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         dev->si_uid = uid;
 296         __va_end(ap);
 297
 298         return (dev);
 299 }
 300
 301 /*
 302  * This function is similar to make_dev() but no cred information or name
 303  * need be specified.
 304  */
 305 cdev_t
 306 make_adhoc_dev(struct dev_ops *ops, int minor)
 307 {
 308         cdev_t dev;
 309
 310         dev = hashdev(ops, ops->head.maj, minor, FALSE);
 311         return(dev);
 312 }
 313
 314 /*
 315  * This function is similar to make_dev() except the new device is created
 316  * using an old device as a template.
 317  */
 318 cdev_t
 319 make_sub_dev(cdev_t odev, int minor)
 320 {
 321         cdev_t  dev;
 322
 323         dev = hashdev(odev->si_ops, odev->si_umajor, minor, FALSE);
 324
 325         /*
 326          * Copy cred requirements and name info XXX DEVFS.
 327          */
 328         if (dev->si_name[0] == 0 && odev->si_name[0])
 329                 bcopy(odev->si_name, dev->si_name, sizeof(dev->si_name));
 330         return (dev);
 331 }
 332
 333 cdev_t
 334 get_dev(int x, int y)
 335 {
 336         cdev_t dev;
 337         struct dev_ops *ops;
 338
 339         if (x == NOUDEV)
 340                 return(NULL);
 341         ops = dev_ops_get(x, y);
 342         if (ops == NULL)
 343                 return(NULL);
 344         dev = hashdev(ops, x, y, TRUE);
 345         return(dev);
 346 }
 347
 348 /*
 349  * destroy_dev() removes the adhoc association for a device and revectors
 350  * its ops to &dead_dev_ops.
 351  *
 352  * This routine releases the reference count associated with the ADHOC
 353  * entry, plus releases the reference count held by the caller.  What this
 354  * means is that you should not call destroy_dev(make_dev(...)), because
 355  * make_dev() does not bump the reference count (beyond what it needs to
 356  * create the ad-hoc association).  Any procedure that intends to destroy
 357  * a device must have its own reference to it first.
 358  */
 359 void
 360 destroy_dev(cdev_t dev)
 361 {
 362         int hash;
 363
 364         if (dev == NULL)
 365                 return;
 366         if ((dev->si_flags & SI_ADHOC) == 0) {
 367                 release_dev(dev);
 368                 return;
 369         }
 370         if (dev_ref_debug) {
 371                 kprintf("destroy   dev %p %s(minor=%08x) refs=%d\n",
 372                         dev, devtoname(dev), dev->si_uminor,
 373                         dev->si_sysref.refcnt);
 374         }
 375         if (dev->si_sysref.refcnt < 2) {
 376                 kprintf("destroy_dev(): too few references on device! "
 377                         "%p %s(minor=%08x) refs=%d\n",
 378                     dev, devtoname(dev), dev->si_uminor,
 379                     dev->si_sysref.refcnt);
 380         }
 381         dev->si_flags &= ~SI_ADHOC;
 382         if (dev->si_flags & SI_HASHED) {
 383                 hash = __devthash(dev->si_umajor, dev->si_uminor);
 384                 LIST_REMOVE(dev, si_hash);
 385                 dev->si_flags &= ~SI_HASHED;
 386         }
 387
 388         /*
 389          * We have to release the ops reference before we replace the
 390          * device switch with dead_dev_ops.
 391          */
 392         if (dead_dev_ops.d_strategy == NULL)
 393                 compile_dev_ops(&dead_dev_ops);
 394         if (dev->si_ops && dev->si_ops != &dead_dev_ops)
 395                 dev_ops_release(dev->si_ops);
 396         dev->si_drv1 = NULL;
 397         dev->si_drv2 = NULL;
 398         dev->si_ops = &dead_dev_ops;
 399         sysref_put(&dev->si_sysref);    /* release adhoc association */
 400         release_dev(dev);               /* release callers reference */
 401 }
 402
 403 /*
 404  * Destroy all ad-hoc device associations associated with a domain within a
 405  * device switch.  Only the minor numbers are included in the mask/match
 406  * values.
 407  *
 408  * Unlike the ops functions whos link structures do not contain
 409  * any major bits, this function scans through the dev list via
 410  * si_umajor/si_uminor.
 411  *
 412  * The caller must not include any major bits in the match value.
 413  */
 414 void
 415 destroy_all_devs(struct dev_ops *ops, u_int mask, u_int match)
 416 {
 417         int i;
 418         cdev_t dev;
 419         cdev_t ndev;
 420
 421         for (i = 0; i < DEVT_HASH; ++i) {
 422                 ndev = LIST_FIRST(&dev_hash[i]);
 423                 while ((dev = ndev) != NULL) {
 424                     ndev = LIST_NEXT(dev, si_hash);
 425                     if (dev->si_ops == ops &&
 426                         ((u_int)dev->si_uminor & mask) == match
 427                     ) {
 428                         KKASSERT(dev->si_flags & SI_ADHOC);
 429                         reference_dev(dev);
 430                         destroy_dev(dev);
 431                     }
 432                 }
 433         }
 434 }
 435
 436 /*
 437  * Add a reference to a device.  Callers generally add their own references
 438  * when they are going to store a device node in a variable for long periods
 439  * of time, to prevent a disassociation from free()ing the node.
 440  *
 441  * Also note that a caller that intends to call destroy_dev() must first
 442  * obtain a reference on the device.  The ad-hoc reference you get with
 443  * make_dev() and friends is NOT sufficient to be able to call destroy_dev().
 444  */
 445 cdev_t
 446 reference_dev(cdev_t dev)
 447 {
 448         if (dev != NULL) {
 449                 sysref_get(&dev->si_sysref);
 450                 if (dev_ref_debug) {
 451                         kprintf("reference dev %p %s(minor=%08x) refs=%d\n",
 452                             dev, devtoname(dev), dev->si_uminor,
 453                             dev->si_sysref.refcnt);
 454                 }
 455         }
 456         return(dev);
 457 }
 458
 459 /*
 460  * release a reference on a device.  The device will be terminated when the
 461  * last reference has been released.
 462  *
 463  * NOTE: we must use si_umajor to figure out the original major number,
 464  * because si_ops could already be pointing at dead_dev_ops.
 465  */
 466 void
 467 release_dev(cdev_t dev)
 468 {
 469         if (dev == NULL)
 470                 return;
 471         sysref_put(&dev->si_sysref);
 472 }
 473
 474 static
 475 void
 476 cdev_terminate(struct cdev *dev)
 477 {
 478         int messedup = 0;
 479
 480         if (dev_ref_debug) {
 481                 kprintf("release   dev %p %s(minor=%08x) refs=%d\n",
 482                         dev, devtoname(dev), dev->si_uminor,
 483                         dev->si_sysref.refcnt);
 484         }
 485         if (dev->si_flags & SI_ADHOC) {
 486                 kprintf("Warning: illegal final release on ADHOC"
 487                         " device %p(%s), the device was never"
 488                         " destroyed!\n",
 489                         dev, devtoname(dev));
 490                 messedup = 1;
 491         }
 492         if (dev->si_flags & SI_HASHED) {
 493                 kprintf("Warning: last release on device, no call"
 494                         " to destroy_dev() was made! dev %p(%s)\n",
 495                         dev, devtoname(dev));
 496                 reference_dev(dev);
 497                 destroy_dev(dev);
 498                 messedup = 1;
 499         }
 500         if (SLIST_FIRST(&dev->si_hlist) != NULL) {
 501                 kprintf("Warning: last release on device, vnode"
 502                         " associations still exist! dev %p(%s)\n",
 503                         dev, devtoname(dev));
 504                 messedup = 1;
 505         }
 506         if (dev->si_ops && dev->si_ops != &dead_dev_ops) {
 507                 dev_ops_release(dev->si_ops);
 508                 dev->si_ops = NULL;
 509         }
 510         if (messedup == 0)
 511                 sysref_put(&dev->si_sysref);
 512 }
 513
 514 const char *
 515 devtoname(cdev_t dev)
 516 {
 517         int mynor;
 518         int len;
 519         char *p;
 520         const char *dname;
 521
 522         if (dev == NULL)
 523                 return("#nodev");
 524         if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
 525                 p = dev->si_name;
 526                 len = sizeof(dev->si_name);
 527                 if ((dname = dev_dname(dev)) != NULL)
 528                         ksnprintf(p, len, "#%s/", dname);
 529                 else
 530                         ksnprintf(p, len, "#%d/", major(dev));
 531                 len -= strlen(p);
 532                 p += strlen(p);
 533                 mynor = minor(dev);
 534                 if (mynor < 0 || mynor > 255)
 535                         ksnprintf(p, len, "%#x", (u_int)mynor);
 536                 else
 537                         ksnprintf(p, len, "%d", mynor);
 538         }
 539         return (dev->si_name);
 540 }
 541