CTLTYPE_ULONG: 241 if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0) 242 return; 243 req.newlen = sizeof(val_long); 244 req.newptr = &val_long; 245 break; 246 case CTLTYPE_S64: 247 if (getenv_quad(path + rem, &val_quad) == 0) 248 return; 249 val_64 = val_quad; 250 req.newlen = sizeof(val_64); 251 req.newptr = &val_64; 252 break; 253 case CTLTYPE_U64: 254 /* XXX there is no getenv_uquad() */ 255 if (getenv_quad(path + rem, &val_quad) == 0) 256 return; 257 val_64 = val_quad; 258 req.newlen = sizeof(val_64); 259 req.newptr = &val_64; 260 break; 261 case CTLTYPE_STRING: 262 penv = kern_getenv(path + rem); 263 if (penv == NULL) 264 return; 265 req.newlen = strlen(penv); 266 req.newptr = penv; 267 break; 268 default: 269 return; 270 } 271 error = sysctl_root_handler_locked(oidp, oidp->oid_arg1, 272 oidp->oid_arg2, &req, NULL); 273 if (error != 0) 274 printf("Setting sysctl %s failed: %d\n", path + rem, error); 275 if (penv != NULL) 276 freeenv(penv); 277} 278 279void 280sysctl_register_oid(struct sysctl_oid *oidp) 281{ 282 struct sysctl_oid_list *parent = oidp->oid_parent; 283 struct sysctl_oid *p; 284 struct sysctl_oid *q; 285 int oid_number; 286 int timeout = 2; 287 288 /* 289 * First check if another oid with the same name already 290 * exists in the parent's list. 291 */ 292 SYSCTL_ASSERT_WLOCKED(); 293 p = sysctl_find_oidname(oidp->oid_name, parent); 294 if (p != NULL) { 295 if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 296 p->oid_refcnt++; 297 return; 298 } else { 299 printf("can't re-use a leaf (%s)!\n", p->oid_name); 300 return; 301 } 302 } 303 /* get current OID number */ 304 oid_number = oidp->oid_number; 305 306#if (OID_AUTO >= 0) 307#error "OID_AUTO is expected to be a negative value" 308#endif 309 /* 310 * Any negative OID number qualifies as OID_AUTO. Valid OID 311 * numbers should always be positive. 312 * 313 * NOTE: DO NOT change the starting value here, change it in 314 * <sys/sysctl.h>, and make sure it is at least 256 to 315 * accomodate e.g. net.inet.raw as a static sysctl node. 316 */ 317 if (oid_number < 0) { 318 static int newoid; 319 320 /* 321 * By decrementing the next OID number we spend less 322 * time inserting the OIDs into a sorted list. 323 */ 324 if (--newoid < CTL_AUTO_START) 325 newoid = 0x7fffffff; 326 327 oid_number = newoid; 328 } 329 330 /* 331 * Insert the OID into the parent's list sorted by OID number. 332 */ 333retry: 334 q = NULL; 335 SLIST_FOREACH(p, parent, oid_link) { 336 /* check if the current OID number is in use */ 337 if (oid_number == p->oid_number) { 338 /* get the next valid OID number */ 339 if (oid_number < CTL_AUTO_START || 340 oid_number == 0x7fffffff) { 341 /* wraparound - restart */ 342 oid_number = CTL_AUTO_START; 343 /* don't loop forever */ 344 if (!timeout--) 345 panic("sysctl: Out of OID numbers\n"); 346 goto retry; 347 } else { 348 oid_number++; 349 } 350 } else if (oid_number < p->oid_number) 351 break; 352 q = p; 353 } 354 /* check for non-auto OID number collision */ 355 if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START && 356 oid_number >= CTL_AUTO_START) { 357 printf("sysctl: OID number(%d) is already in use for '%s'\n", 358 oidp->oid_number, oidp->oid_name); 359 } 360 /* update the OID number, if any */ 361 oidp->oid_number = oid_number; 362 if (q != NULL) 363 SLIST_INSERT_AFTER(q, oidp, oid_link); 364 else 365 SLIST_INSERT_HEAD(parent, oidp, oid_link); 366 367 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE && 368#ifdef VIMAGE 369 (oidp->oid_kind & CTLFLAG_VNET) == 0 && 370#endif 371 (oidp->oid_kind & CTLFLAG_TUN) != 0 && 372 (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) { 373 /* only fetch value once */ 374 oidp->oid_kind |= CTLFLAG_NOFETCH; 375 /* try to fetch value from kernel environment */ 376 sysctl_load_tunable_by_oid_locked(oidp); 377 } 378} 379 380void 381sysctl_unregister_oid(struct sysctl_oid *oidp) 382{ 383 struct sysctl_oid *p; 384 int error; 385 386 SYSCTL_ASSERT_WLOCKED(); 387 error = ENOENT; 388 if (oidp->oid_number == OID_AUTO) { 389 error = EINVAL; 390 } else { 391 SLIST_FOREACH(p, oidp->oid_parent, oid_link) { 392 if (p == oidp) { 393 SLIST_REMOVE(oidp->oid_parent, oidp, 394 sysctl_oid, oid_link); 395 error = 0; 396 break; 397 } 398 } 399 } 400 401 /* 402 * This can happen when a module fails to register and is 403 * being unloaded afterwards. It should not be a panic() 404 * for normal use. 405 */ 406 if (error) 407 printf("%s: failed to unregister sysctl\n", __func__); 408} 409 410/* Initialize a new context to keep track of dynamically added sysctls. */ 411int 412sysctl_ctx_init(struct sysctl_ctx_list *c) 413{ 414 415 if (c == NULL) { 416 return (EINVAL); 417 } 418 419 /* 420 * No locking here, the caller is responsible for not adding 421 * new nodes to a context until after this function has 422 * returned. 423 */ 424 TAILQ_INIT(c); 425 return (0); 426} 427 428/* Free the context, and destroy all dynamic oids registered in this context */ 429int 430sysctl_ctx_free(struct sysctl_ctx_list *clist) 431{ 432 struct sysctl_ctx_entry *e, *e1; 433 int error; 434 435 error = 0; 436 /* 437 * First perform a "dry run" to check if it's ok to remove oids. 438 * XXX FIXME 439 * XXX This algorithm is a hack. But I don't know any 440 * XXX better solution for now... 441 */ 442 SYSCTL_WLOCK(); 443 TAILQ_FOREACH(e, clist, link) { 444 error = sysctl_remove_oid_locked(e->entry, 0, 0); 445 if (error) 446 break; 447 } 448 /* 449 * Restore deregistered entries, either from the end, 450 * or from the place where error occured. 451 * e contains the entry that was not unregistered 452 */ 453 if (error) 454 e1 = TAILQ_PREV(e, sysctl_ctx_list, link); 455 else 456 e1 = TAILQ_LAST(clist, sysctl_ctx_list); 457 while (e1 != NULL) { 458 sysctl_register_oid(e1->entry); 459 e1 = TAILQ_PREV(e1, sysctl_ctx_list, link); 460 } 461 if (error) { 462 SYSCTL_WUNLOCK(); 463 return(EBUSY); 464 } 465 /* Now really delete the entries */ 466 e = TAILQ_FIRST(clist); 467 while (e != NULL) { 468 e1 = TAILQ_NEXT(e, link); 469 error = sysctl_remove_oid_locked(e->entry, 1, 0); 470 if (error) 471 panic("sysctl_remove_oid: corrupt tree, entry: %s", 472 e->entry->oid_name); 473 free(e, M_SYSCTLOID); 474 e = e1; 475 } 476 SYSCTL_WUNLOCK(); 477 return (error); 478} 479 480/* Add an entry to the context */ 481struct sysctl_ctx_entry * 482sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 483{ 484 struct sysctl_ctx_entry *e; 485 486 SYSCTL_ASSERT_WLOCKED(); 487 if (clist == NULL || oidp == NULL) 488 return(NULL); 489 e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK); 490 e->entry = oidp; 491 TAILQ_INSERT_HEAD(clist, e, link); 492 return (e); 493} 494 495/* Find an entry in the context */ 496struct sysctl_ctx_entry * 497sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 498{ 499 struct sysctl_ctx_entry *e; 500 501 SYSCTL_ASSERT_WLOCKED(); 502 if (clist == NULL || oidp == NULL) 503 return(NULL); 504 TAILQ_FOREACH(e, clist, link) { 505 if(e->entry == oidp) 506 return(e); 507 } 508 return (e); 509} 510 511/* 512 * Delete an entry from the context. 513 * NOTE: this function doesn't free oidp! You have to remove it 514 * with sysctl_remove_oid(). 515 */ 516int 517sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) 518{ 519 struct sysctl_ctx_entry *e; 520 521 if (clist == NULL || oidp == NULL) 522 return (EINVAL); 523 SYSCTL_WLOCK(); 524 e = sysctl_ctx_entry_find(clist, oidp); 525 if (e != NULL) { 526 TAILQ_REMOVE(clist, e, link); 527 SYSCTL_WUNLOCK(); 528 free(e, M_SYSCTLOID); 529 return (0); 530 } else { 531 SYSCTL_WUNLOCK(); 532 return (ENOENT); 533 } 534} 535 536/* 537 * Remove dynamically created sysctl trees. 538 * oidp - top of the tree to be removed 539 * del - if 0 - just deregister, otherwise free up entries as well 540 * recurse - if != 0 traverse the subtree to be deleted 541 */ 542int 543sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse) 544{ 545 int error; 546 547 SYSCTL_WLOCK(); 548 error = sysctl_remove_oid_locked(oidp, del, recurse); 549 SYSCTL_WUNLOCK(); 550 return (error); 551} 552 553int 554sysctl_remove_name(struct sysctl_oid *parent, const char *name, 555 int del, int recurse) 556{ 557 struct sysctl_oid *p, *tmp; 558 int error; 559 560 error = ENOENT; 561 SYSCTL_WLOCK(); 562 SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) { 563 if (strcmp(p->oid_name, name) == 0) { 564 error = sysctl_remove_oid_locked(p, del, recurse); 565 break; 566 } 567 } 568 SYSCTL_WUNLOCK(); 569 570 return (error); 571} 572 573 574static int 575sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse) 576{ 577 struct sysctl_oid *p, *tmp; 578 int error; 579 580 SYSCTL_ASSERT_WLOCKED(); 581 if (oidp == NULL) 582 return(EINVAL); 583 if ((oidp->oid_kind & CTLFLAG_DYN) == 0) { 584 printf("can't remove non-dynamic nodes!\n"); 585 return (EINVAL); 586 } 587 /* 588 * WARNING: normal method to do this should be through 589 * sysctl_ctx_free(). Use recursing as the last resort 590 * method to purge your sysctl tree of leftovers... 591 * However, if some other code still references these nodes, 592 * it will panic. 593 */ 594 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 595 if (oidp->oid_refcnt == 1) { 596 SLIST_FOREACH_SAFE(p, 597 SYSCTL_CHILDREN(oidp), oid_link, tmp) { 598 if (!recurse) { 599 printf("Warning: failed attempt to " 600 "remove oid %s with child %s\n", 601 oidp->oid_name, p->oid_name); 602 return (ENOTEMPTY); 603 } 604 error = sysctl_remove_oid_locked(p, del, 605 recurse); 606 if (error) 607 return (error); 608 } 609 } 610 } 611 if (oidp->oid_refcnt > 1 ) { 612 oidp->oid_refcnt--; 613 } else { 614 if (oidp->oid_refcnt == 0) { 615 printf("Warning: bad oid_refcnt=%u (%s)!\n", 616 oidp->oid_refcnt, oidp->oid_name); 617 return (EINVAL); 618 } 619 sysctl_unregister_oid(oidp); 620 if (del) { 621 /* 622 * Wait for all threads running the handler to drain. 623 * This preserves the previous behavior when the 624 * sysctl lock was held across a handler invocation, 625 * and is necessary for module unload correctness. 626 */ 627 while (oidp->oid_running > 0) { 628 oidp->oid_kind |= CTLFLAG_DYING; 629 SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0); 630 } 631 if (oidp->oid_descr) 632 free(__DECONST(char *, oidp->oid_descr), 633 M_SYSCTLOID); 634 free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID); 635 free(oidp, M_SYSCTLOID); 636 } 637 } 638 return (0); 639} 640/* 641 * Create new sysctls at run time. 642 * clist may point to a valid context initialized with sysctl_ctx_init(). 643 */ 644struct sysctl_oid * 645sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent, 646 int number, const char *name, int kind, void *arg1, intmax_t arg2, 647 int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr) 648{ 649 struct sysctl_oid *oidp; 650 651 /* You have to hook up somewhere.. */ 652 if (parent == NULL) 653 return(NULL); 654 /* Check if the node already exists, otherwise create it */ 655 SYSCTL_WLOCK(); 656 oidp = sysctl_find_oidname(name, parent); 657 if (oidp != NULL) { 658 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 659 oidp->oid_refcnt++; 660 /* Update the context */ 661 if (clist != NULL) 662 sysctl_ctx_entry_add(clist, oidp); 663 SYSCTL_WUNLOCK(); 664 return (oidp); 665 } else { 666 SYSCTL_WUNLOCK(); 667 printf("can't re-use a leaf (%s)!\n", name); 668 return (NULL); 669 } 670 } 671 oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO); 672 oidp->oid_parent = parent; 673 SLIST_INIT(&oidp->oid_children); 674 oidp->oid_number = number; 675 oidp->oid_refcnt = 1; 676 oidp->oid_name = strdup(name, M_SYSCTLOID); 677 oidp->oid_handler = handler; 678 oidp->oid_kind = CTLFLAG_DYN | kind; 679 oidp->oid_arg1 = arg1; 680 oidp->oid_arg2 = arg2; 681 oidp->oid_fmt = fmt; 682 if (descr != NULL) 683 oidp->oid_descr = strdup(descr, M_SYSCTLOID); 684 /* Update the context, if used */ 685 if (clist != NULL) 686 sysctl_ctx_entry_add(clist, oidp); 687 /* Register this oid */ 688 sysctl_register_oid(oidp); 689 SYSCTL_WUNLOCK(); 690 return (oidp); 691} 692 693/* 694 * Rename an existing oid. 695 */ 696void 697sysctl_rename_oid(struct sysctl_oid *oidp, const char *name) 698{ 699 char *newname; 700 char *oldname; 701 702 newname = strdup(name, M_SYSCTLOID); 703 SYSCTL_WLOCK(); 704 oldname = __DECONST(char *, oidp->oid_name); 705 oidp->oid_name = newname; 706 SYSCTL_WUNLOCK(); 707 free(oldname, M_SYSCTLOID); 708} 709 710/* 711 * Reparent an existing oid. 712 */ 713int 714sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent) 715{ 716 struct sysctl_oid *oidp; 717 718 SYSCTL_WLOCK(); 719 if (oid->oid_parent == parent) { 720 SYSCTL_WUNLOCK(); 721 return (0); 722 } 723 oidp = sysctl_find_oidname(oid->oid_name, parent); 724 if (oidp != NULL) { 725 SYSCTL_WUNLOCK(); 726 return (EEXIST); 727 } 728 sysctl_unregister_oid(oid); 729 oid->oid_parent = parent; 730 oid->oid_number = OID_AUTO; 731 sysctl_register_oid(oid); 732 SYSCTL_WUNLOCK(); 733 return (0); 734} 735 736/* 737 * Register the kernel's oids on startup. 738 */ 739SET_DECLARE(sysctl_set, struct sysctl_oid); 740 741static void 742sysctl_register_all(void *arg) 743{ 744 struct sysctl_oid **oidp; 745 746 sx_init(&sysctlmemlock, "sysctl mem"); 747 SYSCTL_INIT(); 748 SYSCTL_WLOCK(); 749 SET_FOREACH(oidp, sysctl_set) 750 sysctl_register_oid(*oidp); 751 SYSCTL_WUNLOCK(); 752} 753SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0); 754 755/* 756 * "Staff-functions" 757 * 758 * These functions implement a presently undocumented interface 759 * used by the sysctl program to walk the tree, and get the type 760 * so it can print the value. 761 * This interface is under work and consideration, and should probably 762 * be killed with a big axe by the first person who can find the time. 763 * (be aware though, that the proper interface isn't as obvious as it 764 * may seem, there are various conflicting requirements. 765 * 766 * {0,0} printf the entire MIB-tree. 767 * {0,1,...} return the name of the "..." OID. 768 * {0,2,...} return the next OID. 769 * {0,3} return the OID of the name in "new" 770 * {0,4,...} return the kind & format info for the "..." OID. 771 * {0,5,...} return the description the "..." OID. 772 */ 773 774#ifdef SYSCTL_DEBUG 775static void 776sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i) 777{ 778 int k; 779 struct sysctl_oid *oidp; 780 781 SYSCTL_ASSERT_LOCKED(); 782 SLIST_FOREACH(oidp, l, oid_link) { 783 784 for (k=0; k<i; k++) 785 printf(" "); 786 787 printf("%d %s ", oidp->oid_number, oidp->oid_name); 788 789 printf("%c%c", 790 oidp->oid_kind & CTLFLAG_RD ? 'R':' ', 791 oidp->oid_kind & CTLFLAG_WR ? 'W':' '); 792 793 if (oidp->oid_handler) 794 printf(" *Handler"); 795 796 switch (oidp->oid_kind & CTLTYPE) { 797 case CTLTYPE_NODE: 798 printf(" Node\n"); 799 if (!oidp->oid_handler) { 800 sysctl_sysctl_debug_dump_node( 801 SYSCTL_CHILDREN(oidp), i + 2); 802 } 803 break; 804 case CTLTYPE_INT: printf(" Int\n"); break; 805 case CTLTYPE_UINT: printf(" u_int\n"); break; 806 case CTLTYPE_LONG: printf(" Long\n"); break; 807 case CTLTYPE_ULONG: printf(" u_long\n"); break; 808 case CTLTYPE_STRING: printf(" String\n"); break; 809 case CTLTYPE_U64: printf(" uint64_t\n"); break; 810 case CTLTYPE_S64: printf(" int64_t\n"); break; 811 case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break; 812 default: printf("\n"); 813 } 814 815 } 816} 817 818static int 819sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS) 820{ 821 struct rm_priotracker tracker; 822 int error; 823 824 error = priv_check(req->td, PRIV_SYSCTL_DEBUG); 825 if (error) 826 return (error); 827 SYSCTL_RLOCK(&tracker); 828 sysctl_sysctl_debug_dump_node(&sysctl__children, 0); 829 SYSCTL_RUNLOCK(&tracker); 830 return (ENOENT); 831} 832 833SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE, 834 0, 0, sysctl_sysctl_debug, "-", ""); 835#endif 836 837static int 838sysctl_sysctl_name(SYSCTL_HANDLER_ARGS) 839{ 840 int *name = (int *) arg1; 841 u_int namelen = arg2; 842 int error = 0; 843 struct sysctl_oid *oid; 844 struct sysctl_oid_list *lsp = &sysctl__children, *lsp2; 845 struct rm_priotracker tracker; 846 char buf[10]; 847 848 SYSCTL_RLOCK(&tracker); 849 while (namelen) { 850 if (!lsp) { 851 snprintf(buf,sizeof(buf),"%d",*name); 852 if (req->oldidx) 853 error = SYSCTL_OUT(req, ".", 1); 854 if (!error) 855 error = SYSCTL_OUT(req, buf, strlen(buf)); 856 if (error) 857 goto out; 858 namelen--; 859 name++; 860 continue; 861 } 862 lsp2 = 0; 863 SLIST_FOREACH(oid, lsp, oid_link) { 864 if (oid->oid_number != *name) 865 continue; 866 867 if (req->oldidx) 868 error = SYSCTL_OUT(req, ".", 1); 869 if (!error) 870 error = SYSCTL_OUT(req, oid->oid_name, 871 strlen(oid->oid_name)); 872 if (error) 873 goto out; 874 875 namelen--; 876 name++; 877 878 if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE) 879 break; 880 881 if (oid->oid_handler) 882 break; 883 884 lsp2 = SYSCTL_CHILDREN(oid); 885 break; 886 } 887 lsp = lsp2; 888 } 889 error = SYSCTL_OUT(req, "", 1); 890 out: 891 SYSCTL_RUNLOCK(&tracker); 892 return (error); 893} 894 895/* 896 * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in 897 * capability mode. 898 */ 899static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, 900 sysctl_sysctl_name, ""); 901 902static int 903sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen, 904 int *next, int *len, int level, struct sysctl_oid **oidpp) 905{ 906 struct sysctl_oid *oidp; 907 908 SYSCTL_ASSERT_LOCKED(); 909 *len = level; 910 SLIST_FOREACH(oidp, lsp, oid_link) { 911 *next = oidp->oid_number; 912 *oidpp = oidp; 913 914 if (oidp->oid_kind & CTLFLAG_SKIP) 915 continue; 916 917 if (!namelen) { 918 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 919 return (0); 920 if (oidp->oid_handler) 921 /* We really should call the handler here...*/ 922 return (0); 923 lsp = SYSCTL_CHILDREN(oidp); 924 if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1, 925 len, level+1, oidpp)) 926 return (0); 927 goto emptynode; 928 } 929 930 if (oidp->oid_number < *name) 931 continue; 932 933 if (oidp->oid_number > *name) { 934 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 935 return (0); 936 if (oidp->oid_handler) 937 return (0); 938 lsp = SYSCTL_CHILDREN(oidp); 939 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, 940 next+1, len, level+1, oidpp)) 941 return (0); 942 goto next; 943 } 944 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 945 continue; 946 947 if (oidp->oid_handler) 948 continue; 949 950 lsp = SYSCTL_CHILDREN(oidp); 951 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, 952 len, level+1, oidpp)) 953 return (0); 954 next: 955 namelen = 1; 956 emptynode: 957 *len = level; 958 } 959 return (1); 960} 961 962static int 963sysctl_sysctl_next(SYSCTL_HANDLER_ARGS) 964{ 965 int *name = (int *) arg1; 966 u_int namelen = arg2; 967 int i, j, error; 968 struct sysctl_oid *oid; 969 struct sysctl_oid_list *lsp = &sysctl__children; 970 struct rm_priotracker tracker; 971 int newoid[CTL_MAXNAME]; 972 973 SYSCTL_RLOCK(&tracker); 974 i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid); 975 SYSCTL_RUNLOCK(&tracker); 976 if (i) 977 return (ENOENT); 978 error = SYSCTL_OUT(req, newoid, j * sizeof (int)); 979 return (error); 980} 981 982/* 983 * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in 984 * capability mode. 985 */ 986static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, 987 sysctl_sysctl_next, ""); 988 989static int 990name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp) 991{ 992 struct sysctl_oid *oidp; 993 struct sysctl_oid_list *lsp = &sysctl__children; 994 char *p; 995 996 SYSCTL_ASSERT_LOCKED(); 997 998 for (*len = 0; *len < CTL_MAXNAME;) { 999 p = strsep(&name, "."); 1000 1001 oidp = SLIST_FIRST(lsp); 1002 for (;; oidp = SLIST_NEXT(oidp, oid_link)) { 1003 if (oidp == NULL) 1004 return (ENOENT); 1005 if (strcmp(p, oidp->oid_name) == 0) 1006 break; 1007 } 1008 *oid++ = oidp->oid_number; 1009 (*len)++; 1010 1011 if (name == NULL || *name == '\0') { 1012 if (oidpp) 1013 *oidpp = oidp; 1014 return (0); 1015 } 1016 1017 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) 1018 break; 1019 1020 if (oidp->oid_handler) 1021 break; 1022 1023 lsp = SYSCTL_CHILDREN(oidp); 1024 } 1025 return (ENOENT); 1026} 1027 1028static int 1029sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS) 1030{ 1031 char *p; 1032 int error, oid[CTL_MAXNAME], len = 0; 1033 struct sysctl_oid *op = 0; 1034 struct rm_priotracker tracker; 1035 1036 if (!req->newlen) 1037 return (ENOENT); 1038 if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */ 1039 return (ENAMETOOLONG); 1040 1041 p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK); 1042 1043 error = SYSCTL_IN(req, p, req->newlen); 1044 if (error) { 1045 free(p, M_SYSCTL); 1046 return (error); 1047 } 1048 1049 p [req->newlen] = '\0'; 1050 1051 SYSCTL_RLOCK(&tracker); 1052 error = name2oid(p, oid, &len, &op); 1053 SYSCTL_RUNLOCK(&tracker); 1054 1055 free(p, M_SYSCTL); 1056 1057 if (error) 1058 return (error); 1059 1060 error = SYSCTL_OUT(req, oid, len * sizeof *oid); 1061 return (error); 1062} 1063 1064/* 1065 * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in 1066 * capability mode. 1067 */ 1068SYSCTL_PROC(_sysctl, 3, name2oid, 1069 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE 1070 | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", ""); 1071 1072static int 1073sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS) 1074{ 1075 struct sysctl_oid *oid; 1076 struct rm_priotracker tracker; 1077 int error; 1078 1079 SYSCTL_RLOCK(&tracker); 1080 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 1081 if (error) 1082 goto out; 1083 1084 if (oid->oid_fmt == NULL) { 1085 error = ENOENT; 1086 goto out; 1087 } 1088 error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind)); 1089 if (error) 1090 goto out; 1091 error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1); 1092 out: 1093 SYSCTL_RUNLOCK(&tracker); 1094 return (error); 1095} 1096 1097 1098static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, 1099 sysctl_sysctl_oidfmt, ""); 1100 1101static int 1102sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS) 1103{ 1104 struct sysctl_oid *oid; 1105 struct rm_priotracker tracker; 1106 int error; 1107 1108 SYSCTL_RLOCK(&tracker); 1109 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); 1110 if (error) 1111 goto out; 1112 1113 if (oid->oid_descr == NULL) { 1114 error = ENOENT; 1115 goto out; 1116 } 1117 error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1); 1118 out: 1119 SYSCTL_RUNLOCK(&tracker); 1120 return (error); 1121} 1122 1123static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, 1124 sysctl_sysctl_oiddescr, ""); 1125 1126/* 1127 * Default "handler" functions. 1128 */ 1129 1130/* 1131 * Handle an int8_t, signed or unsigned. 1132 * Two cases: 1133 * a variable: point arg1 at it. 1134 * a constant: pass it in arg2. 1135 */ 1136 1137int 1138sysctl_handle_8(SYSCTL_HANDLER_ARGS) 1139{ 1140 int8_t tmpout; 1141 int error = 0; 1142 1143 /* 1144 * Attempt to get a coherent snapshot by making a copy of the data. 1145 */ 1146 if (arg1) 1147 tmpout = *(int8_t *)arg1; 1148 else 1149 tmpout = arg2; 1150 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1151 1152 if (error || !req->newptr) 1153 return (error); 1154 1155 if (!arg1) 1156 error = EPERM; 1157 else 1158 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1159 return (error); 1160} 1161 1162/* 1163 * Handle an int16_t, signed or unsigned. 1164 * Two cases: 1165 * a variable: point arg1 at it. 1166 * a constant: pass it in arg2. 1167 */ 1168 1169int 1170sysctl_handle_16(SYSCTL_HANDLER_ARGS) 1171{ 1172 int16_t tmpout; 1173 int error = 0; 1174 1175 /* 1176 * Attempt to get a coherent snapshot by making a copy of the data. 1177 */ 1178 if (arg1) 1179 tmpout = *(int16_t *)arg1; 1180 else 1181 tmpout = arg2; 1182 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1183 1184 if (error || !req->newptr) 1185 return (error); 1186 1187 if (!arg1) 1188 error = EPERM; 1189 else 1190 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1191 return (error); 1192} 1193 1194/* 1195 * Handle an int32_t, signed or unsigned. 1196 * Two cases: 1197 * a variable: point arg1 at it. 1198 * a constant: pass it in arg2. 1199 */ 1200 1201int 1202sysctl_handle_32(SYSCTL_HANDLER_ARGS) 1203{ 1204 int32_t tmpout; 1205 int error = 0; 1206 1207 /* 1208 * Attempt to get a coherent snapshot by making a copy of the data. 1209 */ 1210 if (arg1) 1211 tmpout = *(int32_t *)arg1; 1212 else 1213 tmpout = arg2; 1214 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); 1215 1216 if (error || !req->newptr) 1217 return (error); 1218 1219 if (!arg1) 1220 error = EPERM; 1221 else 1222 error = SYSCTL_IN(req, arg1, sizeof(tmpout)); 1223 return (error); 1224} 1225 1226/* 1227 * Handle an int, signed or unsigned. 1228 * Two cases: 1229 * a variable: point arg1 at it. 1230 * a constant: pass it in arg2. 1231 */ 1232 1233int 1234sysctl_handle_int(SYSCTL_HANDLER_ARGS) 1235{ 1236 int tmpout, error = 0; 1237 1238 /* 1239 * Attempt to get a coherent snapshot by making a copy of the data. 1240 */ 1241 if (arg1) 1242 tmpout = *(int *)arg1; 1243 else 1244 tmpout = arg2; 1245 error = SYSCTL_OUT(req, &tmpout, sizeof(int)); 1246 1247 if (error || !req->newptr) 1248 return (error); 1249 1250 if (!arg1) 1251 error = EPERM; 1252 else 1253 error = SYSCTL_IN(req, arg1, sizeof(int)); 1254 return (error); 1255} 1256 1257/* 1258 * Based on on sysctl_handle_int() convert milliseconds into ticks. 1259 * Note: this is used by TCP. 1260 */ 1261 1262int 1263sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS) 1264{ 1265 int error, s, tt; 1266 1267 tt = *(int *)arg1; 1268 s = (int)((int64_t)tt * 1000 / hz); 1269 1270 error = sysctl_handle_int(oidp, &s, 0, req); 1271 if (error || !req->newptr) 1272 return (error); 1273 1274 tt = (int)((int64_t)s * hz / 1000); 1275 if (tt < 1) 1276 return (EINVAL); 1277 1278 *(int *)arg1 = tt; 1279 return (0); 1280} 1281 1282 1283/* 1284 * Handle a long, signed or unsigned. 1285 * Two cases: 1286 * a variable: point arg1 at it. 1287 * a constant: pass it in arg2. 1288 */ 1289 1290int 1291sysctl_handle_long(SYSCTL_HANDLER_ARGS) 1292{ 1293 int error = 0; 1294 long tmplong; 1295#ifdef SCTL_MASK32 1296 int tmpint; 1297#endif 1298 1299 /* 1300 * Attempt to get a coherent snapshot by making a copy of the data. 1301 */ 1302 if (arg1) 1303 tmplong = *(long *)arg1; 1304 else 1305 tmplong = arg2; 1306#ifdef SCTL_MASK32 1307 if (req->flags & SCTL_MASK32) { 1308 tmpint = tmplong; 1309 error = SYSCTL_OUT(req, &tmpint, sizeof(int)); 1310 } else 1311#endif 1312 error = SYSCTL_OUT(req, &tmplong, sizeof(long)); 1313 1314 if (error || !req->newptr) 1315 return (error); 1316 1317 if (!arg1) 1318 error = EPERM; 1319#ifdef SCTL_MASK32 1320 else if (req->flags & SCTL_MASK32) { 1321 error = SYSCTL_IN(req, &tmpint, sizeof(int)); 1322 *(long *)arg1 = (long)tmpint; 1323 } 1324#endif 1325 else 1326 error = SYSCTL_IN(req, arg1, sizeof(long)); 1327 return (error); 1328} 1329 1330/* 1331 * Handle a 64 bit int, signed or unsigned. 1332 * Two cases: 1333 * a variable: point arg1 at it. 1334 * a constant: pass it in arg2. 1335 */ 1336int 1337sysctl_handle_64(SYSCTL_HANDLER_ARGS) 1338{ 1339 int error = 0; 1340 uint64_t tmpout; 1341 1342 /* 1343 * Attempt to get a coherent snapshot by making a copy of the data. 1344 */ 1345 if (arg1) 1346 tmpout = *(uint64_t *)arg1; 1347 else 1348 tmpout = arg2; 1349 error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t)); 1350 1351 if (error || !req->newptr) 1352 return (error); 1353 1354 if (!arg1) 1355 error = EPERM; 1356 else 1357 error = SYSCTL_IN(req, arg1, sizeof(uint64_t)); 1358 return (error); 1359} 1360 1361/* 1362 * Handle our generic '\0' terminated 'C' string. 1363 * Two cases: 1364 * a variable string: point arg1 at it, arg2 is max length. 1365 * a constant string: point arg1 at it, arg2 is zero. 1366 */ 1367 1368int 1369sysctl_handle_string(SYSCTL_HANDLER_ARGS) 1370{ 1371 size_t outlen; 1372 int error = 0, ro_string = 0; 1373 1374 /* 1375 * A zero-length buffer indicates a fixed size read-only 1376 * string: 1377 */ 1378 if (arg2 == 0) { 1379 arg2 = strlen((char *)arg1) + 1; 1380 ro_string = 1; 1381 } 1382 1383 if (req->oldptr != NULL) { 1384 char *tmparg; 1385 1386 if (ro_string) { 1387 tmparg = arg1; 1388 } else { 1389 /* try to make a coherent snapshot of the string */ 1390 tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK); 1391 memcpy(tmparg, arg1, arg2); 1392 } 1393 1394 outlen = strnlen(tmparg, arg2 - 1) + 1; 1395 error = SYSCTL_OUT(req, tmparg, outlen); 1396 1397 if (!ro_string) 1398 free(tmparg, M_SYSCTLTMP); 1399 } else { 1400 outlen = strnlen((char *)arg1, arg2 - 1) + 1; 1401 error = SYSCTL_OUT(req, NULL, outlen); 1402 } 1403 if (error || !req->newptr) 1404 return (error); 1405 1406 if ((req->newlen - req->newidx) >= arg2) { 1407 error = EINVAL; 1408 } else { 1409 arg2 = (req->newlen - req->newidx); 1410 error = SYSCTL_IN(req, arg1, arg2); 1411 ((char *)arg1)[arg2] = '\0'; 1412 } 1413 return (error); 1414} 1415 1416/* 1417 * Handle any kind of opaque data. 1418 * arg1 points to it, arg2 is the size. 1419 */ 1420 1421int 1422sysctl_handle_opaque(SYSCTL_HANDLER_ARGS) 1423{ 1424 int error, tries; 1425 u_int generation; 1426 struct sysctl_req req2; 1427 1428 /* 1429 * Attempt to get a coherent snapshot, by using the thread 1430 * pre-emption counter updated from within mi_switch() to 1431 * determine if we were pre-empted during a bcopy() or 1432 * copyout(). Make 3 attempts at doing this before giving up. 1433 * If we encounter an error, stop immediately. 1434 */ 1435 tries = 0; 1436 req2 = *req; 1437retry: 1438 generation = curthread->td_generation; 1439 error = SYSCTL_OUT(req, arg1, arg2); 1440 if (error) 1441 return (error); 1442 tries++; 1443 if (generation != curthread->td_generation && tries < 3) { 1444 *req = req2; 1445 goto retry; 1446 } 1447 1448 error = SYSCTL_IN(req, arg1, arg2); 1449 1450 return (error); 1451} 1452 1453/* 1454 * Transfer functions to/from kernel space. 1455 * XXX: rather untested at this point 1456 */ 1457static int 1458sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l) 1459{ 1460 size_t i = 0; 1461 1462 if (req->oldptr) { 1463 i = l; 1464 if (req->oldlen <= req->oldidx) 1465 i = 0; 1466 else 1467 if (i > req->oldlen - req->oldidx) 1468 i = req->oldlen - req->oldidx; 1469 if (i > 0) 1470 bcopy(p, (char *)req->oldptr + req->oldidx, i); 1471 } 1472 req->oldidx += l; 1473 if (req->oldptr && i != l) 1474 return (ENOMEM); 1475 return (0); 1476} 1477 1478static int 1479sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l) 1480{ 1481 if (!req->newptr) 1482 return (0); 1483 if (req->newlen - req->newidx < l) 1484 return (EINVAL); 1485 bcopy((char *)req->newptr + req->newidx, p, l); 1486 req->newidx += l; 1487 return (0); 1488} 1489 1490int 1491kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old, 1492 size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags) 1493{ 1494 int error = 0; 1495 struct sysctl_req req; 1496 1497 bzero(&req, sizeof req); 1498 1499 req.td = td; 1500 req.flags = flags; 1501 1502 if (oldlenp) { 1503 req.oldlen = *oldlenp; 1504 } 1505 req.validlen = req.oldlen; 1506 1507 if (old) { 1508 req.oldptr= old; 1509 } 1510 1511 if (new != NULL) { 1512 req.newlen = newlen; 1513 req.newptr = new; 1514 } 1515 1516 req.oldfunc = sysctl_old_kernel; 1517 req.newfunc = sysctl_new_kernel; 1518 req.lock = REQ_UNWIRED; 1519 1520 error = sysctl_root(0, name, namelen, &req); 1521 1522 if (req.lock == REQ_WIRED && req.validlen > 0) 1523 vsunlock(req.oldptr, req.validlen); 1524 1525 if (error && error != ENOMEM) 1526 return (error); 1527 1528 if (retval) { 1529 if (req.oldptr && req.oldidx > req.validlen) 1530 *retval = req.validlen; 1531 else 1532 *retval = req.oldidx; 1533 } 1534 return (error); 1535} 1536 1537int 1538kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp, 1539 void *new, size_t newlen, size_t *retval, int flags) 1540{ 1541 int oid[CTL_MAXNAME]; 1542 size_t oidlen, plen; 1543 int error; 1544 1545 oid[0] = 0; /* sysctl internal magic */ 1546 oid[1] = 3; /* name2oid */ 1547 oidlen = sizeof(oid); 1548 1549 error = kernel_sysctl(td, oid, 2, oid, &oidlen, 1550 (void *)name, strlen(name), &plen, flags); 1551 if (error) 1552 return (error); 1553 1554 error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp, 1555 new, newlen, retval, flags); 1556 return (error); 1557} 1558 1559/* 1560 * Transfer function to/from user space. 1561 */ 1562static int 1563sysctl_old_user(struct sysctl_req *req, const void *p, size_t l) 1564{ 1565 size_t i, len, origidx; 1566 int error; 1567 1568 origidx = req->oldidx; 1569 req->oldidx += l; 1570 if (req->oldptr == NULL) 1571 return (0); 1572 /* 1573 * If we have not wired the user supplied buffer and we are currently 1574 * holding locks, drop a witness warning, as it's possible that 1575 * write operations to the user page can sleep. 1576 */ 1577 if (req->lock != REQ_WIRED) 1578 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 1579 "sysctl_old_user()"); 1580 i = l; 1581 len = req->validlen; 1582 if (len <= origidx) 1583 i = 0; 1584 else { 1585 if (i > len - origidx) 1586 i = len - origidx; 1587 if (req->lock == REQ_WIRED) { 1588 error = copyout_nofault(p, (char *)req->oldptr + 1589 origidx, i); 1590 } else 1591 error = copyout(p, (char *)req->oldptr + origidx, i); 1592 if (error != 0) 1593 return (error); 1594 } 1595 if (i < l) 1596 return (ENOMEM); 1597 return (0); 1598} 1599 1600static int 1601sysctl_new_user(struct sysctl_req *req, void *p, size_t l) 1602{ 1603 int error; 1604 1605 if (!req->newptr) 1606 return (0); 1607 if (req->newlen - req->newidx < l) 1608 return (EINVAL); 1609 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 1610 "sysctl_new_user()"); 1611 error = copyin((char *)req->newptr + req->newidx, p, l); 1612 req->newidx += l; 1613 return (error); 1614} 1615 1616/* 1617 * Wire the user space destination buffer. If set to a value greater than 1618 * zero, the len parameter limits the maximum amount of wired memory. 1619 */ 1620int 1621sysctl_wire_old_buffer(struct sysctl_req *req, size_t len) 1622{ 1623 int ret; 1624 size_t wiredlen; 1625 1626 wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen; 1627 ret = 0; 1628 if (req->lock != REQ_WIRED && req->oldptr && 1629 req->oldfunc == sysctl_old_user) { 1630 if (wiredlen != 0) { 1631 ret = vslock(req->oldptr, wiredlen); 1632 if (ret != 0) { 1633 if (ret != ENOMEM) 1634 return (ret); 1635 wiredlen = 0; 1636 } 1637 } 1638 req->lock = REQ_WIRED; 1639 req->validlen = wiredlen; 1640 } 1641 return (0); 1642} 1643 1644int 1645sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid, 1646 int *nindx, struct sysctl_req *req) 1647{ 1648 struct sysctl_oid_list *lsp; 1649 struct sysctl_oid *oid; 1650 int indx; 1651 1652 SYSCTL_ASSERT_LOCKED(); 1653 lsp = &sysctl__children; 1654 indx = 0; 1655 while (indx < CTL_MAXNAME) { 1656 SLIST_FOREACH(oid, lsp, oid_link) { 1657 if (oid->oid_number == name[indx]) 1658 break; 1659 } 1660 if (oid == NULL) 1661 return (ENOENT); 1662 1663 indx++; 1664 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1665 if (oid->oid_handler != NULL || indx == namelen) { 1666 *noid = oid; 1667 if (nindx != NULL) 1668 *nindx = indx; 1669 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1670 ("%s found DYING node %p", __func__, oid)); 1671 return (0); 1672 } 1673 lsp = SYSCTL_CHILDREN(oid); 1674 } else if (indx == namelen) { 1675 *noid = oid; 1676 if (nindx != NULL) 1677 *nindx = indx; 1678 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, 1679 ("%s found DYING node %p", __func__, oid)); 1680 return (0); 1681 } else { 1682 return (ENOTDIR); 1683 } 1684 } 1685 return (ENOENT); 1686} 1687 1688/* 1689 * Traverse our tree, and find the right node, execute whatever it points 1690 * to, and return the resulting error code. 1691 */ 1692 1693static int 1694sysctl_root(SYSCTL_HANDLER_ARGS) 1695{ 1696 struct sysctl_oid *oid; 1697 struct rm_priotracker tracker; 1698 int error, indx, lvl; 1699 1700 SYSCTL_RLOCK(&tracker); 1701 1702 error = sysctl_find_oid(arg1, arg2, &oid, &indx, req); 1703 if (error) 1704 goto out; 1705 1706 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1707 /* 1708 * You can't call a sysctl when it's a node, but has 1709 * no handler. Inform the user that it's a node. 1710 * The indx may or may not be the same as namelen. 1711 */ 1712 if (oid->oid_handler == NULL) { 1713 error = EISDIR; 1714 goto out; 1715 } 1716 } 1717 1718 /* Is this sysctl writable? */ 1719 if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) { 1720 error = EPERM; 1721 goto out; 1722 } 1723 1724 KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL")); 1725 1726#ifdef CAPABILITY_MODE 1727 /* 1728 * If the process is in capability mode, then don't permit reading or 1729 * writing unless specifically granted for the node. 1730 */ 1731 if (IN_CAPABILITY_MODE(req->td)) { 1732 if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) || 1733 (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) { 1734 error = EPERM; 1735 goto out; 1736 } 1737 } 1738#endif 1739 1740 /* Is this sysctl sensitive to securelevels? */ 1741 if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) { 1742 lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE; 1743 error = securelevel_gt(req->td->td_ucred, lvl); 1744 if (error) 1745 goto out; 1746 } 1747 1748 /* Is this sysctl writable by only privileged users? */ 1749 if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) { 1750 int priv; 1751 1752 if (oid->oid_kind & CTLFLAG_PRISON) 1753 priv = PRIV_SYSCTL_WRITEJAIL; 1754#ifdef VIMAGE 1755 else if ((oid->oid_kind & CTLFLAG_VNET) && 1756 prison_owns_vnet(req->td->td_ucred)) 1757 priv = PRIV_SYSCTL_WRITEJAIL; 1758#endif 1759 else 1760 priv = PRIV_SYSCTL_WRITE; 1761 error = priv_check(req->td, priv); 1762 if (error) 1763 goto out; 1764 } 1765 1766 if (!oid->oid_handler) { 1767 error = EINVAL; 1768 goto out; 1769 } 1770 1771 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { 1772 arg1 = (int *)arg1 + indx; 1773 arg2 -= indx; 1774 } else { 1775 arg1 = oid->oid_arg1; 1776 arg2 = oid->oid_arg2; 1777 } 1778#ifdef MAC 1779 error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2, 1780 req); 1781 if (error != 0) 1782 goto out; 1783#endif 1784#ifdef VIMAGE 1785 if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL) 1786 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1); 1787#endif 1788 error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker); 1789 1790 KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error); 1791 1792out: 1793 SYSCTL_RUNLOCK(&tracker); 1794 return (error); 1795} 1796 1797#ifndef _SYS_SYSPROTO_H_ 1798struct sysctl_args { 1799 int *name; 1800 u_int namelen; 1801 void *old; 1802 size_t *oldlenp; 1803 void *new; 1804 size_t newlen; 1805}; 1806#endif 1807int 1808sys___sysctl(struct thread *td, struct sysctl_args *uap) 1809{ 1810 int error, i, name[CTL_MAXNAME]; 1811 size_t j; 1812 1813 if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) 1814 return (EINVAL); 1815 1816 error = copyin(uap->name, &name, uap->namelen * sizeof(int)); 1817 if (error) 1818 return (error); 1819 1820 error = userland_sysctl(td, name, uap->namelen, 1821 uap->old, uap->oldlenp, 0, 1822 uap->new, uap->newlen, &j, 0); 1823 if (error && error != ENOMEM) 1824 return (error); 1825 if (uap->oldlenp) { 1826 i = copyout(&j, uap->oldlenp, sizeof(j)); 1827 if (i) 1828 return (i); 1829 } 1830 return (error); 1831} 1832 1833/* 1834 * This is used from various compatibility syscalls too. That's why name 1835 * must be in kernel space. 1836 */ 1837int 1838userland_sysctl(struct thread *td, int *name, u_int namelen, void *old, 1839 size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval, 1840 int flags) 1841{ 1842 int error = 0, memlocked; 1843 struct sysctl_req req; 1844 1845 bzero(&req, sizeof req); 1846 1847 req.td = td; 1848 req.flags = flags; 1849 1850 if (oldlenp) { 1851 if (inkernel) { 1852 req.oldlen = *oldlenp; 1853 } else { 1854 error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp)); 1855 if (error) 1856 return (error); 1857 } 1858 } 1859 req.validlen = req.oldlen; 1860 1861 if (old) { 1862 if (!useracc(old, req.oldlen, VM_PROT_WRITE)) 1863 return (EFAULT); 1864 req.oldptr= old; 1865 } 1866 1867 if (new != NULL) { 1868 if (!useracc(new, newlen, VM_PROT_READ)) 1869 return (EFAULT); 1870 req.newlen = newlen; 1871 req.newptr = new; 1872 } 1873 1874 req.oldfunc = sysctl_old_user; 1875 req.newfunc = sysctl_new_user; 1876 req.lock = REQ_UNWIRED; 1877 1878#ifdef KTRACE 1879 if (KTRPOINT(curthread, KTR_SYSCTL)) 1880 ktrsysctl(name, namelen); 1881#endif 1882 1883 if (req.oldptr && req.oldlen > PAGE_SIZE) { 1884 memlocked = 1; 1885 sx_xlock(&sysctlmemlock); 1886 } else 1887 memlocked = 0; 1888 CURVNET_SET(TD_TO_VNET(td)); 1889 1890 for (;;) { 1891 req.oldidx = 0; 1892 req.newidx = 0; 1893 error = sysctl_root(0, name, namelen, &req); 1894 if (error != EAGAIN) 1895 break; 1896 kern_yield(PRI_USER); 1897 } 1898 1899 CURVNET_RESTORE(); 1900 1901 if (req.lock == REQ_WIRED && req.validlen > 0) 1902 vsunlock(req.oldptr, req.validlen); 1903 if (memlocked) 1904 sx_xunlock(&sysctlmemlock); 1905 1906 if (error && error != ENOMEM) 1907 return (error); 1908 1909 if (retval) { 1910 if (req.oldptr && req.oldidx > req.validlen) 1911 *retval = req.validlen; 1912 else 1913 *retval = req.oldidx; 1914 } 1915 return (error); 1916} 1917 1918/* 1919 * Drain into a sysctl struct. The user buffer should be wired if a page 1920 * fault would cause issue. 1921 */ 1922static int 1923sbuf_sysctl_drain(void *arg, const char *data, int len) 1924{ 1925 struct sysctl_req *req = arg; 1926 int error; 1927 1928 error = SYSCTL_OUT(req, data, len); 1929 KASSERT(error >= 0, ("Got unexpected negative value %d", error)); 1930 return (error == 0 ? len : -error); 1931} 1932 1933struct sbuf * 1934sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length, 1935 struct sysctl_req *req) 1936{ 1937 1938 /* Supply a default buffer size if none given. */ 1939 if (buf == NULL && length == 0) 1940 length = 64; 1941 s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL); 1942 sbuf_set_drain(s, sbuf_sysctl_drain, req); 1943 return (s); 1944} 1945