sys/kern/kern_conf.c

   1 /*-
   2  * Parts Copyright (c) 1995 Terrence R. Lambert
   3  * Copyright (c) 1995 Julian R. Elischer
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 3. All advertising materials mentioning features or use of this software
  15  *    must display the following acknowledgement:
  16  *      This product includes software developed by Terrence R. Lambert.
  17  * 4. The name Terrence R. Lambert may not be used to endorse or promote
  18  *    products derived from this software without specific prior written
  19  *    permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY Julian R. Elischer ``AS IS'' AND ANY
  22  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE TERRENCE R. LAMBERT BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  * $FreeBSD: src/sys/kern/kern_conf.c,v 1.73.2.3 2003/03/10 02:18:25 imp Exp $
  34  * $DragonFly: src/sys/kern/kern_conf.c,v 1.23 2007/05/09 00:53:34 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/kernel.h>
  39 #include <sys/sysctl.h>
  40 #include <sys/systm.h>
  41 #include <sys/module.h>
  42 #include <sys/malloc.h>
  43 #include <sys/conf.h>
  44 #include <sys/vnode.h>
  45 #include <sys/queue.h>
  46 #include <sys/device.h>
  47 #include <machine/stdarg.h>
  48
  49 #include <sys/sysref2.h>
  50
  51 static void cdev_terminate(struct cdev *dev);
  52
  53 MALLOC_DEFINE(M_DEVT, "cdev_t", "dev_t storage");
  54
  55 /*
  56  * SYSREF Integration - reference counting, allocation,
  57  * sysid and syslink integration.
  58  */
  59 static struct sysref_class     cdev_sysref_class = {
  60         .name =         "cdev",
  61         .mtype =        M_DEVT,
  62         .proto =        SYSREF_PROTO_DEV,
  63         .offset =       offsetof(struct cdev, si_sysref),
  64         .objsize =      sizeof(struct cdev),
  65         .mag_capacity = 32,
  66         .flags =        0,
  67         .ops =  {
  68                 .terminate = (sysref_terminate_func_t)cdev_terminate
  69         }
  70 };
  71
  72 /*
  73  * This is the number of hash-buckets.  Experiements with 'real-life'
  74  * udev_t's show that a prime halfway between two powers of two works
  75  * best.
  76  */
  77 #define DEVT_HASH 128   /* must be power of 2 */
  78 static LIST_HEAD(, cdev) dev_hash[DEVT_HASH];
  79
  80 static int free_devt;
  81 SYSCTL_INT(_debug, OID_AUTO, free_devt, CTLFLAG_RW, &free_devt, 0, "");
  82 int dev_ref_debug = 0;
  83 SYSCTL_INT(_debug, OID_AUTO, dev_refs, CTLFLAG_RW, &dev_ref_debug, 0, "");
  84
  85 /*
  86  * cdev_t and u_dev_t primitives.  Note that the major number is always
  87  * extracted from si_umajor, not from si_devsw, because si_devsw is replaced
  88  * when a device is destroyed.
  89  */
  90 int
  91 major(cdev_t dev)
  92 {
  93         if (dev == NULL)
  94                 return NOUDEV;
  95         return(dev->si_umajor);
  96 }
  97
  98 int
  99 minor(cdev_t dev)
 100 {
 101         if (dev == NULL)
 102                 return NOUDEV;
 103         return(dev->si_uminor);
 104 }
 105
 106 /*
 107  * Compatibility function with old udev_t format to convert the
 108  * non-consecutive minor space into a consecutive minor space.
 109  */
 110 int
 111 lminor(cdev_t dev)
 112 {
 113         int y;
 114
 115         if (dev == NULL)
 116                 return NOUDEV;
 117         y = dev->si_uminor;
 118         if (y & 0x0000ff00)
 119                 return NOUDEV;
 120         return ((y & 0xff) | (y >> 8));
 121 }
 122
 123 /*
 124  * This is a bit complex because devices are always created relative to
 125  * a particular cdevsw, including 'hidden' cdevsw's (such as the raw device
 126  * backing a disk subsystem overlay), so we have to compare both the
 127  * devsw and udev fields to locate the correct device.
 128  *
 129  * The device is created if it does not already exist.  If SI_ADHOC is not
 130  * set the device will be referenced (once) and SI_ADHOC will be set.
 131  * The caller must explicitly add additional references to the device if
 132  * the caller wishes to track additional references.
 133  *
 134  * NOTE: The passed ops vector must normally match the device.  This is
 135  * because the kernel may create shadow devices that are INVISIBLE TO
 136  * USERLAND.  For example, the block device backing a disk is created
 137  * as a shadow underneath the user-visible disklabel management device.
 138  * Sometimes a device ops vector can be overridden, such as by /dev/console.
 139  * In this case and this case only we allow a match when the ops vector
 140  * otherwise would not match.
 141  */
 142 static
 143 int
 144 __devthash(int x, int y)
 145 {
 146         return(((x << 2) ^ y) & (DEVT_HASH - 1));
 147 }
 148
 149 static
 150 cdev_t
 151 hashdev(struct dev_ops *ops, int x, int y, int allow_intercept)
 152 {
 153         struct cdev *si;
 154         int hash;
 155
 156         hash = __devthash(x, y);
 157         LIST_FOREACH(si, &dev_hash[hash], si_hash) {
 158                 if (si->si_umajor == x && si->si_uminor == y) {
 159                         if (si->si_ops == ops)
 160                                 return (si);
 161                         if (allow_intercept && (si->si_flags & SI_INTERCEPTED))
 162                                 return (si);
 163                 }
 164         }
 165         si = sysref_alloc(&cdev_sysref_class);
 166         si->si_ops = ops;
 167         si->si_flags |= SI_HASHED | SI_ADHOC;
 168         si->si_umajor = x;
 169         si->si_uminor = y;
 170         LIST_INSERT_HEAD(&dev_hash[hash], si, si_hash);
 171         sysref_activate(&si->si_sysref);
 172
 173         dev_dclone(si);
 174         if (ops != &dead_dev_ops)
 175                 ++ops->head.refs;
 176         if (dev_ref_debug) {
 177                 kprintf("create    dev %p %s(minor=%08x) refs=%d\n",
 178                         si, devtoname(si), y,
 179                         si->si_sysref.refcnt);
 180         }
 181         return (si);
 182 }
 183
 184 /*
 185  * Convert a device pointer to an old style device number.  Return NOUDEV
 186  * if the device is invalid or if the device (maj,min) cannot be converted
 187  * to an old style udev_t.
 188  */
 189 udev_t
 190 dev2udev(cdev_t dev)
 191 {
 192         if (dev == NULL)
 193                 return NOUDEV;
 194         if ((dev->si_umajor & 0xffffff00) ||
 195             (dev->si_uminor & 0x0000ff00)) {
 196                 return NOUDEV;
 197         }
 198         return((dev->si_umajor << 8) | dev->si_uminor);
 199 }
 200
 201 /*
 202  * Convert a device number to a device pointer.  The device is referenced
 203  * ad-hoc, meaning that the caller should call reference_dev() if it wishes
 204  * to keep ahold of the returned structure long term.
 205  *
 206  * The returned device is associated with the currently installed cdevsw
 207  * for the requested major number.  NULL is returned if the major number
 208  * has not been registered.
 209  */
 210 cdev_t
 211 udev2dev(udev_t x, int b)
 212 {
 213         cdev_t dev;
 214         struct dev_ops *ops;
 215
 216         if (x == NOUDEV || b != 0)
 217                 return(NULL);
 218         ops = dev_ops_get(umajor(x), uminor(x));
 219         if (ops == NULL)
 220                 return(NULL);
 221         dev = hashdev(ops, umajor(x), uminor(x), TRUE);
 222         return(dev);
 223 }
 224
 225 int
 226 dev_is_good(cdev_t dev)
 227 {
 228         if (dev != NULL && dev->si_ops != &dead_dev_ops)
 229                 return(1);
 230         return(0);
 231 }
 232
 233 /*
 234  * Various user device number extraction and conversion routines
 235  */
 236 int
 237 uminor(udev_t dev)
 238 {
 239         if (dev == NOUDEV)
 240                 return(-1);
 241         return(dev & 0xffff00ff);
 242 }
 243
 244 int
 245 umajor(udev_t dev)
 246 {
 247         if (dev == NOUDEV)
 248                 return(-1);
 249         return((dev & 0xff00) >> 8);
 250 }
 251
 252 udev_t
 253 makeudev(int x, int y)
 254 {
 255         if ((x & 0xffffff00) || (y & 0x0000ff00))
 256                 return NOUDEV;
 257         return ((x << 8) | y);
 258 }
 259
 260 /*
 261  * Create an internal or external device.
 262  *
 263  * Device majors can be overloaded and used directly by the kernel without
 264  * conflict, but userland will only see the particular device major that
 265  * has been installed with dev_ops_add().
 266  *
 267  * This routine creates and returns an unreferenced ad-hoc entry for the
 268  * device which will remain intact until the device is destroyed.  If the
 269  * caller intends to store the device pointer it must call reference_dev()
 270  * to retain a real reference to the device.
 271  *
 272  * If an entry already exists, this function will set (or override)
 273  * its cred requirements and name (XXX DEVFS interface).
 274  */
 275 cdev_t
 276 make_dev(struct dev_ops *ops, int minor, uid_t uid, gid_t gid,
 277         int perms, const char *fmt, ...)
 278 {
 279         cdev_t  dev;
 280         __va_list ap;
 281         int i;
 282
 283         /*
 284          * compile the cdevsw and install the device
 285          */
 286         compile_dev_ops(ops);
 287         dev = hashdev(ops, ops->head.maj, minor, FALSE);
 288
 289         /*
 290          * Set additional fields (XXX DEVFS interface goes here)
 291          */
 292         __va_start(ap, fmt);
 293         i = kvcprintf(fmt, NULL, dev->si_name, 32, ap);
 294         dev->si_name[i] = '\0';
 295         __va_end(ap);
 296
 297         return (dev);
 298 }
 299
 300 /*
 301  * This function is similar to make_dev() but no cred information or name
 302  * need be specified.
 303  */
 304 cdev_t
 305 make_adhoc_dev(struct dev_ops *ops, int minor)
 306 {
 307         cdev_t dev;
 308
 309         dev = hashdev(ops, ops->head.maj, minor, FALSE);
 310         return(dev);
 311 }
 312
 313 /*
 314  * This function is similar to make_dev() except the new device is created
 315  * using an old device as a template.
 316  */
 317 cdev_t
 318 make_sub_dev(cdev_t odev, int minor)
 319 {
 320         cdev_t  dev;
 321
 322         dev = hashdev(odev->si_ops, odev->si_umajor, minor, FALSE);
 323
 324         /*
 325          * Copy cred requirements and name info XXX DEVFS.
 326          */
 327         if (dev->si_name[0] == 0 && odev->si_name[0])
 328                 bcopy(odev->si_name, dev->si_name, sizeof(dev->si_name));
 329         return (dev);
 330 }
 331
 332 cdev_t
 333 get_dev(int x, int y)
 334 {
 335         cdev_t dev;
 336         struct dev_ops *ops;
 337
 338         if (x == NOUDEV)
 339                 return(NULL);
 340         ops = dev_ops_get(x, y);
 341         if (ops == NULL)
 342                 return(NULL);
 343         dev = hashdev(ops, x, y, TRUE);
 344         return(dev);
 345 }
 346
 347 /*
 348  * destroy_dev() removes the adhoc association for a device and revectors
 349  * its ops to &dead_dev_ops.
 350  *
 351  * This routine releases the reference count associated with the ADHOC
 352  * entry, plus releases the reference count held by the caller.  What this
 353  * means is that you should not call destroy_dev(make_dev(...)), because
 354  * make_dev() does not bump the reference count (beyond what it needs to
 355  * create the ad-hoc association).  Any procedure that intends to destroy
 356  * a device must have its own reference to it first.
 357  */
 358 void
 359 destroy_dev(cdev_t dev)
 360 {
 361         int hash;
 362
 363         if (dev == NULL)
 364                 return;
 365         if ((dev->si_flags & SI_ADHOC) == 0) {
 366                 release_dev(dev);
 367                 return;
 368         }
 369         if (dev_ref_debug) {
 370                 kprintf("destroy   dev %p %s(minor=%08x) refs=%d\n",
 371                         dev, devtoname(dev), dev->si_uminor,
 372                         dev->si_sysref.refcnt);
 373         }
 374         if (dev->si_sysref.refcnt < 2) {
 375                 kprintf("destroy_dev(): too few references on device! "
 376                         "%p %s(minor=%08x) refs=%d\n",
 377                     dev, devtoname(dev), dev->si_uminor,
 378                     dev->si_sysref.refcnt);
 379         }
 380         dev->si_flags &= ~SI_ADHOC;
 381         if (dev->si_flags & SI_HASHED) {
 382                 hash = __devthash(dev->si_umajor, dev->si_uminor);
 383                 LIST_REMOVE(dev, si_hash);
 384                 dev->si_flags &= ~SI_HASHED;
 385         }
 386
 387         /*
 388          * We have to release the ops reference before we replace the
 389          * device switch with dead_dev_ops.
 390          */
 391         if (dead_dev_ops.d_strategy == NULL)
 392                 compile_dev_ops(&dead_dev_ops);
 393         if (dev->si_ops && dev->si_ops != &dead_dev_ops)
 394                 dev_ops_release(dev->si_ops);
 395         dev->si_drv1 = NULL;
 396         dev->si_drv2 = NULL;
 397         dev->si_ops = &dead_dev_ops;
 398         sysref_put(&dev->si_sysref);    /* release adhoc association */
 399         release_dev(dev);               /* release callers reference */
 400 }
 401
 402 /*
 403  * Destroy all ad-hoc device associations associated with a domain within a
 404  * device switch.  Only the minor numbers are included in the mask/match
 405  * values.
 406  *
 407  * Unlike the ops functions whos link structures do not contain
 408  * any major bits, this function scans through the dev list via
 409  * si_umajor/si_uminor.
 410  *
 411  * The caller must not include any major bits in the match value.
 412  */
 413 void
 414 destroy_all_devs(struct dev_ops *ops, u_int mask, u_int match)
 415 {
 416         int i;
 417         cdev_t dev;
 418         cdev_t ndev;
 419
 420         for (i = 0; i < DEVT_HASH; ++i) {
 421                 ndev = LIST_FIRST(&dev_hash[i]);
 422                 while ((dev = ndev) != NULL) {
 423                     ndev = LIST_NEXT(dev, si_hash);
 424                     KKASSERT(dev->si_flags & SI_ADHOC);
 425                     if (dev->si_ops == ops &&
 426                         ((u_int)dev->si_uminor & mask) == match
 427                     ) {
 428                         reference_dev(dev);
 429                         destroy_dev(dev);
 430                     }
 431                 }
 432         }
 433 }
 434
 435 /*
 436  * Add a reference to a device.  Callers generally add their own references
 437  * when they are going to store a device node in a variable for long periods
 438  * of time, to prevent a disassociation from free()ing the node.
 439  *
 440  * Also note that a caller that intends to call destroy_dev() must first
 441  * obtain a reference on the device.  The ad-hoc reference you get with
 442  * make_dev() and friends is NOT sufficient to be able to call destroy_dev().
 443  */
 444 cdev_t
 445 reference_dev(cdev_t dev)
 446 {
 447         if (dev != NULL) {
 448                 sysref_get(&dev->si_sysref);
 449                 if (dev_ref_debug) {
 450                         kprintf("reference dev %p %s(minor=%08x) refs=%d\n",
 451                             dev, devtoname(dev), dev->si_uminor,
 452                             dev->si_sysref.refcnt);
 453                 }
 454         }
 455         return(dev);
 456 }
 457
 458 /*
 459  * release a reference on a device.  The device will be terminated when the
 460  * last reference has been released.
 461  *
 462  * NOTE: we must use si_umajor to figure out the original major number,
 463  * because si_ops could already be pointing at dead_dev_ops.
 464  */
 465 void
 466 release_dev(cdev_t dev)
 467 {
 468         if (dev == NULL)
 469                 return;
 470         sysref_put(&dev->si_sysref);
 471 }
 472
 473 static
 474 void
 475 cdev_terminate(struct cdev *dev)
 476 {
 477         int messedup = 0;
 478
 479         if (dev_ref_debug) {
 480                 kprintf("release   dev %p %s(minor=%08x) refs=%d\n",
 481                         dev, devtoname(dev), dev->si_uminor,
 482                         dev->si_sysref.refcnt);
 483         }
 484         if (dev->si_flags & SI_ADHOC) {
 485                 kprintf("Warning: illegal final release on ADHOC"
 486                         " device %p(%s), the device was never"
 487                         " destroyed!\n",
 488                         dev, devtoname(dev));
 489                 messedup = 1;
 490         }
 491         if (dev->si_flags & SI_HASHED) {
 492                 kprintf("Warning: last release on device, no call"
 493                         " to destroy_dev() was made! dev %p(%s)\n",
 494                         dev, devtoname(dev));
 495                 reference_dev(dev);
 496                 destroy_dev(dev);
 497                 messedup = 1;
 498         }
 499         if (SLIST_FIRST(&dev->si_hlist) != NULL) {
 500                 kprintf("Warning: last release on device, vnode"
 501                         " associations still exist! dev %p(%s)\n",
 502                         dev, devtoname(dev));
 503                 messedup = 1;
 504         }
 505         if (dev->si_ops && dev->si_ops != &dead_dev_ops) {
 506                 dev_ops_release(dev->si_ops);
 507                 dev->si_ops = NULL;
 508         }
 509         if (messedup == 0)
 510                 sysref_put(&dev->si_sysref);
 511 }
 512
 513 const char *
 514 devtoname(cdev_t dev)
 515 {
 516         int mynor;
 517         int len;
 518         char *p;
 519         const char *dname;
 520
 521         if (dev == NULL)
 522                 return("#nodev");
 523         if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
 524                 p = dev->si_name;
 525                 len = sizeof(dev->si_name);
 526                 if ((dname = dev_dname(dev)) != NULL)
 527                         ksnprintf(p, len, "#%s/", dname);
 528                 else
 529                         ksnprintf(p, len, "#%d/", major(dev));
 530                 len -= strlen(p);
 531                 p += strlen(p);
 532                 mynor = minor(dev);
 533                 if (mynor < 0 || mynor > 255)
 534                         ksnprintf(p, len, "%#x", (u_int)mynor);
 535                 else
 536                         ksnprintf(p, len, "%d", mynor);
 537         }
 538         return (dev->si_name);
 539 }
 540