sys/kern/kern_sysctl.c

   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("$FreeBSD$");
  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
  73 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
  74 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
  75 static 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  */
  90 static struct rmlock sysctllock;
  91 static 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
 106 static int sysctl_root(SYSCTL_HANDLER_ARGS);
 107
 108 /* Root list */
 109 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
 110
 111 static int      sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
 112                     int recurse);
 113 static int      sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
 114 static int      sysctl_new_kernel(struct sysctl_req *, void *, size_t);
 115
 116 static struct sysctl_oid *
 117 sysctl_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  */
 135 void
 136 sysctl_wlock(void)
 137 {
 138
 139         SYSCTL_WLOCK();
 140 }
 141
 142 void
 143 sysctl_wunlock(void)
 144 {
 145
 146         SYSCTL_WUNLOCK();
 147 }
 148
 149 static int
 150 sysctl_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
 185 static void
 186 sysctl_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
 324 void
 325 sysctl_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          */
 378 retry:
 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
 425 void
 426 sysctl_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
 441 void
 442 sysctl_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
 456 void
 457 sysctl_unregister_oid(struct sysctl_oid *oidp)
 458 {
 459         struct sysctl_oid *p;
 460         int error;
 461
 462         SYSCTL_ASSERT_WLOCKED();
 463         if (oidp->oid_number == OID_AUTO) {
 464                 error = EINVAL;
 465         } else {
 466                 error = ENOENT;
 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(%d) to unregister sysctl(%s)\n",
 484                     __func__, error, oidp->oid_name);
 485         }
 486 }
 487
 488 /* Initialize a new context to keep track of dynamically added sysctls. */
 489 int
 490 sysctl_ctx_init(struct sysctl_ctx_list *c)
 491 {
 492
 493         if (c == NULL) {
 494                 return (EINVAL);
 495         }
 496
 497         /*
 498          * No locking here, the caller is responsible for not adding
 499          * new nodes to a context until after this function has
 500          * returned.
 501          */
 502         TAILQ_INIT(c);
 503         return (0);
 504 }
 505
 506 /* Free the context, and destroy all dynamic oids registered in this context */
 507 int
 508 sysctl_ctx_free(struct sysctl_ctx_list *clist)
 509 {
 510         struct sysctl_ctx_entry *e, *e1;
 511         int error;
 512
 513         error = 0;
 514         /*
 515          * First perform a "dry run" to check if it's ok to remove oids.
 516          * XXX FIXME
 517          * XXX This algorithm is a hack. But I don't know any
 518          * XXX better solution for now...
 519          */
 520         SYSCTL_WLOCK();
 521         TAILQ_FOREACH(e, clist, link) {
 522                 error = sysctl_remove_oid_locked(e->entry, 0, 0);
 523                 if (error)
 524                         break;
 525         }
 526         /*
 527          * Restore deregistered entries, either from the end,
 528          * or from the place where error occurred.
 529          * e contains the entry that was not unregistered
 530          */
 531         if (error)
 532                 e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
 533         else
 534                 e1 = TAILQ_LAST(clist, sysctl_ctx_list);
 535         while (e1 != NULL) {
 536                 sysctl_register_oid(e1->entry);
 537                 e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
 538         }
 539         if (error) {
 540                 SYSCTL_WUNLOCK();
 541                 return(EBUSY);
 542         }
 543         /* Now really delete the entries */
 544         e = TAILQ_FIRST(clist);
 545         while (e != NULL) {
 546                 e1 = TAILQ_NEXT(e, link);
 547                 error = sysctl_remove_oid_locked(e->entry, 1, 0);
 548                 if (error)
 549                         panic("sysctl_remove_oid: corrupt tree, entry: %s",
 550                             e->entry->oid_name);
 551                 free(e, M_SYSCTLOID);
 552                 e = e1;
 553         }
 554         SYSCTL_WUNLOCK();
 555         return (error);
 556 }
 557
 558 /* Add an entry to the context */
 559 struct sysctl_ctx_entry *
 560 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
 561 {
 562         struct sysctl_ctx_entry *e;
 563
 564         SYSCTL_ASSERT_WLOCKED();
 565         if (clist == NULL || oidp == NULL)
 566                 return(NULL);
 567         e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
 568         e->entry = oidp;
 569         TAILQ_INSERT_HEAD(clist, e, link);
 570         return (e);
 571 }
 572
 573 /* Find an entry in the context */
 574 struct sysctl_ctx_entry *
 575 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
 576 {
 577         struct sysctl_ctx_entry *e;
 578
 579         SYSCTL_ASSERT_WLOCKED();
 580         if (clist == NULL || oidp == NULL)
 581                 return(NULL);
 582         TAILQ_FOREACH(e, clist, link) {
 583                 if(e->entry == oidp)
 584                         return(e);
 585         }
 586         return (e);
 587 }
 588
 589 /*
 590  * Delete an entry from the context.
 591  * NOTE: this function doesn't free oidp! You have to remove it
 592  * with sysctl_remove_oid().
 593  */
 594 int
 595 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
 596 {
 597         struct sysctl_ctx_entry *e;
 598
 599         if (clist == NULL || oidp == NULL)
 600                 return (EINVAL);
 601         SYSCTL_WLOCK();
 602         e = sysctl_ctx_entry_find(clist, oidp);
 603         if (e != NULL) {
 604                 TAILQ_REMOVE(clist, e, link);
 605                 SYSCTL_WUNLOCK();
 606                 free(e, M_SYSCTLOID);
 607                 return (0);
 608         } else {
 609                 SYSCTL_WUNLOCK();
 610                 return (ENOENT);
 611         }
 612 }
 613
 614 /*
 615  * Remove dynamically created sysctl trees.
 616  * oidp - top of the tree to be removed
 617  * del - if 0 - just deregister, otherwise free up entries as well
 618  * recurse - if != 0 traverse the subtree to be deleted
 619  */
 620 int
 621 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
 622 {
 623         int error;
 624
 625         SYSCTL_WLOCK();
 626         error = sysctl_remove_oid_locked(oidp, del, recurse);
 627         SYSCTL_WUNLOCK();
 628         return (error);
 629 }
 630
 631 int
 632 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
 633     int del, int recurse)
 634 {
 635         struct sysctl_oid *p, *tmp;
 636         int error;
 637
 638         error = ENOENT;
 639         SYSCTL_WLOCK();
 640         SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
 641                 if (strcmp(p->oid_name, name) == 0) {
 642                         error = sysctl_remove_oid_locked(p, del, recurse);
 643                         break;
 644                 }
 645         }
 646         SYSCTL_WUNLOCK();
 647
 648         return (error);
 649 }
 650
 651
 652 static int
 653 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
 654 {
 655         struct sysctl_oid *p, *tmp;
 656         int error;
 657
 658         SYSCTL_ASSERT_WLOCKED();
 659         if (oidp == NULL)
 660                 return(EINVAL);
 661         if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
 662                 printf("Warning: can't remove non-dynamic nodes (%s)!\n",
 663                     oidp->oid_name);
 664                 return (EINVAL);
 665         }
 666         /*
 667          * WARNING: normal method to do this should be through
 668          * sysctl_ctx_free(). Use recursing as the last resort
 669          * method to purge your sysctl tree of leftovers...
 670          * However, if some other code still references these nodes,
 671          * it will panic.
 672          */
 673         if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
 674                 if (oidp->oid_refcnt == 1) {
 675                         SLIST_FOREACH_SAFE(p,
 676                             SYSCTL_CHILDREN(oidp), oid_link, tmp) {
 677                                 if (!recurse) {
 678                                         printf("Warning: failed attempt to "
 679                                             "remove oid %s with child %s\n",
 680                                             oidp->oid_name, p->oid_name);
 681                                         return (ENOTEMPTY);
 682                                 }
 683                                 error = sysctl_remove_oid_locked(p, del,
 684                                     recurse);
 685                                 if (error)
 686                                         return (error);
 687                         }
 688                 }
 689         }
 690         if (oidp->oid_refcnt > 1 ) {
 691                 oidp->oid_refcnt--;
 692         } else {
 693                 if (oidp->oid_refcnt == 0) {
 694                         printf("Warning: bad oid_refcnt=%u (%s)!\n",
 695                                 oidp->oid_refcnt, oidp->oid_name);
 696                         return (EINVAL);
 697                 }
 698                 sysctl_unregister_oid(oidp);
 699                 if (del) {
 700                         /*
 701                          * Wait for all threads running the handler to drain.
 702                          * This preserves the previous behavior when the
 703                          * sysctl lock was held across a handler invocation,
 704                          * and is necessary for module unload correctness.
 705                          */
 706                         while (oidp->oid_running > 0) {
 707                                 oidp->oid_kind |= CTLFLAG_DYING;
 708                                 SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
 709                         }
 710                         if (oidp->oid_descr)
 711                                 free(__DECONST(char *, oidp->oid_descr),
 712                                     M_SYSCTLOID);
 713                         free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
 714                         free(oidp, M_SYSCTLOID);
 715                 }
 716         }
 717         return (0);
 718 }
 719 /*
 720  * Create new sysctls at run time.
 721  * clist may point to a valid context initialized with sysctl_ctx_init().
 722  */
 723 struct sysctl_oid *
 724 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
 725         int number, const char *name, int kind, void *arg1, intmax_t arg2,
 726         int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
 727 {
 728         struct sysctl_oid *oidp;
 729
 730         /* You have to hook up somewhere.. */
 731         if (parent == NULL)
 732                 return(NULL);
 733         /* Check if the node already exists, otherwise create it */
 734         SYSCTL_WLOCK();
 735         oidp = sysctl_find_oidname(name, parent);
 736         if (oidp != NULL) {
 737                 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
 738                         oidp->oid_refcnt++;
 739                         /* Update the context */
 740                         if (clist != NULL)
 741                                 sysctl_ctx_entry_add(clist, oidp);
 742                         SYSCTL_WUNLOCK();
 743                         return (oidp);
 744                 } else {
 745                         SYSCTL_WUNLOCK();
 746                         printf("can't re-use a leaf (%s)!\n", name);
 747                         return (NULL);
 748                 }
 749         }
 750         oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
 751         oidp->oid_parent = parent;
 752         SLIST_INIT(&oidp->oid_children);
 753         oidp->oid_number = number;
 754         oidp->oid_refcnt = 1;
 755         oidp->oid_name = strdup(name, M_SYSCTLOID);
 756         oidp->oid_handler = handler;
 757         oidp->oid_kind = CTLFLAG_DYN | kind;
 758         oidp->oid_arg1 = arg1;
 759         oidp->oid_arg2 = arg2;
 760         oidp->oid_fmt = fmt;
 761         if (descr != NULL)
 762                 oidp->oid_descr = strdup(descr, M_SYSCTLOID);
 763         /* Update the context, if used */
 764         if (clist != NULL)
 765                 sysctl_ctx_entry_add(clist, oidp);
 766         /* Register this oid */
 767         sysctl_register_oid(oidp);
 768         SYSCTL_WUNLOCK();
 769         return (oidp);
 770 }
 771
 772 /*
 773  * Rename an existing oid.
 774  */
 775 void
 776 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
 777 {
 778         char *newname;
 779         char *oldname;
 780
 781         newname = strdup(name, M_SYSCTLOID);
 782         SYSCTL_WLOCK();
 783         oldname = __DECONST(char *, oidp->oid_name);
 784         oidp->oid_name = newname;
 785         SYSCTL_WUNLOCK();
 786         free(oldname, M_SYSCTLOID);
 787 }
 788
 789 /*
 790  * Reparent an existing oid.
 791  */
 792 int
 793 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
 794 {
 795         struct sysctl_oid *oidp;
 796
 797         SYSCTL_WLOCK();
 798         if (oid->oid_parent == parent) {
 799                 SYSCTL_WUNLOCK();
 800                 return (0);
 801         }
 802         oidp = sysctl_find_oidname(oid->oid_name, parent);
 803         if (oidp != NULL) {
 804                 SYSCTL_WUNLOCK();
 805                 return (EEXIST);
 806         }
 807         sysctl_unregister_oid(oid);
 808         oid->oid_parent = parent;
 809         oid->oid_number = OID_AUTO;
 810         sysctl_register_oid(oid);
 811         SYSCTL_WUNLOCK();
 812         return (0);
 813 }
 814
 815 /*
 816  * Register the kernel's oids on startup.
 817  */
 818 SET_DECLARE(sysctl_set, struct sysctl_oid);
 819
 820 static void
 821 sysctl_register_all(void *arg)
 822 {
 823         struct sysctl_oid **oidp;
 824
 825         sx_init(&sysctlmemlock, "sysctl mem");
 826         SYSCTL_INIT();
 827         SYSCTL_WLOCK();
 828         SET_FOREACH(oidp, sysctl_set)
 829                 sysctl_register_oid(*oidp);
 830         SYSCTL_WUNLOCK();
 831 }
 832 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
 833
 834 /*
 835  * "Staff-functions"
 836  *
 837  * These functions implement a presently undocumented interface
 838  * used by the sysctl program to walk the tree, and get the type
 839  * so it can print the value.
 840  * This interface is under work and consideration, and should probably
 841  * be killed with a big axe by the first person who can find the time.
 842  * (be aware though, that the proper interface isn't as obvious as it
 843  * may seem, there are various conflicting requirements.
 844  *
 845  * {0,0}        printf the entire MIB-tree.
 846  * {0,1,...}    return the name of the "..." OID.
 847  * {0,2,...}    return the next OID.
 848  * {0,3}        return the OID of the name in "new"
 849  * {0,4,...}    return the kind & format info for the "..." OID.
 850  * {0,5,...}    return the description the "..." OID.
 851  */
 852
 853 #ifdef SYSCTL_DEBUG
 854 static void
 855 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
 856 {
 857         int k;
 858         struct sysctl_oid *oidp;
 859
 860         SYSCTL_ASSERT_LOCKED();
 861         SLIST_FOREACH(oidp, l, oid_link) {
 862
 863                 for (k=0; k<i; k++)
 864                         printf(" ");
 865
 866                 printf("%d %s ", oidp->oid_number, oidp->oid_name);
 867
 868                 printf("%c%c",
 869                         oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
 870                         oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
 871
 872                 if (oidp->oid_handler)
 873                         printf(" *Handler");
 874
 875                 switch (oidp->oid_kind & CTLTYPE) {
 876                         case CTLTYPE_NODE:
 877                                 printf(" Node\n");
 878                                 if (!oidp->oid_handler) {
 879                                         sysctl_sysctl_debug_dump_node(
 880                                             SYSCTL_CHILDREN(oidp), i + 2);
 881                                 }
 882                                 break;
 883                         case CTLTYPE_INT:    printf(" Int\n"); break;
 884                         case CTLTYPE_UINT:   printf(" u_int\n"); break;
 885                         case CTLTYPE_LONG:   printf(" Long\n"); break;
 886                         case CTLTYPE_ULONG:  printf(" u_long\n"); break;
 887                         case CTLTYPE_STRING: printf(" String\n"); break;
 888                         case CTLTYPE_S8:     printf(" int8_t\n"); break;
 889                         case CTLTYPE_S16:    printf(" int16_t\n"); break;
 890                         case CTLTYPE_S32:    printf(" int32_t\n"); break;
 891                         case CTLTYPE_S64:    printf(" int64_t\n"); break;
 892                         case CTLTYPE_U8:     printf(" uint8_t\n"); break;
 893                         case CTLTYPE_U16:    printf(" uint16_t\n"); break;
 894                         case CTLTYPE_U32:    printf(" uint32_t\n"); break;
 895                         case CTLTYPE_U64:    printf(" uint64_t\n"); break;
 896                         case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
 897                         default:             printf("\n");
 898                 }
 899
 900         }
 901 }
 902
 903 static int
 904 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
 905 {
 906         struct rm_priotracker tracker;
 907         int error;
 908
 909         error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
 910         if (error)
 911                 return (error);
 912         SYSCTL_RLOCK(&tracker);
 913         sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
 914         SYSCTL_RUNLOCK(&tracker);
 915         return (ENOENT);
 916 }
 917
 918 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
 919         0, 0, sysctl_sysctl_debug, "-", "");
 920 #endif
 921
 922 static int
 923 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
 924 {
 925         int *name = (int *) arg1;
 926         u_int namelen = arg2;
 927         int error = 0;
 928         struct sysctl_oid *oid;
 929         struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
 930         struct rm_priotracker tracker;
 931         char buf[10];
 932
 933         SYSCTL_RLOCK(&tracker);
 934         while (namelen) {
 935                 if (!lsp) {
 936                         snprintf(buf,sizeof(buf),"%d",*name);
 937                         if (req->oldidx)
 938                                 error = SYSCTL_OUT(req, ".", 1);
 939                         if (!error)
 940                                 error = SYSCTL_OUT(req, buf, strlen(buf));
 941                         if (error)
 942                                 goto out;
 943                         namelen--;
 944                         name++;
 945                         continue;
 946                 }
 947                 lsp2 = NULL;
 948                 SLIST_FOREACH(oid, lsp, oid_link) {
 949                         if (oid->oid_number != *name)
 950                                 continue;
 951
 952                         if (req->oldidx)
 953                                 error = SYSCTL_OUT(req, ".", 1);
 954                         if (!error)
 955                                 error = SYSCTL_OUT(req, oid->oid_name,
 956                                         strlen(oid->oid_name));
 957                         if (error)
 958                                 goto out;
 959
 960                         namelen--;
 961                         name++;
 962
 963                         if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
 964                                 break;
 965
 966                         if (oid->oid_handler)
 967                                 break;
 968
 969                         lsp2 = SYSCTL_CHILDREN(oid);
 970                         break;
 971                 }
 972                 lsp = lsp2;
 973         }
 974         error = SYSCTL_OUT(req, "", 1);
 975  out:
 976         SYSCTL_RUNLOCK(&tracker);
 977         return (error);
 978 }
 979
 980 /*
 981  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
 982  * capability mode.
 983  */
 984 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
 985     sysctl_sysctl_name, "");
 986
 987 static int
 988 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
 989         int *next, int *len, int level, struct sysctl_oid **oidpp)
 990 {
 991         struct sysctl_oid *oidp;
 992
 993         SYSCTL_ASSERT_LOCKED();
 994         *len = level;
 995         SLIST_FOREACH(oidp, lsp, oid_link) {
 996                 *next = oidp->oid_number;
 997                 *oidpp = oidp;
 998
 999                 if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0)
1000                         continue;
1001
1002                 if (!namelen) {
1003                         if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1004                                 return (0);
1005                         if (oidp->oid_handler)
1006                                 /* We really should call the handler here...*/
1007                                 return (0);
1008                         lsp = SYSCTL_CHILDREN(oidp);
1009                         if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1010                                 len, level+1, oidpp))
1011                                 return (0);
1012                         goto emptynode;
1013                 }
1014
1015                 if (oidp->oid_number < *name)
1016                         continue;
1017
1018                 if (oidp->oid_number > *name) {
1019                         if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1020                                 return (0);
1021                         if (oidp->oid_handler)
1022                                 return (0);
1023                         lsp = SYSCTL_CHILDREN(oidp);
1024                         if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1025                                 next+1, len, level+1, oidpp))
1026                                 return (0);
1027                         goto next;
1028                 }
1029                 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1030                         continue;
1031
1032                 if (oidp->oid_handler)
1033                         continue;
1034
1035                 lsp = SYSCTL_CHILDREN(oidp);
1036                 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1037                         len, level+1, oidpp))
1038                         return (0);
1039         next:
1040                 namelen = 1;
1041         emptynode:
1042                 *len = level;
1043         }
1044         return (1);
1045 }
1046
1047 static int
1048 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1049 {
1050         int *name = (int *) arg1;
1051         u_int namelen = arg2;
1052         int i, j, error;
1053         struct sysctl_oid *oid;
1054         struct sysctl_oid_list *lsp = &sysctl__children;
1055         struct rm_priotracker tracker;
1056         int newoid[CTL_MAXNAME];
1057
1058         SYSCTL_RLOCK(&tracker);
1059         i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1060         SYSCTL_RUNLOCK(&tracker);
1061         if (i)
1062                 return (ENOENT);
1063         error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1064         return (error);
1065 }
1066
1067 /*
1068  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1069  * capability mode.
1070  */
1071 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1072     sysctl_sysctl_next, "");
1073
1074 static int
1075 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1076 {
1077         struct sysctl_oid *oidp;
1078         struct sysctl_oid_list *lsp = &sysctl__children;
1079         char *p;
1080
1081         SYSCTL_ASSERT_LOCKED();
1082
1083         for (*len = 0; *len < CTL_MAXNAME;) {
1084                 p = strsep(&name, ".");
1085
1086                 oidp = SLIST_FIRST(lsp);
1087                 for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1088                         if (oidp == NULL)
1089                                 return (ENOENT);
1090                         if (strcmp(p, oidp->oid_name) == 0)
1091                                 break;
1092                 }
1093                 *oid++ = oidp->oid_number;
1094                 (*len)++;
1095
1096                 if (name == NULL || *name == '\0') {
1097                         if (oidpp)
1098                                 *oidpp = oidp;
1099                         return (0);
1100                 }
1101
1102                 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1103                         break;
1104
1105                 if (oidp->oid_handler)
1106                         break;
1107
1108                 lsp = SYSCTL_CHILDREN(oidp);
1109         }
1110         return (ENOENT);
1111 }
1112
1113 static int
1114 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1115 {
1116         char *p;
1117         int error, oid[CTL_MAXNAME], len = 0;
1118         struct sysctl_oid *op = NULL;
1119         struct rm_priotracker tracker;
1120         char buf[32];
1121
1122         if (!req->newlen)
1123                 return (ENOENT);
1124         if (req->newlen >= MAXPATHLEN)  /* XXX arbitrary, undocumented */
1125                 return (ENAMETOOLONG);
1126
1127         p = buf;
1128         if (req->newlen >= sizeof(buf))
1129                 p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1130
1131         error = SYSCTL_IN(req, p, req->newlen);
1132         if (error) {
1133                 if (p != buf)
1134                         free(p, M_SYSCTL);
1135                 return (error);
1136         }
1137
1138         p [req->newlen] = '\0';
1139
1140         SYSCTL_RLOCK(&tracker);
1141         error = name2oid(p, oid, &len, &op);
1142         SYSCTL_RUNLOCK(&tracker);
1143
1144         if (p != buf)
1145                 free(p, M_SYSCTL);
1146
1147         if (error)
1148                 return (error);
1149
1150         error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1151         return (error);
1152 }
1153
1154 /*
1155  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1156  * capability mode.
1157  */
1158 SYSCTL_PROC(_sysctl, 3, name2oid,
1159     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1160     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1161
1162 static int
1163 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1164 {
1165         struct sysctl_oid *oid;
1166         struct rm_priotracker tracker;
1167         int error;
1168
1169         SYSCTL_RLOCK(&tracker);
1170         error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1171         if (error)
1172                 goto out;
1173
1174         if (oid->oid_fmt == NULL) {
1175                 error = ENOENT;
1176                 goto out;
1177         }
1178         error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1179         if (error)
1180                 goto out;
1181         error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1182  out:
1183         SYSCTL_RUNLOCK(&tracker);
1184         return (error);
1185 }
1186
1187
1188 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1189     sysctl_sysctl_oidfmt, "");
1190
1191 static int
1192 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1193 {
1194         struct sysctl_oid *oid;
1195         struct rm_priotracker tracker;
1196         int error;
1197
1198         SYSCTL_RLOCK(&tracker);
1199         error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1200         if (error)
1201                 goto out;
1202
1203         if (oid->oid_descr == NULL) {
1204                 error = ENOENT;
1205                 goto out;
1206         }
1207         error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1208  out:
1209         SYSCTL_RUNLOCK(&tracker);
1210         return (error);
1211 }
1212
1213 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1214     sysctl_sysctl_oiddescr, "");
1215
1216 /*
1217  * Default "handler" functions.
1218  */
1219
1220 /*
1221  * Handle a bool.
1222  * Two cases:
1223  *     a variable:  point arg1 at it.
1224  *     a constant:  pass it in arg2.
1225  */
1226
1227 int
1228 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1229 {
1230         uint8_t temp;
1231         int error;
1232
1233         /*
1234          * Attempt to get a coherent snapshot by making a copy of the data.
1235          */
1236         if (arg1)
1237                 temp = *(bool *)arg1 ? 1 : 0;
1238         else
1239                 temp = arg2 ? 1 : 0;
1240
1241         error = SYSCTL_OUT(req, &temp, sizeof(temp));
1242         if (error || !req->newptr)
1243                 return (error);
1244
1245         if (!arg1)
1246                 error = EPERM;
1247         else {
1248                 error = SYSCTL_IN(req, &temp, sizeof(temp));
1249                 if (!error)
1250                         *(bool *)arg1 = temp ? 1 : 0;
1251         }
1252         return (error);
1253 }
1254
1255 /*
1256  * Handle an int8_t, signed or unsigned.
1257  * Two cases:
1258  *     a variable:  point arg1 at it.
1259  *     a constant:  pass it in arg2.
1260  */
1261
1262 int
1263 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1264 {
1265         int8_t tmpout;
1266         int error = 0;
1267
1268         /*
1269          * Attempt to get a coherent snapshot by making a copy of the data.
1270          */
1271         if (arg1)
1272                 tmpout = *(int8_t *)arg1;
1273         else
1274                 tmpout = arg2;
1275         error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1276
1277         if (error || !req->newptr)
1278                 return (error);
1279
1280         if (!arg1)
1281                 error = EPERM;
1282         else
1283                 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1284         return (error);
1285 }
1286
1287 /*
1288  * Handle an int16_t, signed or unsigned.
1289  * Two cases:
1290  *     a variable:  point arg1 at it.
1291  *     a constant:  pass it in arg2.
1292  */
1293
1294 int
1295 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1296 {
1297         int16_t tmpout;
1298         int error = 0;
1299
1300         /*
1301          * Attempt to get a coherent snapshot by making a copy of the data.
1302          */
1303         if (arg1)
1304                 tmpout = *(int16_t *)arg1;
1305         else
1306                 tmpout = arg2;
1307         error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1308
1309         if (error || !req->newptr)
1310                 return (error);
1311
1312         if (!arg1)
1313                 error = EPERM;
1314         else
1315                 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1316         return (error);
1317 }
1318
1319 /*
1320  * Handle an int32_t, signed or unsigned.
1321  * Two cases:
1322  *     a variable:  point arg1 at it.
1323  *     a constant:  pass it in arg2.
1324  */
1325
1326 int
1327 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1328 {
1329         int32_t tmpout;
1330         int error = 0;
1331
1332         /*
1333          * Attempt to get a coherent snapshot by making a copy of the data.
1334          */
1335         if (arg1)
1336                 tmpout = *(int32_t *)arg1;
1337         else
1338                 tmpout = arg2;
1339         error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1340
1341         if (error || !req->newptr)
1342                 return (error);
1343
1344         if (!arg1)
1345                 error = EPERM;
1346         else
1347                 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1348         return (error);
1349 }
1350
1351 /*
1352  * Handle an int, signed or unsigned.
1353  * Two cases:
1354  *     a variable:  point arg1 at it.
1355  *     a constant:  pass it in arg2.
1356  */
1357
1358 int
1359 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1360 {
1361         int tmpout, error = 0;
1362
1363         /*
1364          * Attempt to get a coherent snapshot by making a copy of the data.
1365          */
1366         if (arg1)
1367                 tmpout = *(int *)arg1;
1368         else
1369                 tmpout = arg2;
1370         error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1371
1372         if (error || !req->newptr)
1373                 return (error);
1374
1375         if (!arg1)
1376                 error = EPERM;
1377         else
1378                 error = SYSCTL_IN(req, arg1, sizeof(int));
1379         return (error);
1380 }
1381
1382 /*
1383  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1384  * Note: this is used by TCP.
1385  */
1386
1387 int
1388 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1389 {
1390         int error, s, tt;
1391
1392         tt = *(int *)arg1;
1393         s = (int)((int64_t)tt * 1000 / hz);
1394
1395         error = sysctl_handle_int(oidp, &s, 0, req);
1396         if (error || !req->newptr)
1397                 return (error);
1398
1399         tt = (int)((int64_t)s * hz / 1000);
1400         if (tt < 1)
1401                 return (EINVAL);
1402
1403         *(int *)arg1 = tt;
1404         return (0);
1405 }
1406
1407
1408 /*
1409  * Handle a long, signed or unsigned.
1410  * Two cases:
1411  *     a variable:  point arg1 at it.
1412  *     a constant:  pass it in arg2.
1413  */
1414
1415 int
1416 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1417 {
1418         int error = 0;
1419         long tmplong;
1420 #ifdef SCTL_MASK32
1421         int tmpint;
1422 #endif
1423
1424         /*
1425          * Attempt to get a coherent snapshot by making a copy of the data.
1426          */
1427         if (arg1)
1428                 tmplong = *(long *)arg1;
1429         else
1430                 tmplong = arg2;
1431 #ifdef SCTL_MASK32
1432         if (req->flags & SCTL_MASK32) {
1433                 tmpint = tmplong;
1434                 error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1435         } else
1436 #endif
1437                 error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1438
1439         if (error || !req->newptr)
1440                 return (error);
1441
1442         if (!arg1)
1443                 error = EPERM;
1444 #ifdef SCTL_MASK32
1445         else if (req->flags & SCTL_MASK32) {
1446                 error = SYSCTL_IN(req, &tmpint, sizeof(int));
1447                 *(long *)arg1 = (long)tmpint;
1448         }
1449 #endif
1450         else
1451                 error = SYSCTL_IN(req, arg1, sizeof(long));
1452         return (error);
1453 }
1454
1455 /*
1456  * Handle a 64 bit int, signed or unsigned.
1457  * Two cases:
1458  *     a variable:  point arg1 at it.
1459  *     a constant:  pass it in arg2.
1460  */
1461 int
1462 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1463 {
1464         int error = 0;
1465         uint64_t tmpout;
1466
1467         /*
1468          * Attempt to get a coherent snapshot by making a copy of the data.
1469          */
1470         if (arg1)
1471                 tmpout = *(uint64_t *)arg1;
1472         else
1473                 tmpout = arg2;
1474         error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1475
1476         if (error || !req->newptr)
1477                 return (error);
1478
1479         if (!arg1)
1480                 error = EPERM;
1481         else
1482                 error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1483         return (error);
1484 }
1485
1486 /*
1487  * Handle our generic '\0' terminated 'C' string.
1488  * Two cases:
1489  *      a variable string:  point arg1 at it, arg2 is max length.
1490  *      a constant string:  point arg1 at it, arg2 is zero.
1491  */
1492
1493 int
1494 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1495 {
1496         size_t outlen;
1497         int error = 0, ro_string = 0;
1498
1499         /*
1500          * A zero-length buffer indicates a fixed size read-only
1501          * string:
1502          */
1503         if (arg2 == 0) {
1504                 arg2 = strlen((char *)arg1) + 1;
1505                 ro_string = 1;
1506         }
1507
1508         if (req->oldptr != NULL) {
1509                 char *tmparg;
1510
1511                 if (ro_string) {
1512                         tmparg = arg1;
1513                 } else {
1514                         /* try to make a coherent snapshot of the string */
1515                         tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1516                         memcpy(tmparg, arg1, arg2);
1517                 }
1518
1519                 outlen = strnlen(tmparg, arg2 - 1) + 1;
1520                 error = SYSCTL_OUT(req, tmparg, outlen);
1521
1522                 if (!ro_string)
1523                         free(tmparg, M_SYSCTLTMP);
1524         } else {
1525                 outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1526                 error = SYSCTL_OUT(req, NULL, outlen);
1527         }
1528         if (error || !req->newptr)
1529                 return (error);
1530
1531         if ((req->newlen - req->newidx) >= arg2) {
1532                 error = EINVAL;
1533         } else {
1534                 arg2 = (req->newlen - req->newidx);
1535                 error = SYSCTL_IN(req, arg1, arg2);
1536                 ((char *)arg1)[arg2] = '\0';
1537         }
1538         return (error);
1539 }
1540
1541 /*
1542  * Handle any kind of opaque data.
1543  * arg1 points to it, arg2 is the size.
1544  */
1545
1546 int
1547 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1548 {
1549         int error, tries;
1550         u_int generation;
1551         struct sysctl_req req2;
1552
1553         /*
1554          * Attempt to get a coherent snapshot, by using the thread
1555          * pre-emption counter updated from within mi_switch() to
1556          * determine if we were pre-empted during a bcopy() or
1557          * copyout(). Make 3 attempts at doing this before giving up.
1558          * If we encounter an error, stop immediately.
1559          */
1560         tries = 0;
1561         req2 = *req;
1562 retry:
1563         generation = curthread->td_generation;
1564         error = SYSCTL_OUT(req, arg1, arg2);
1565         if (error)
1566                 return (error);
1567         tries++;
1568         if (generation != curthread->td_generation && tries < 3) {
1569                 *req = req2;
1570                 goto retry;
1571         }
1572
1573         error = SYSCTL_IN(req, arg1, arg2);
1574
1575         return (error);
1576 }
1577
1578 /*
1579  * Transfer functions to/from kernel space.
1580  * XXX: rather untested at this point
1581  */
1582 static int
1583 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1584 {
1585         size_t i = 0;
1586
1587         if (req->oldptr) {
1588                 i = l;
1589                 if (req->oldlen <= req->oldidx)
1590                         i = 0;
1591                 else
1592                         if (i > req->oldlen - req->oldidx)
1593                                 i = req->oldlen - req->oldidx;
1594                 if (i > 0)
1595                         bcopy(p, (char *)req->oldptr + req->oldidx, i);
1596         }
1597         req->oldidx += l;
1598         if (req->oldptr && i != l)
1599                 return (ENOMEM);
1600         return (0);
1601 }
1602
1603 static int
1604 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1605 {
1606         if (!req->newptr)
1607                 return (0);
1608         if (req->newlen - req->newidx < l)
1609                 return (EINVAL);
1610         bcopy((char *)req->newptr + req->newidx, p, l);
1611         req->newidx += l;
1612         return (0);
1613 }
1614
1615 int
1616 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1617     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1618 {
1619         int error = 0;
1620         struct sysctl_req req;
1621
1622         bzero(&req, sizeof req);
1623
1624         req.td = td;
1625         req.flags = flags;
1626
1627         if (oldlenp) {
1628                 req.oldlen = *oldlenp;
1629         }
1630         req.validlen = req.oldlen;
1631
1632         if (old) {
1633                 req.oldptr= old;
1634         }
1635
1636         if (new != NULL) {
1637                 req.newlen = newlen;
1638                 req.newptr = new;
1639         }
1640
1641         req.oldfunc = sysctl_old_kernel;
1642         req.newfunc = sysctl_new_kernel;
1643         req.lock = REQ_UNWIRED;
1644
1645         error = sysctl_root(0, name, namelen, &req);
1646
1647         if (req.lock == REQ_WIRED && req.validlen > 0)
1648                 vsunlock(req.oldptr, req.validlen);
1649
1650         if (error && error != ENOMEM)
1651                 return (error);
1652
1653         if (retval) {
1654                 if (req.oldptr && req.oldidx > req.validlen)
1655                         *retval = req.validlen;
1656                 else
1657                         *retval = req.oldidx;
1658         }
1659         return (error);
1660 }
1661
1662 int
1663 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1664     void *new, size_t newlen, size_t *retval, int flags)
1665 {
1666         int oid[CTL_MAXNAME];
1667         size_t oidlen, plen;
1668         int error;
1669
1670         oid[0] = 0;             /* sysctl internal magic */
1671         oid[1] = 3;             /* name2oid */
1672         oidlen = sizeof(oid);
1673
1674         error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1675             (void *)name, strlen(name), &plen, flags);
1676         if (error)
1677                 return (error);
1678
1679         error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1680             new, newlen, retval, flags);
1681         return (error);
1682 }
1683
1684 /*
1685  * Transfer function to/from user space.
1686  */
1687 static int
1688 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1689 {
1690         size_t i, len, origidx;
1691         int error;
1692
1693         origidx = req->oldidx;
1694         req->oldidx += l;
1695         if (req->oldptr == NULL)
1696                 return (0);
1697         /*
1698          * If we have not wired the user supplied buffer and we are currently
1699          * holding locks, drop a witness warning, as it's possible that
1700          * write operations to the user page can sleep.
1701          */
1702         if (req->lock != REQ_WIRED)
1703                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1704                     "sysctl_old_user()");
1705         i = l;
1706         len = req->validlen;
1707         if (len <= origidx)
1708                 i = 0;
1709         else {
1710                 if (i > len - origidx)
1711                         i = len - origidx;
1712                 if (req->lock == REQ_WIRED) {
1713                         error = copyout_nofault(p, (char *)req->oldptr +
1714                             origidx, i);
1715                 } else
1716                         error = copyout(p, (char *)req->oldptr + origidx, i);
1717                 if (error != 0)
1718                         return (error);
1719         }
1720         if (i < l)
1721                 return (ENOMEM);
1722         return (0);
1723 }
1724
1725 static int
1726 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1727 {
1728         int error;
1729
1730         if (!req->newptr)
1731                 return (0);
1732         if (req->newlen - req->newidx < l)
1733                 return (EINVAL);
1734         WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1735             "sysctl_new_user()");
1736         error = copyin((char *)req->newptr + req->newidx, p, l);
1737         req->newidx += l;
1738         return (error);
1739 }
1740
1741 /*
1742  * Wire the user space destination buffer.  If set to a value greater than
1743  * zero, the len parameter limits the maximum amount of wired memory.
1744  */
1745 int
1746 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1747 {
1748         int ret;
1749         size_t wiredlen;
1750
1751         wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1752         ret = 0;
1753         if (req->lock != REQ_WIRED && req->oldptr &&
1754             req->oldfunc == sysctl_old_user) {
1755                 if (wiredlen != 0) {
1756                         ret = vslock(req->oldptr, wiredlen);
1757                         if (ret != 0) {
1758                                 if (ret != ENOMEM)
1759                                         return (ret);
1760                                 wiredlen = 0;
1761                         }
1762                 }
1763                 req->lock = REQ_WIRED;
1764                 req->validlen = wiredlen;
1765         }
1766         return (0);
1767 }
1768
1769 int
1770 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1771     int *nindx, struct sysctl_req *req)
1772 {
1773         struct sysctl_oid_list *lsp;
1774         struct sysctl_oid *oid;
1775         int indx;
1776
1777         SYSCTL_ASSERT_LOCKED();
1778         lsp = &sysctl__children;
1779         indx = 0;
1780         while (indx < CTL_MAXNAME) {
1781                 SLIST_FOREACH(oid, lsp, oid_link) {
1782                         if (oid->oid_number == name[indx])
1783                                 break;
1784                 }
1785                 if (oid == NULL)
1786                         return (ENOENT);
1787
1788                 indx++;
1789                 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1790                         if (oid->oid_handler != NULL || indx == namelen) {
1791                                 *noid = oid;
1792                                 if (nindx != NULL)
1793                                         *nindx = indx;
1794                                 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1795                                     ("%s found DYING node %p", __func__, oid));
1796                                 return (0);
1797                         }
1798                         lsp = SYSCTL_CHILDREN(oid);
1799                 } else if (indx == namelen) {
1800                         if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
1801                                 return (ENOENT);
1802                         *noid = oid;
1803                         if (nindx != NULL)
1804                                 *nindx = indx;
1805                         KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1806                             ("%s found DYING node %p", __func__, oid));
1807                         return (0);
1808                 } else {
1809                         return (ENOTDIR);
1810                 }
1811         }
1812         return (ENOENT);
1813 }
1814
1815 /*
1816  * Traverse our tree, and find the right node, execute whatever it points
1817  * to, and return the resulting error code.
1818  */
1819
1820 static int
1821 sysctl_root(SYSCTL_HANDLER_ARGS)
1822 {
1823         struct sysctl_oid *oid;
1824         struct rm_priotracker tracker;
1825         int error, indx, lvl;
1826
1827         SYSCTL_RLOCK(&tracker);
1828
1829         error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1830         if (error)
1831                 goto out;
1832
1833         if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1834                 /*
1835                  * You can't call a sysctl when it's a node, but has
1836                  * no handler.  Inform the user that it's a node.
1837                  * The indx may or may not be the same as namelen.
1838                  */
1839                 if (oid->oid_handler == NULL) {
1840                         error = EISDIR;
1841                         goto out;
1842                 }
1843         }
1844
1845         /* Is this sysctl writable? */
1846         if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
1847                 error = EPERM;
1848                 goto out;
1849         }
1850
1851         KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1852
1853 #ifdef CAPABILITY_MODE
1854         /*
1855          * If the process is in capability mode, then don't permit reading or
1856          * writing unless specifically granted for the node.
1857          */
1858         if (IN_CAPABILITY_MODE(req->td)) {
1859                 if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
1860                     (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
1861                         error = EPERM;
1862                         goto out;
1863                 }
1864         }
1865 #endif
1866
1867         /* Is this sysctl sensitive to securelevels? */
1868         if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1869                 lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1870                 error = securelevel_gt(req->td->td_ucred, lvl);
1871                 if (error)
1872                         goto out;
1873         }
1874
1875         /* Is this sysctl writable by only privileged users? */
1876         if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1877                 int priv;
1878
1879                 if (oid->oid_kind & CTLFLAG_PRISON)
1880                         priv = PRIV_SYSCTL_WRITEJAIL;
1881 #ifdef VIMAGE
1882                 else if ((oid->oid_kind & CTLFLAG_VNET) &&
1883                      prison_owns_vnet(req->td->td_ucred))
1884                         priv = PRIV_SYSCTL_WRITEJAIL;
1885 #endif
1886                 else
1887                         priv = PRIV_SYSCTL_WRITE;
1888                 error = priv_check(req->td, priv);
1889                 if (error)
1890                         goto out;
1891         }
1892
1893         if (!oid->oid_handler) {
1894                 error = EINVAL;
1895                 goto out;
1896         }
1897
1898         if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1899                 arg1 = (int *)arg1 + indx;
1900                 arg2 -= indx;
1901         } else {
1902                 arg1 = oid->oid_arg1;
1903                 arg2 = oid->oid_arg2;
1904         }
1905 #ifdef MAC
1906         error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1907             req);
1908         if (error != 0)
1909                 goto out;
1910 #endif
1911 #ifdef VIMAGE
1912         if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1913                 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1914 #endif
1915         error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
1916
1917 out:
1918         SYSCTL_RUNLOCK(&tracker);
1919         return (error);
1920 }
1921
1922 #ifndef _SYS_SYSPROTO_H_
1923 struct sysctl_args {
1924         int     *name;
1925         u_int   namelen;
1926         void    *old;
1927         size_t  *oldlenp;
1928         void    *new;
1929         size_t  newlen;
1930 };
1931 #endif
1932 int
1933 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1934 {
1935         int error, i, name[CTL_MAXNAME];
1936         size_t j;
1937
1938         if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1939                 return (EINVAL);
1940
1941         error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1942         if (error)
1943                 return (error);
1944
1945         error = userland_sysctl(td, name, uap->namelen,
1946                 uap->old, uap->oldlenp, 0,
1947                 uap->new, uap->newlen, &j, 0);
1948         if (error && error != ENOMEM)
1949                 return (error);
1950         if (uap->oldlenp) {
1951                 i = copyout(&j, uap->oldlenp, sizeof(j));
1952                 if (i)
1953                         return (i);
1954         }
1955         return (error);
1956 }
1957
1958 /*
1959  * This is used from various compatibility syscalls too.  That's why name
1960  * must be in kernel space.
1961  */
1962 int
1963 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1964     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1965     int flags)
1966 {
1967         int error = 0, memlocked;
1968         struct sysctl_req req;
1969
1970         bzero(&req, sizeof req);
1971
1972         req.td = td;
1973         req.flags = flags;
1974
1975         if (oldlenp) {
1976                 if (inkernel) {
1977                         req.oldlen = *oldlenp;
1978                 } else {
1979                         error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1980                         if (error)
1981                                 return (error);
1982                 }
1983         }
1984         req.validlen = req.oldlen;
1985         req.oldptr = old;
1986
1987         if (new != NULL) {
1988                 req.newlen = newlen;
1989                 req.newptr = new;
1990         }
1991
1992         req.oldfunc = sysctl_old_user;
1993         req.newfunc = sysctl_new_user;
1994         req.lock = REQ_UNWIRED;
1995
1996 #ifdef KTRACE
1997         if (KTRPOINT(curthread, KTR_SYSCTL))
1998                 ktrsysctl(name, namelen);
1999 #endif
2000
2001         if (req.oldptr && req.oldlen > PAGE_SIZE) {
2002                 memlocked = 1;
2003                 sx_xlock(&sysctlmemlock);
2004         } else
2005                 memlocked = 0;
2006         CURVNET_SET(TD_TO_VNET(td));
2007
2008         for (;;) {
2009                 req.oldidx = 0;
2010                 req.newidx = 0;
2011                 error = sysctl_root(0, name, namelen, &req);
2012                 if (error != EAGAIN)
2013                         break;
2014                 kern_yield(PRI_USER);
2015         }
2016
2017         CURVNET_RESTORE();
2018
2019         if (req.lock == REQ_WIRED && req.validlen > 0)
2020                 vsunlock(req.oldptr, req.validlen);
2021         if (memlocked)
2022                 sx_xunlock(&sysctlmemlock);
2023
2024         if (error && error != ENOMEM)
2025                 return (error);
2026
2027         if (retval) {
2028                 if (req.oldptr && req.oldidx > req.validlen)
2029                         *retval = req.validlen;
2030                 else
2031                         *retval = req.oldidx;
2032         }
2033         return (error);
2034 }
2035
2036 /*
2037  * Drain into a sysctl struct.  The user buffer should be wired if a page
2038  * fault would cause issue.
2039  */
2040 static int
2041 sbuf_sysctl_drain(void *arg, const char *data, int len)
2042 {
2043         struct sysctl_req *req = arg;
2044         int error;
2045
2046         error = SYSCTL_OUT(req, data, len);
2047         KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2048         return (error == 0 ? len : -error);
2049 }
2050
2051 struct sbuf *
2052 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2053     struct sysctl_req *req)
2054 {
2055
2056         /* Supply a default buffer size if none given. */
2057         if (buf == NULL && length == 0)
2058                 length = 64;
2059         s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2060         sbuf_set_drain(s, sbuf_sysctl_drain, req);
2061         return (s);
2062 }