sys/vfs/devfs/devfs_core.c

   1 /*
   2  * Copyright (c) 2009-2019 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Alex Hornung <ahornung@gmail.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34 #include <sys/param.h>
  35 #include <sys/systm.h>
  36 #include <sys/kernel.h>
  37 #include <sys/bus.h>
  38 #include <sys/malloc.h>
  39 #include <sys/mount.h>
  40 #include <sys/vnode.h>
  41 #include <sys/lock.h>
  42 #include <sys/file.h>
  43 #include <sys/msgport.h>
  44 #include <sys/sysctl.h>
  45 #include <sys/ucred.h>
  46 #include <sys/devfs.h>
  47 #include <sys/devfs_rules.h>
  48 #include <sys/udev.h>
  49
  50 #include <sys/msgport2.h>
  51 #include <sys/spinlock2.h>
  52 #include <sys/sysref2.h>
  53
  54 MALLOC_DEFINE(M_DEVFS, "devfs", "Device File System (devfs) allocations");
  55 DEVFS_DEFINE_CLONE_BITMAP(ops_id);
  56 /*
  57  * SYSREF Integration - reference counting, allocation,
  58  * sysid and syslink integration.
  59  */
  60 static void devfs_cdev_terminate(cdev_t dev);
  61 static void devfs_cdev_lock(cdev_t dev);
  62 static void devfs_cdev_unlock(cdev_t dev);
  63 static struct sysref_class     cdev_sysref_class = {
  64         .name =         "cdev",
  65         .mtype =        M_DEVFS,
  66         .proto =        SYSREF_PROTO_DEV,
  67         .offset =       offsetof(struct cdev, si_sysref),
  68         .objsize =      sizeof(struct cdev),
  69         .nom_cache =    32,
  70         .flags =        0,
  71         .ops =  {
  72                 .terminate = (sysref_terminate_func_t)devfs_cdev_terminate,
  73                 .lock = (sysref_lock_func_t)devfs_cdev_lock,
  74                 .unlock = (sysref_unlock_func_t)devfs_cdev_unlock
  75         }
  76 };
  77
  78 static struct objcache  *devfs_node_cache;
  79 static struct objcache  *devfs_msg_cache;
  80 static struct objcache  *devfs_dev_cache;
  81
  82 static struct objcache_malloc_args devfs_node_malloc_args = {
  83         sizeof(struct devfs_node), M_DEVFS };
  84 struct objcache_malloc_args devfs_msg_malloc_args = {
  85         sizeof(struct devfs_msg), M_DEVFS };
  86 struct objcache_malloc_args devfs_dev_malloc_args = {
  87         sizeof(struct cdev), M_DEVFS };
  88
  89 static struct devfs_dev_head devfs_dev_list =
  90                 TAILQ_HEAD_INITIALIZER(devfs_dev_list);
  91 static struct devfs_mnt_head devfs_mnt_list =
  92                 TAILQ_HEAD_INITIALIZER(devfs_mnt_list);
  93 static struct devfs_chandler_head devfs_chandler_list =
  94                 TAILQ_HEAD_INITIALIZER(devfs_chandler_list);
  95 static struct devfs_alias_head devfs_alias_list =
  96                 TAILQ_HEAD_INITIALIZER(devfs_alias_list);
  97 static struct devfs_dev_ops_head devfs_dev_ops_list =
  98                 TAILQ_HEAD_INITIALIZER(devfs_dev_ops_list);
  99
 100 struct lock             devfs_lock;
 101 struct lwkt_token       devfs_token;
 102 static struct lwkt_port devfs_dispose_port;
 103 static struct lwkt_port devfs_msg_port;
 104 static struct thread    *td_core;
 105
 106 static struct spinlock  ino_lock;
 107 static ino_t    d_ino;
 108 static int      devfs_debug_enable;
 109 static int      devfs_run;
 110
 111 static ino_t devfs_fetch_ino(void);
 112 static int devfs_reference_ops(struct dev_ops *ops);
 113 static void devfs_release_ops(struct dev_ops *ops);
 114 static int devfs_create_all_dev_worker(struct devfs_node *);
 115 static int devfs_create_dev_worker(cdev_t, uid_t, gid_t, int);
 116 static int devfs_destroy_dev_worker(cdev_t);
 117 static int devfs_destroy_related_worker(cdev_t);
 118 static int devfs_destroy_dev_by_ops_worker(struct dev_ops *, int);
 119 static int devfs_propagate_dev(cdev_t, int);
 120 static int devfs_unlink_dev(cdev_t dev);
 121 static void devfs_msg_exec(devfs_msg_t msg);
 122
 123 static int devfs_chandler_add_worker(const char *, d_clone_t *);
 124 static int devfs_chandler_del_worker(const char *);
 125
 126 static void devfs_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
 127 static void devfs_msg_core(void *);
 128
 129 static int devfs_find_device_by_name_worker(devfs_msg_t);
 130 static int devfs_find_device_by_devid_worker(devfs_msg_t);
 131
 132 static int devfs_apply_reset_rules_caller(char *, int);
 133
 134 static int devfs_scan_callback_worker(devfs_scan_t *, void *);
 135
 136 static struct devfs_node *devfs_resolve_or_create_dir(struct devfs_node *,
 137                 char *, size_t, int);
 138
 139 static int devfs_make_alias_worker(struct devfs_alias *);
 140 static int devfs_destroy_alias_worker(struct devfs_alias *);
 141 static int devfs_alias_remove(cdev_t);
 142 static int devfs_alias_reap(void);
 143 static int devfs_alias_propagate(struct devfs_alias *, int);
 144 static int devfs_alias_apply(struct devfs_node *, struct devfs_alias *);
 145 static int devfs_alias_check_create(struct devfs_node *);
 146
 147 static int devfs_clr_related_flag_worker(cdev_t, uint32_t);
 148 static int devfs_destroy_related_without_flag_worker(cdev_t, uint32_t);
 149
 150 static void *devfs_reaperp_callback(struct devfs_node *, void *);
 151 static void devfs_iterate_orphans_unmount(struct mount *mp);
 152 static void *devfs_gc_dirs_callback(struct devfs_node *, void *);
 153 static void *devfs_gc_links_callback(struct devfs_node *, struct devfs_node *);
 154 static void *
 155 devfs_inode_to_vnode_worker_callback(struct devfs_node *, ino_t *);
 156
 157 /*
 158  * devfs_debug() is a SYSCTL and TUNABLE controlled debug output function
 159  * using kvprintf
 160  */
 161 int
 162 devfs_debug(int level, char *fmt, ...)
 163 {
 164         __va_list ap;
 165
 166         __va_start(ap, fmt);
 167         if (level <= devfs_debug_enable)
 168                 kvprintf(fmt, ap);
 169         __va_end(ap);
 170
 171         return 0;
 172 }
 173
 174 /*
 175  * devfs_allocp() Allocates a new devfs node with the specified
 176  * parameters. The node is also automatically linked into the topology
 177  * if a parent is specified. It also calls the rule and alias stuff to
 178  * be applied on the new node
 179  */
 180 struct devfs_node *
 181 devfs_allocp(devfs_nodetype devfsnodetype, char *name,
 182              struct devfs_node *parent, struct mount *mp, cdev_t dev)
 183 {
 184         struct devfs_node *node = NULL;
 185         size_t namlen = strlen(name);
 186
 187         node = objcache_get(devfs_node_cache, M_WAITOK);
 188         bzero(node, sizeof(*node));
 189
 190         atomic_add_long(&DEVFS_MNTDATA(mp)->leak_count, 1);
 191
 192         node->d_dev = NULL;
 193         node->nchildren = 1;
 194         node->mp = mp;
 195         node->d_dir.d_ino = devfs_fetch_ino();
 196
 197         /*
 198          * Cookie jar for children. Leave 0 and 1 for '.' and '..' entries
 199          * respectively.
 200          */
 201         node->cookie_jar = 2;
 202
 203         /*
 204          * Access Control members
 205          */
 206         node->mode = DEVFS_DEFAULT_MODE;
 207         node->uid = DEVFS_DEFAULT_UID;
 208         node->gid = DEVFS_DEFAULT_GID;
 209
 210         switch (devfsnodetype) {
 211         case Nroot:
 212                 /*
 213                  * Ensure that we don't recycle the root vnode by marking it as
 214                  * linked into the topology.
 215                  */
 216                 node->flags |= DEVFS_NODE_LINKED;
 217         case Ndir:
 218                 TAILQ_INIT(DEVFS_DENODE_HEAD(node));
 219                 node->d_dir.d_type = DT_DIR;
 220                 node->nchildren = 2;
 221                 break;
 222
 223         case Nlink:
 224                 node->d_dir.d_type = DT_LNK;
 225                 break;
 226
 227         case Nreg:
 228                 node->d_dir.d_type = DT_REG;
 229                 break;
 230
 231         case Ndev:
 232                 if (dev != NULL) {
 233                         node->d_dir.d_type = DT_CHR;
 234                         node->d_dev = dev;
 235
 236                         node->mode = dev->si_perms;
 237                         node->uid = dev->si_uid;
 238                         node->gid = dev->si_gid;
 239
 240                         devfs_alias_check_create(node);
 241                 }
 242                 break;
 243
 244         default:
 245                 panic("devfs_allocp: unknown node type");
 246         }
 247
 248         node->v_node = NULL;
 249         node->node_type = devfsnodetype;
 250
 251         /* Initialize the dirent structure of each devfs vnode */
 252         node->d_dir.d_namlen = namlen;
 253         node->d_dir.d_name = kmalloc(namlen+1, M_DEVFS, M_WAITOK);
 254         memcpy(node->d_dir.d_name, name, namlen);
 255         node->d_dir.d_name[namlen] = '\0';
 256
 257         /* Initialize the parent node element */
 258         node->parent = parent;
 259
 260         /* Initialize *time members */
 261         vfs_timestamp(&node->atime);
 262         node->mtime = node->ctime = node->atime;
 263
 264         /*
 265          * Associate with parent as last step, clean out namecache
 266          * reference.
 267          */
 268         if (parent) {
 269                 if (parent->node_type == Nroot ||
 270                     parent->node_type == Ndir) {
 271                         parent->nchildren++;
 272                         node->cookie = parent->cookie_jar++;
 273                         node->flags |= DEVFS_NODE_LINKED;
 274                         TAILQ_INSERT_TAIL(DEVFS_DENODE_HEAD(parent), node, link);
 275
 276                         /* This forces negative namecache lookups to clear */
 277                         ++mp->mnt_namecache_gen;
 278                 } else {
 279                         kprintf("devfs: Cannot link node %p (%s) "
 280                                 "into %p (%s)\n",
 281                                 node, node->d_dir.d_name,
 282                                 parent, parent->d_dir.d_name);
 283                         print_backtrace(-1);
 284                 }
 285         }
 286
 287         /*
 288          * Apply rules (requires root node, skip if we are creating the root
 289          * node)
 290          */
 291         if (DEVFS_MNTDATA(mp)->root_node)
 292                 devfs_rule_check_apply(node, NULL);
 293
 294         atomic_add_long(&DEVFS_MNTDATA(mp)->file_count, 1);
 295
 296         return node;
 297 }
 298
 299 /*
 300  * devfs_allocv() allocates a new vnode based on a devfs node.
 301  */
 302 int
 303 devfs_allocv(struct vnode **vpp, struct devfs_node *node)
 304 {
 305         struct vnode *vp;
 306         int error = 0;
 307
 308         KKASSERT(node);
 309
 310         /*
 311          * devfs master lock must not be held across a vget() call, we have
 312          * to hold our ad-hoc vp to avoid a free race from destroying the
 313          * contents of the structure.  The vget() will interlock recycles
 314          * for us.
 315          */
 316 try_again:
 317         while ((vp = node->v_node) != NULL) {
 318                 vhold(vp);
 319                 lockmgr(&devfs_lock, LK_RELEASE);
 320                 error = vget(vp, LK_EXCLUSIVE);
 321                 vdrop(vp);
 322                 lockmgr(&devfs_lock, LK_EXCLUSIVE);
 323                 if (error == 0) {
 324                         *vpp = vp;
 325                         goto out;
 326                 }
 327                 if (error != ENOENT) {
 328                         *vpp = NULL;
 329                         goto out;
 330                 }
 331         }
 332
 333         /*
 334          * devfs master lock must not be held across a getnewvnode() call.
 335          */
 336         lockmgr(&devfs_lock, LK_RELEASE);
 337         if ((error = getnewvnode(VT_DEVFS, node->mp, vpp, 0, 0)) != 0) {
 338                 lockmgr(&devfs_lock, LK_EXCLUSIVE);
 339                 goto out;
 340         }
 341         lockmgr(&devfs_lock, LK_EXCLUSIVE);
 342
 343         vp = *vpp;
 344
 345         if (node->v_node != NULL) {
 346                 vp->v_type = VBAD;
 347                 vx_put(vp);
 348                 goto try_again;
 349         }
 350
 351         vp->v_data = node;
 352         node->v_node = vp;
 353
 354         switch (node->node_type) {
 355         case Nroot:
 356                 vsetflags(vp, VROOT);
 357                 /* fall through */
 358         case Ndir:
 359                 vp->v_type = VDIR;
 360                 break;
 361
 362         case Nlink:
 363                 vp->v_type = VLNK;
 364                 break;
 365
 366         case Nreg:
 367                 vp->v_type = VREG;
 368                 break;
 369
 370         case Ndev:
 371                 vp->v_type = VCHR;
 372                 KKASSERT(node->d_dev);
 373
 374                 vp->v_uminor = node->d_dev->si_uminor;
 375                 vp->v_umajor = node->d_dev->si_umajor;
 376
 377                 v_associate_rdev(vp, node->d_dev);
 378                 vp->v_ops = &node->mp->mnt_vn_spec_ops;
 379                 if (node->d_dev->si_ops->head.flags & D_KVABIO)
 380                         vsetflags(vp, VKVABIO);
 381                 break;
 382
 383         default:
 384                 panic("devfs_allocv: unknown node type");
 385         }
 386         vx_downgrade(vp);       /* downgrade VX lock to VN lock */
 387
 388 out:
 389         return error;
 390 }
 391
 392 /*
 393  * devfs_allocvp allocates both a devfs node (with the given settings) and a vnode
 394  * based on the newly created devfs node.
 395  */
 396 int
 397 devfs_allocvp(struct mount *mp, struct vnode **vpp, devfs_nodetype devfsnodetype,
 398                 char *name, struct devfs_node *parent, cdev_t dev)
 399 {
 400         struct devfs_node *node;
 401
 402         node = devfs_allocp(devfsnodetype, name, parent, mp, dev);
 403
 404         if (node != NULL)
 405                 devfs_allocv(vpp, node);
 406         else
 407                 *vpp = NULL;
 408
 409         return 0;
 410 }
 411
 412 /*
 413  * Destroy the devfs_node.  The node must be unlinked from the topology.
 414  *
 415  * This function will also destroy any vnode association with the node
 416  * and device.
 417  *
 418  * The cdev_t itself remains intact.
 419  *
 420  * The core lock is not necessarily held on call and must be temporarily
 421  * released if it is to avoid a deadlock.
 422  */
 423 void
 424 devfs_freep(struct devfs_node *node)
 425 {
 426         struct vnode *vp;
 427         int maxloops;
 428
 429         KKASSERT(node);
 430
 431         /*
 432          * It is possible for devfs_freep() to race a destruction due
 433          * to having to release the lock below.  We use DEVFS_DESTROYED
 434          * to interlock the race (mediated by devfs_lock)
 435          *
 436          * We use NLINKSWAIT to indicate that the node couldn't be
 437          * freed due to having pending nlinks.  We can free
 438          * the node when nlinks drops to 0.  This should never print
 439          * a "(null)" name, if it ever does there are still unresolved
 440          * issues.
 441          */
 442         if (node->flags & DEVFS_DESTROYED) {
 443                 if ((node->flags & DEVFS_NLINKSWAIT) &&
 444                     node->nlinks == 0) {
 445                         kprintf("devfs: final node '%s' on nlinks\n",
 446                                 node->d_dir.d_name);
 447                         if (node->d_dir.d_name) {
 448                                 kfree(node->d_dir.d_name, M_DEVFS);
 449                                 node->d_dir.d_name = NULL;
 450                         }
 451                         objcache_put(devfs_node_cache, node);
 452                 } else {
 453                         kprintf("devfs: race avoided node '%s' (%p)\n",
 454                                 node->d_dir.d_name, node);
 455                 }
 456                 return;
 457         }
 458         node->flags |= DEVFS_DESTROYED;
 459
 460         /*
 461          * Items we have to dispose of before potentially releasing
 462          * devfs_lock.
 463          *
 464          * Remove the node from the orphan list if it is still on it.
 465          */
 466         atomic_subtract_long(&DEVFS_MNTDATA(node->mp)->leak_count, 1);
 467         atomic_subtract_long(&DEVFS_MNTDATA(node->mp)->file_count, 1);
 468         if (node->flags & DEVFS_ORPHANED)
 469                 devfs_tracer_del_orphan(node);
 470
 471         /*
 472          * At this point only the vp points to node, and node cannot be
 473          * physically freed because we own DEVFS_DESTROYED.
 474          *
 475          * We must dispose of the vnode without deadlocking or racing
 476          * against e.g. a vnode reclaim.
 477          *
 478          * This also prevents the vnode reclaim code from double-freeing
 479          * the node.  The vget() is required to safely modified the vp
 480          * and cycle the refs to terminate an inactive vp.
 481          */
 482         maxloops = 1000;
 483         while ((vp = node->v_node) != NULL) {
 484                 int relock;
 485
 486                 vhold(vp);
 487                 if (lockstatus(&devfs_lock, curthread) == LK_EXCLUSIVE) {
 488                         lockmgr(&devfs_lock, LK_RELEASE);
 489                         relock = 1;
 490                 } else {
 491                         relock = 0;
 492                 }
 493                 if (node->v_node == NULL) {
 494                         /* reclaim race, mediated by devfs_lock */
 495                         vdrop(vp);
 496                 } else if (vget(vp, LK_EXCLUSIVE | LK_RETRY) == 0) {
 497                         vdrop(vp);
 498                         v_release_rdev(vp);
 499                         vp->v_data = NULL;
 500                         node->v_node = NULL;
 501                         vput(vp);
 502                 } else {
 503                         /* reclaim race, mediated by devfs_lock */
 504                         vdrop(vp);
 505                 }
 506                 if (relock)
 507                         lockmgr(&devfs_lock, LK_EXCLUSIVE);
 508                 if (--maxloops == 0) {
 509                         kprintf("devfs_freep: livelock on node %p\n", node);
 510                         break;
 511                 }
 512         }
 513
 514         /*
 515          * Remaining cleanup
 516          */
 517         if (node->symlink_name) {
 518                 kfree(node->symlink_name, M_DEVFS);
 519                 node->symlink_name = NULL;
 520         }
 521
 522         /*
 523          * We cannot actually free the node if it still has
 524          * nlinks.
 525          */
 526         if (node->nlinks) {
 527                 node->flags |= DEVFS_NLINKSWAIT;
 528         } else {
 529                 if (node->d_dir.d_name) {
 530                         kfree(node->d_dir.d_name, M_DEVFS);
 531                         node->d_dir.d_name = NULL;
 532                 }
 533                 objcache_put(devfs_node_cache, node);
 534         }
 535 }
 536
 537 /*
 538  * Returns a valid vp associated with the devfs alias node or NULL
 539  */
 540 static void *devfs_alias_getvp(struct devfs_node *node)
 541 {
 542         struct devfs_node *found = node;
 543         int depth = 0;
 544
 545         while ((found->node_type == Nlink) && (found->link_target)) {
 546                 if (depth >= 8) {
 547                         devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
 548                         break;
 549                 }
 550
 551                 found = found->link_target;
 552                 ++depth;
 553         }
 554
 555         return found->v_node;
 556 }
 557
 558 /*
 559  * Unlink the devfs node from the topology and add it to the orphan list.
 560  * The node will later be destroyed by freep.
 561  *
 562  * Any vnode association, including the v_rdev and v_data, remains intact
 563  * until the freep.
 564  */
 565 void
 566 devfs_unlinkp(struct devfs_node *node)
 567 {
 568         struct devfs_node *parent;
 569         struct devfs_node *target;
 570         struct vnode *vp;
 571         KKASSERT(node);
 572
 573         /*
 574          * Add the node to the orphan list, so it is referenced somewhere, to
 575          * so we don't leak it.
 576          */
 577         devfs_tracer_add_orphan(node);
 578
 579         parent = node->parent;
 580         node->parent = NULL;
 581
 582         /*
 583          * If the parent is known we can unlink the node out of the topology
 584          */
 585         if (node->flags & DEVFS_NODE_LINKED) {
 586                 if (parent) {
 587                         TAILQ_REMOVE(DEVFS_DENODE_HEAD(parent), node, link);
 588                         parent->nchildren--;
 589                 } else if (node == DEVFS_MNTDATA(node->mp)->root_node) {
 590                         DEVFS_MNTDATA(node->mp)->root_node = NULL;
 591                 }
 592                 node->flags &= ~DEVFS_NODE_LINKED;
 593         }
 594
 595         /*
 596          * Namecache invalidation.
 597          *
 598          * devfs alias nodes are special: their v_node entry is always null
 599          * and they use the one from their link target.  We thus use the
 600          * target node's vp to invalidate both alias and target entries in
 601          * the namecache.
 602          *
 603          * Doing so for the target is not necessary but it would be more
 604          * expensive to resolve only the namecache entry of the alias node
 605          * from the information available in this function.
 606          *
 607          * WARNING! We do not disassociate the vnode here.  That can only
 608          *          be safely done in devfs_freep().
 609          */
 610         if (node->node_type == Nlink) {
 611                 if ((target = node->link_target) != NULL) {
 612                         vp = devfs_alias_getvp(node);
 613                         node->link_target = NULL;
 614                         target->nlinks--;
 615                         if (target->nlinks == 0 &&
 616                             (target->flags & DEVFS_DESTROYED)) {
 617                                 devfs_freep(target);
 618                         }
 619                 } else {
 620                         vp = NULL;
 621                 }
 622         } else {
 623                 vp = node->v_node;
 624         }
 625
 626         if (vp != NULL)
 627                 cache_inval_vp(vp, CINV_DESTROY);
 628 }
 629
 630 void *
 631 devfs_iterate_topology(struct devfs_node *node,
 632                 devfs_iterate_callback_t *callback, void *arg1)
 633 {
 634         struct devfs_node *node1, *node2;
 635         void *ret = NULL;
 636
 637         if (((node->node_type == Nroot) || (node->node_type == Ndir)) &&
 638             node->nchildren > 2) {
 639                 TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node),
 640                                       link, node2) {
 641                         ret = devfs_iterate_topology(node1, callback, arg1);
 642                         if (ret)
 643                                 return ret;
 644                 }
 645         }
 646         ret = callback(node, arg1);
 647
 648         return ret;
 649 }
 650
 651 static void *
 652 devfs_alias_reaper_callback(struct devfs_node *node, void *unused)
 653 {
 654         if (node->node_type == Nlink) {
 655                 devfs_unlinkp(node);
 656                 devfs_freep(node);
 657         }
 658
 659         return NULL;
 660 }
 661
 662 /*
 663  * devfs_reaperp() is a recursive function that iterates through all the
 664  * topology, unlinking and freeing all devfs nodes.
 665  */
 666 static void *
 667 devfs_reaperp_callback(struct devfs_node *node, void *unused)
 668 {
 669         devfs_unlinkp(node);
 670         devfs_freep(node);
 671
 672         return NULL;
 673 }
 674
 675 /*
 676  * Report any orphans that we couldn't delete.  The mp and mnt_data
 677  * are both disappearing, so we must also clean up the nodes a bit.
 678  */
 679 static void
 680 devfs_iterate_orphans_unmount(struct mount *mp)
 681 {
 682         struct devfs_orphan *orphan;
 683
 684         while ((orphan = TAILQ_FIRST(DEVFS_ORPHANLIST(mp))) != NULL) {
 685                 devfs_freep(orphan->node);
 686                 /* orphan stale */
 687         }
 688 }
 689
 690 static void *
 691 devfs_gc_dirs_callback(struct devfs_node *node, void *unused)
 692 {
 693         if (node->node_type == Ndir) {
 694                 if ((node->nchildren == 2) &&
 695                     !(node->flags & DEVFS_USER_CREATED)) {
 696                         devfs_unlinkp(node);
 697                         devfs_freep(node);
 698                 }
 699         }
 700
 701         return NULL;
 702 }
 703
 704 static void *
 705 devfs_gc_links_callback(struct devfs_node *node, struct devfs_node *target)
 706 {
 707         if ((node->node_type == Nlink) && (node->link_target == target)) {
 708                 devfs_unlinkp(node);
 709                 devfs_freep(node);
 710         }
 711
 712         return NULL;
 713 }
 714
 715 /*
 716  * devfs_gc() is devfs garbage collector. It takes care of unlinking and
 717  * freeing a node, but also removes empty directories and links that link
 718  * via devfs auto-link mechanism to the node being deleted.
 719  */
 720 int
 721 devfs_gc(struct devfs_node *node)
 722 {
 723         struct devfs_node *root_node = DEVFS_MNTDATA(node->mp)->root_node;
 724
 725         if (node->nlinks > 0)
 726                 devfs_iterate_topology(root_node,
 727                                 (devfs_iterate_callback_t *)devfs_gc_links_callback, node);
 728
 729         devfs_unlinkp(node);
 730         devfs_iterate_topology(root_node,
 731                         (devfs_iterate_callback_t *)devfs_gc_dirs_callback, NULL);
 732
 733         devfs_freep(node);
 734
 735         return 0;
 736 }
 737
 738 /*
 739  * devfs_create_dev() is the asynchronous entry point for device creation.
 740  * It just sends a message with the relevant details to the devfs core.
 741  *
 742  * This function will reference the passed device.  The reference is owned
 743  * by devfs and represents all of the device's node associations.
 744  */
 745 int
 746 devfs_create_dev(cdev_t dev, uid_t uid, gid_t gid, int perms)
 747 {
 748         reference_dev(dev);
 749         devfs_msg_send_dev(DEVFS_DEVICE_CREATE, dev, uid, gid, perms);
 750
 751         return 0;
 752 }
 753
 754 /*
 755  * devfs_destroy_dev() is the asynchronous entry point for device destruction.
 756  * It just sends a message with the relevant details to the devfs core.
 757  */
 758 int
 759 devfs_destroy_dev(cdev_t dev)
 760 {
 761         devfs_msg_send_dev(DEVFS_DEVICE_DESTROY, dev, 0, 0, 0);
 762         return 0;
 763 }
 764
 765 /*
 766  * devfs_mount_add() is the synchronous entry point for adding a new devfs
 767  * mount.  It sends a synchronous message with the relevant details to the
 768  * devfs core.
 769  */
 770 int
 771 devfs_mount_add(struct devfs_mnt_data *mnt)
 772 {
 773         devfs_msg_t msg;
 774
 775         msg = devfs_msg_get();
 776         msg->mdv_mnt = mnt;
 777         devfs_msg_send_sync(DEVFS_MOUNT_ADD, msg);
 778         devfs_msg_put(msg);
 779
 780         return 0;
 781 }
 782
 783 /*
 784  * devfs_mount_del() is the synchronous entry point for removing a devfs mount.
 785  * It sends a synchronous message with the relevant details to the devfs core.
 786  */
 787 int
 788 devfs_mount_del(struct devfs_mnt_data *mnt)
 789 {
 790         devfs_msg_t msg;
 791
 792         msg = devfs_msg_get();
 793         msg->mdv_mnt = mnt;
 794         devfs_msg_send_sync(DEVFS_MOUNT_DEL, msg);
 795         devfs_msg_put(msg);
 796
 797         return 0;
 798 }
 799
 800 /*
 801  * devfs_destroy_related() is the synchronous entry point for device
 802  * destruction by subname. It just sends a message with the relevant details to
 803  * the devfs core.
 804  */
 805 int
 806 devfs_destroy_related(cdev_t dev)
 807 {
 808         devfs_msg_t msg;
 809
 810         msg = devfs_msg_get();
 811         msg->mdv_load = dev;
 812         devfs_msg_send_sync(DEVFS_DESTROY_RELATED, msg);
 813         devfs_msg_put(msg);
 814         return 0;
 815 }
 816
 817 int
 818 devfs_clr_related_flag(cdev_t dev, uint32_t flag)
 819 {
 820         devfs_msg_t msg;
 821
 822         msg = devfs_msg_get();
 823         msg->mdv_flags.dev = dev;
 824         msg->mdv_flags.flag = flag;
 825         devfs_msg_send_sync(DEVFS_CLR_RELATED_FLAG, msg);
 826         devfs_msg_put(msg);
 827
 828         return 0;
 829 }
 830
 831 int
 832 devfs_destroy_related_without_flag(cdev_t dev, uint32_t flag)
 833 {
 834         devfs_msg_t msg;
 835
 836         msg = devfs_msg_get();
 837         msg->mdv_flags.dev = dev;
 838         msg->mdv_flags.flag = flag;
 839         devfs_msg_send_sync(DEVFS_DESTROY_RELATED_WO_FLAG, msg);
 840         devfs_msg_put(msg);
 841
 842         return 0;
 843 }
 844
 845 /*
 846  * devfs_create_all_dev is the asynchronous entry point to trigger device
 847  * node creation.  It just sends a message with the relevant details to
 848  * the devfs core.
 849  */
 850 int
 851 devfs_create_all_dev(struct devfs_node *root)
 852 {
 853         devfs_msg_send_generic(DEVFS_CREATE_ALL_DEV, root);
 854         return 0;
 855 }
 856
 857 /*
 858  * devfs_destroy_dev_by_ops is the asynchronous entry point to destroy all
 859  * devices with a specific set of dev_ops and minor.  It just sends a
 860  * message with the relevant details to the devfs core.
 861  */
 862 int
 863 devfs_destroy_dev_by_ops(struct dev_ops *ops, int minor)
 864 {
 865         devfs_msg_send_ops(DEVFS_DESTROY_DEV_BY_OPS, ops, minor);
 866         return 0;
 867 }
 868
 869 /*
 870  * devfs_clone_handler_add is the synchronous entry point to add a new
 871  * clone handler.  It just sends a message with the relevant details to
 872  * the devfs core.
 873  */
 874 int
 875 devfs_clone_handler_add(const char *name, d_clone_t *nhandler)
 876 {
 877         devfs_msg_t msg;
 878
 879         msg = devfs_msg_get();
 880         msg->mdv_chandler.name = name;
 881         msg->mdv_chandler.nhandler = nhandler;
 882         devfs_msg_send_sync(DEVFS_CHANDLER_ADD, msg);
 883         devfs_msg_put(msg);
 884         return 0;
 885 }
 886
 887 /*
 888  * devfs_clone_handler_del is the synchronous entry point to remove a
 889  * clone handler.  It just sends a message with the relevant details to
 890  * the devfs core.
 891  */
 892 int
 893 devfs_clone_handler_del(const char *name)
 894 {
 895         devfs_msg_t msg;
 896
 897         msg = devfs_msg_get();
 898         msg->mdv_chandler.name = name;
 899         msg->mdv_chandler.nhandler = NULL;
 900         devfs_msg_send_sync(DEVFS_CHANDLER_DEL, msg);
 901         devfs_msg_put(msg);
 902         return 0;
 903 }
 904
 905 /*
 906  * devfs_find_device_by_name is the synchronous entry point to find a
 907  * device given its name.  It sends a synchronous message with the
 908  * relevant details to the devfs core and returns the answer.
 909  */
 910 cdev_t
 911 devfs_find_device_by_name(const char *fmt, ...)
 912 {
 913         cdev_t found = NULL;
 914         devfs_msg_t msg;
 915         char *target;
 916         __va_list ap;
 917
 918         if (fmt == NULL)
 919                 return NULL;
 920
 921         __va_start(ap, fmt);
 922         kvasnprintf(&target, PATH_MAX, fmt, ap);
 923         __va_end(ap);
 924
 925         msg = devfs_msg_get();
 926         msg->mdv_name = target;
 927         devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_NAME, msg);
 928         found = msg->mdv_cdev;
 929         devfs_msg_put(msg);
 930         kvasfree(&target);
 931
 932         return found;
 933 }
 934
 935 /*
 936  * devfs_find_device_by_devid is the synchronous entry point to find a
 937  * device given its udev number.  It sends a synchronous message with
 938  * the relevant details to the devfs core and returns the answer.
 939  */
 940 cdev_t
 941 devfs_find_device_by_devid(dev_t udev)
 942 {
 943         cdev_t found = NULL;
 944         devfs_msg_t msg;
 945
 946         msg = devfs_msg_get();
 947         msg->mdv_udev = udev;
 948         devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_DEVID, msg);
 949         found = msg->mdv_cdev;
 950         devfs_msg_put(msg);
 951
 952         devfs_debug(DEVFS_DEBUG_DEBUG,
 953                     "devfs_find_device_by_devid found? %s  -end:3-\n",
 954                     ((found) ? found->si_name:"NO"));
 955         return found;
 956 }
 957
 958 struct vnode *
 959 devfs_inode_to_vnode(struct mount *mp, ino_t target)
 960 {
 961         struct vnode *vp = NULL;
 962         devfs_msg_t msg;
 963
 964         if (mp == NULL)
 965                 return NULL;
 966
 967         msg = devfs_msg_get();
 968         msg->mdv_ino.mp = mp;
 969         msg->mdv_ino.ino = target;
 970         devfs_msg_send_sync(DEVFS_INODE_TO_VNODE, msg);
 971         vp = msg->mdv_ino.vp;
 972         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 973         devfs_msg_put(msg);
 974
 975         return vp;
 976 }
 977
 978 /*
 979  * devfs_make_alias is the asynchronous entry point to register an alias
 980  * for a device.  It just sends a message with the relevant details to the
 981  * devfs core.
 982  */
 983 int
 984 devfs_make_alias(const char *name, cdev_t dev_target)
 985 {
 986         struct devfs_alias *alias;
 987         size_t len;
 988
 989         len = strlen(name);
 990
 991         alias = kmalloc(sizeof(struct devfs_alias), M_DEVFS, M_WAITOK);
 992         alias->name = kstrdup(name, M_DEVFS);
 993         alias->namlen = len;
 994         alias->dev_target = dev_target;
 995
 996         devfs_msg_send_generic(DEVFS_MAKE_ALIAS, alias);
 997         return 0;
 998 }
 999
1000 /*
1001  * devfs_destroy_alias is the asynchronous entry point to deregister an alias
1002  * for a device.  It just sends a message with the relevant details to the
1003  * devfs core.
1004  */
1005 int
1006 devfs_destroy_alias(const char *name, cdev_t dev_target)
1007 {
1008         struct devfs_alias *alias;
1009         size_t len;
1010
1011         len = strlen(name);
1012
1013         alias = kmalloc(sizeof(struct devfs_alias), M_DEVFS, M_WAITOK);
1014         alias->name = kstrdup(name, M_DEVFS);
1015         alias->namlen = len;
1016         alias->dev_target = dev_target;
1017
1018         devfs_msg_send_generic(DEVFS_DESTROY_ALIAS, alias);
1019         return 0;
1020 }
1021
1022 /*
1023  * devfs_apply_rules is the asynchronous entry point to trigger application
1024  * of all rules.  It just sends a message with the relevant details to the
1025  * devfs core.
1026  */
1027 int
1028 devfs_apply_rules(char *mntto)
1029 {
1030         char *new_name;
1031
1032         new_name = kstrdup(mntto, M_DEVFS);
1033         devfs_msg_send_name(DEVFS_APPLY_RULES, new_name);
1034
1035         return 0;
1036 }
1037
1038 /*
1039  * devfs_reset_rules is the asynchronous entry point to trigger reset of all
1040  * rules. It just sends a message with the relevant details to the devfs core.
1041  */
1042 int
1043 devfs_reset_rules(char *mntto)
1044 {
1045         char *new_name;
1046
1047         new_name = kstrdup(mntto, M_DEVFS);
1048         devfs_msg_send_name(DEVFS_RESET_RULES, new_name);
1049
1050         return 0;
1051 }
1052
1053
1054 /*
1055  * devfs_scan_callback is the asynchronous entry point to call a callback
1056  * on all cdevs.
1057  * It just sends a message with the relevant details to the devfs core.
1058  */
1059 int
1060 devfs_scan_callback(devfs_scan_t *callback, void *arg)
1061 {
1062         devfs_msg_t msg;
1063
1064         KKASSERT(callback);
1065
1066         msg = devfs_msg_get();
1067         msg->mdv_load = callback;
1068         msg->mdv_load2 = arg;
1069         devfs_msg_send_sync(DEVFS_SCAN_CALLBACK, msg);
1070         devfs_msg_put(msg);
1071
1072         return 0;
1073 }
1074
1075
1076 /*
1077  * Acts as a message drain. Any message that is replied to here gets destroyed
1078  * and the memory freed.
1079  */
1080 static void
1081 devfs_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
1082 {
1083         devfs_msg_put((devfs_msg_t)msg);
1084 }
1085
1086 /*
1087  * devfs_msg_get allocates a new devfs msg and returns it.
1088  */
1089 devfs_msg_t
1090 devfs_msg_get(void)
1091 {
1092         return objcache_get(devfs_msg_cache, M_WAITOK);
1093 }
1094
1095 /*
1096  * devfs_msg_put deallocates a given devfs msg.
1097  */
1098 int
1099 devfs_msg_put(devfs_msg_t msg)
1100 {
1101         objcache_put(devfs_msg_cache, msg);
1102         return 0;
1103 }
1104
1105 /*
1106  * devfs_msg_send is the generic asynchronous message sending facility
1107  * for devfs. By default the reply port is the automatic disposal port.
1108  *
1109  * If the current thread is the devfs_msg_port thread we execute the
1110  * operation synchronously.
1111  */
1112 void
1113 devfs_msg_send(uint32_t cmd, devfs_msg_t devfs_msg)
1114 {
1115         lwkt_port_t port = &devfs_msg_port;
1116
1117         lwkt_initmsg(&devfs_msg->hdr, &devfs_dispose_port, 0);
1118
1119         devfs_msg->hdr.u.ms_result = cmd;
1120
1121         if (port->mpu_td == curthread) {
1122                 devfs_msg_exec(devfs_msg);
1123                 lwkt_replymsg(&devfs_msg->hdr, 0);
1124         } else {
1125                 lwkt_sendmsg(port, (lwkt_msg_t)devfs_msg);
1126         }
1127 }
1128
1129 /*
1130  * devfs_msg_send_sync is the generic synchronous message sending
1131  * facility for devfs. It initializes a local reply port and waits
1132  * for the core's answer. The core will write the answer on the same
1133  * message which is sent back as reply. The caller still has a reference
1134  * to the message, so we don't need to return it.
1135  */
1136 int
1137 devfs_msg_send_sync(uint32_t cmd, devfs_msg_t devfs_msg)
1138 {
1139         struct lwkt_port rep_port;
1140         int     error;
1141         lwkt_port_t port = &devfs_msg_port;
1142
1143         lwkt_initport_thread(&rep_port, curthread);
1144         lwkt_initmsg(&devfs_msg->hdr, &rep_port, 0);
1145
1146         devfs_msg->hdr.u.ms_result = cmd;
1147
1148         error = lwkt_domsg(port, (lwkt_msg_t)devfs_msg, 0);
1149
1150         return error;
1151 }
1152
1153 /*
1154  * sends a message with a generic argument.
1155  */
1156 void
1157 devfs_msg_send_generic(uint32_t cmd, void *load)
1158 {
1159         devfs_msg_t devfs_msg = devfs_msg_get();
1160
1161         devfs_msg->mdv_load = load;
1162         devfs_msg_send(cmd, devfs_msg);
1163 }
1164
1165 /*
1166  * sends a message with a name argument.
1167  */
1168 void
1169 devfs_msg_send_name(uint32_t cmd, char *name)
1170 {
1171         devfs_msg_t devfs_msg = devfs_msg_get();
1172
1173         devfs_msg->mdv_name = name;
1174         devfs_msg_send(cmd, devfs_msg);
1175 }
1176
1177 /*
1178  * sends a message with a mount argument.
1179  */
1180 void
1181 devfs_msg_send_mount(uint32_t cmd, struct devfs_mnt_data *mnt)
1182 {
1183         devfs_msg_t devfs_msg = devfs_msg_get();
1184
1185         devfs_msg->mdv_mnt = mnt;
1186         devfs_msg_send(cmd, devfs_msg);
1187 }
1188
1189 /*
1190  * sends a message with an ops argument.
1191  */
1192 void
1193 devfs_msg_send_ops(uint32_t cmd, struct dev_ops *ops, int minor)
1194 {
1195         devfs_msg_t devfs_msg = devfs_msg_get();
1196
1197         devfs_msg->mdv_ops.ops = ops;
1198         devfs_msg->mdv_ops.minor = minor;
1199         devfs_msg_send(cmd, devfs_msg);
1200 }
1201
1202 /*
1203  * sends a message with a clone handler argument.
1204  */
1205 void
1206 devfs_msg_send_chandler(uint32_t cmd, char *name, d_clone_t handler)
1207 {
1208         devfs_msg_t devfs_msg = devfs_msg_get();
1209
1210         devfs_msg->mdv_chandler.name = name;
1211         devfs_msg->mdv_chandler.nhandler = handler;
1212         devfs_msg_send(cmd, devfs_msg);
1213 }
1214
1215 /*
1216  * sends a message with a device argument.
1217  */
1218 void
1219 devfs_msg_send_dev(uint32_t cmd, cdev_t dev, uid_t uid, gid_t gid, int perms)
1220 {
1221         devfs_msg_t devfs_msg = devfs_msg_get();
1222
1223         devfs_msg->mdv_dev.dev = dev;
1224         devfs_msg->mdv_dev.uid = uid;
1225         devfs_msg->mdv_dev.gid = gid;
1226         devfs_msg->mdv_dev.perms = perms;
1227
1228         devfs_msg_send(cmd, devfs_msg);
1229 }
1230
1231 /*
1232  * sends a message with a link argument.
1233  */
1234 void
1235 devfs_msg_send_link(uint32_t cmd, char *name, char *target, struct mount *mp)
1236 {
1237         devfs_msg_t devfs_msg = devfs_msg_get();
1238
1239         devfs_msg->mdv_link.name = name;
1240         devfs_msg->mdv_link.target = target;
1241         devfs_msg->mdv_link.mp = mp;
1242         devfs_msg_send(cmd, devfs_msg);
1243 }
1244
1245 /*
1246  * devfs_msg_core is the main devfs thread. It handles all incoming messages
1247  * and calls the relevant worker functions. By using messages it's assured
1248  * that events occur in the correct order.
1249  */
1250 static void
1251 devfs_msg_core(void *arg)
1252 {
1253         devfs_msg_t msg;
1254
1255         lwkt_initport_thread(&devfs_msg_port, curthread);
1256
1257         lockmgr(&devfs_lock, LK_EXCLUSIVE);
1258         devfs_run = 1;
1259         wakeup(td_core);
1260         lockmgr(&devfs_lock, LK_RELEASE);
1261
1262         lwkt_gettoken(&devfs_token);
1263
1264         while (devfs_run) {
1265                 msg = (devfs_msg_t)lwkt_waitport(&devfs_msg_port, 0);
1266                 devfs_debug(DEVFS_DEBUG_DEBUG,
1267                                 "devfs_msg_core, new msg: %x\n",
1268                                 (unsigned int)msg->hdr.u.ms_result);
1269                 devfs_msg_exec(msg);
1270                 lwkt_replymsg(&msg->hdr, 0);
1271         }
1272
1273         lwkt_reltoken(&devfs_token);
1274         wakeup(td_core);
1275
1276         lwkt_exit();
1277 }
1278
1279 static void
1280 devfs_msg_exec(devfs_msg_t msg)
1281 {
1282         struct devfs_mnt_data *mnt;
1283         struct devfs_node *node;
1284         cdev_t  dev;
1285
1286         /*
1287          * Acquire the devfs lock to ensure safety of all called functions
1288          */
1289         lockmgr(&devfs_lock, LK_EXCLUSIVE);
1290
1291         switch (msg->hdr.u.ms_result) {
1292         case DEVFS_DEVICE_CREATE:
1293                 dev = msg->mdv_dev.dev;
1294                 devfs_create_dev_worker(dev,
1295                                         msg->mdv_dev.uid,
1296                                         msg->mdv_dev.gid,
1297                                         msg->mdv_dev.perms);
1298                 break;
1299         case DEVFS_DEVICE_DESTROY:
1300                 dev = msg->mdv_dev.dev;
1301                 devfs_destroy_dev_worker(dev);
1302                 break;
1303         case DEVFS_DESTROY_RELATED:
1304                 devfs_destroy_related_worker(msg->mdv_load);
1305                 break;
1306         case DEVFS_DESTROY_DEV_BY_OPS:
1307                 devfs_destroy_dev_by_ops_worker(msg->mdv_ops.ops,
1308                                                 msg->mdv_ops.minor);
1309                 break;
1310         case DEVFS_CREATE_ALL_DEV:
1311                 node = (struct devfs_node *)msg->mdv_load;
1312                 devfs_create_all_dev_worker(node);
1313                 break;
1314         case DEVFS_MOUNT_ADD:
1315                 mnt = msg->mdv_mnt;
1316                 TAILQ_INSERT_TAIL(&devfs_mnt_list, mnt, link);
1317                 devfs_create_all_dev_worker(mnt->root_node);
1318                 break;
1319         case DEVFS_MOUNT_DEL:
1320                 mnt = msg->mdv_mnt;
1321                 TAILQ_REMOVE(&devfs_mnt_list, mnt, link);
1322                 /* Be sure to remove all the aliases first */
1323                 devfs_iterate_topology(mnt->root_node,
1324                                        devfs_alias_reaper_callback,
1325                                        NULL);
1326                 devfs_iterate_topology(mnt->root_node,
1327                                        devfs_reaperp_callback,
1328                                        NULL);
1329                 devfs_iterate_orphans_unmount(mnt->mp);
1330                 if (mnt->leak_count) {
1331                         devfs_debug(DEVFS_DEBUG_SHOW,
1332                                     "Leaked %ld devfs_node elements!\n",
1333                                     mnt->leak_count);
1334                 }
1335                 break;
1336         case DEVFS_CHANDLER_ADD:
1337                 devfs_chandler_add_worker(msg->mdv_chandler.name,
1338                                 msg->mdv_chandler.nhandler);
1339                 break;
1340         case DEVFS_CHANDLER_DEL:
1341                 devfs_chandler_del_worker(msg->mdv_chandler.name);
1342                 break;
1343         case DEVFS_FIND_DEVICE_BY_NAME:
1344                 devfs_find_device_by_name_worker(msg);
1345                 break;
1346         case DEVFS_FIND_DEVICE_BY_DEVID:
1347                 devfs_find_device_by_devid_worker(msg);
1348                 break;
1349         case DEVFS_MAKE_ALIAS:
1350                 devfs_make_alias_worker((struct devfs_alias *)msg->mdv_load);
1351                 break;
1352         case DEVFS_DESTROY_ALIAS:
1353                 devfs_destroy_alias_worker((struct devfs_alias *)msg->mdv_load);
1354                 break;
1355         case DEVFS_APPLY_RULES:
1356                 devfs_apply_reset_rules_caller(msg->mdv_name, 1);
1357                 break;
1358         case DEVFS_RESET_RULES:
1359                 devfs_apply_reset_rules_caller(msg->mdv_name, 0);
1360                 break;
1361         case DEVFS_SCAN_CALLBACK:
1362                 devfs_scan_callback_worker((devfs_scan_t *)msg->mdv_load,
1363                         msg->mdv_load2);
1364                 break;
1365         case DEVFS_CLR_RELATED_FLAG:
1366                 devfs_clr_related_flag_worker(msg->mdv_flags.dev,
1367                                 msg->mdv_flags.flag);
1368                 break;
1369         case DEVFS_DESTROY_RELATED_WO_FLAG:
1370                 devfs_destroy_related_without_flag_worker(msg->mdv_flags.dev,
1371                                 msg->mdv_flags.flag);
1372                 break;
1373         case DEVFS_INODE_TO_VNODE:
1374                 msg->mdv_ino.vp = devfs_iterate_topology(
1375                         DEVFS_MNTDATA(msg->mdv_ino.mp)->root_node,
1376                         (devfs_iterate_callback_t *)devfs_inode_to_vnode_worker_callback,
1377                         &msg->mdv_ino.ino);
1378                 break;
1379         case DEVFS_TERMINATE_CORE:
1380                 devfs_run = 0;
1381                 break;
1382         case DEVFS_SYNC:
1383                 break;
1384         default:
1385                 devfs_debug(DEVFS_DEBUG_WARNING,
1386                             "devfs_msg_core: unknown message "
1387                             "received at core\n");
1388                 break;
1389         }
1390         lockmgr(&devfs_lock, LK_RELEASE);
1391 }
1392
1393 static void
1394 devfs_devctl_notify(cdev_t dev, const char *ev)
1395 {
1396         static const char prefix[] = "cdev=";
1397         char *data;
1398         int namelen;
1399
1400         namelen = strlen(dev->si_name);
1401         data = kmalloc(namelen + sizeof(prefix), M_TEMP, M_WAITOK);
1402         memcpy(data, prefix, sizeof(prefix) - 1);
1403         memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
1404         devctl_notify("DEVFS", "CDEV", ev, data);
1405         kfree(data, M_TEMP);
1406 }
1407
1408 /*
1409  * Worker function to insert a new dev into the dev list and initialize its
1410  * permissions. It also calls devfs_propagate_dev which in turn propagates
1411  * the change to all mount points.
1412  *
1413  * The passed dev is already referenced.  This reference is eaten by this
1414  * function and represents the dev's linkage into devfs_dev_list.
1415  */
1416 static int
1417 devfs_create_dev_worker(cdev_t dev, uid_t uid, gid_t gid, int perms)
1418 {
1419         KKASSERT(dev);
1420
1421         dev->si_uid = uid;
1422         dev->si_gid = gid;
1423         dev->si_perms = perms;
1424
1425         devfs_link_dev(dev);
1426         devfs_propagate_dev(dev, 1);
1427
1428         udev_event_attach(dev, NULL, 0);
1429         devfs_devctl_notify(dev, "CREATE");
1430
1431         return 0;
1432 }
1433
1434 /*
1435  * Worker function to delete a dev from the dev list and free the cdev.
1436  * It also calls devfs_propagate_dev which in turn propagates the change
1437  * to all mount points.
1438  */
1439 static int
1440 devfs_destroy_dev_worker(cdev_t dev)
1441 {
1442         int error;
1443
1444         KKASSERT(dev);
1445         KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1446
1447         error = devfs_unlink_dev(dev);
1448         devfs_propagate_dev(dev, 0);
1449
1450         devfs_devctl_notify(dev, "DESTROY");
1451         udev_event_detach(dev, NULL, 0);
1452
1453         if (error == 0)
1454                 release_dev(dev);       /* link ref */
1455         release_dev(dev);
1456         release_dev(dev);
1457
1458         return 0;
1459 }
1460
1461 /*
1462  * Worker function to destroy all devices with a certain basename.
1463  * Calls devfs_destroy_dev_worker for the actual destruction.
1464  */
1465 static int
1466 devfs_destroy_related_worker(cdev_t needle)
1467 {
1468         cdev_t dev;
1469
1470 restart:
1471         devfs_debug(DEVFS_DEBUG_DEBUG, "related worker: %s\n",
1472             needle->si_name);
1473         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1474                 if (dev->si_parent == needle) {
1475                         devfs_destroy_related_worker(dev);
1476                         devfs_destroy_dev_worker(dev);
1477                         goto restart;
1478                 }
1479         }
1480         return 0;
1481 }
1482
1483 static int
1484 devfs_clr_related_flag_worker(cdev_t needle, uint32_t flag)
1485 {
1486         cdev_t dev, dev1;
1487
1488         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1489                 if (dev->si_parent == needle) {
1490                         devfs_clr_related_flag_worker(dev, flag);
1491                         dev->si_flags &= ~flag;
1492                 }
1493         }
1494
1495         return 0;
1496 }
1497
1498 static int
1499 devfs_destroy_related_without_flag_worker(cdev_t needle, uint32_t flag)
1500 {
1501         cdev_t dev;
1502
1503 restart:
1504         devfs_debug(DEVFS_DEBUG_DEBUG, "related_wo_flag: %s\n",
1505             needle->si_name);
1506
1507         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1508                 if (dev->si_parent == needle) {
1509                         devfs_destroy_related_without_flag_worker(dev, flag);
1510                         if (!(dev->si_flags & flag)) {
1511                                 devfs_destroy_dev_worker(dev);
1512                                 devfs_debug(DEVFS_DEBUG_DEBUG,
1513                                     "related_wo_flag: %s restart\n", dev->si_name);
1514                                 goto restart;
1515                         }
1516                 }
1517         }
1518
1519         return 0;
1520 }
1521
1522 /*
1523  * Worker function that creates all device nodes on top of a devfs
1524  * root node.
1525  */
1526 static int
1527 devfs_create_all_dev_worker(struct devfs_node *root)
1528 {
1529         cdev_t dev;
1530
1531         KKASSERT(root);
1532
1533         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1534                 devfs_create_device_node(root, dev, NULL, NULL, NULL);
1535         }
1536
1537         return 0;
1538 }
1539
1540 /*
1541  * Worker function that destroys all devices that match a specific
1542  * dev_ops and/or minor. If minor is less than 0, it is not matched
1543  * against. It also propagates all changes.
1544  */
1545 static int
1546 devfs_destroy_dev_by_ops_worker(struct dev_ops *ops, int minor)
1547 {
1548         cdev_t dev, dev1;
1549
1550         KKASSERT(ops);
1551
1552         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1553                 if (dev->si_ops != ops)
1554                         continue;
1555                 if ((minor < 0) || (dev->si_uminor == minor)) {
1556                         devfs_destroy_dev_worker(dev);
1557                 }
1558         }
1559
1560         return 0;
1561 }
1562
1563 /*
1564  * Worker function that registers a new clone handler in devfs.
1565  */
1566 static int
1567 devfs_chandler_add_worker(const char *name, d_clone_t *nhandler)
1568 {
1569         struct devfs_clone_handler *chandler = NULL;
1570         u_char len = strlen(name);
1571
1572         if (len == 0)
1573                 return 1;
1574
1575         TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
1576                 if (chandler->namlen != len)
1577                         continue;
1578
1579                 if (!memcmp(chandler->name, name, len)) {
1580                         /* Clonable basename already exists */
1581                         return 1;
1582                 }
1583         }
1584
1585         chandler = kmalloc(sizeof(*chandler), M_DEVFS, M_WAITOK | M_ZERO);
1586         chandler->name = kstrdup(name, M_DEVFS);
1587         chandler->namlen = len;
1588         chandler->nhandler = nhandler;
1589
1590         TAILQ_INSERT_TAIL(&devfs_chandler_list, chandler, link);
1591         return 0;
1592 }
1593
1594 /*
1595  * Worker function that removes a given clone handler from the
1596  * clone handler list.
1597  */
1598 static int
1599 devfs_chandler_del_worker(const char *name)
1600 {
1601         struct devfs_clone_handler *chandler, *chandler2;
1602         u_char len = strlen(name);
1603
1604         if (len == 0)
1605                 return 1;
1606
1607         TAILQ_FOREACH_MUTABLE(chandler, &devfs_chandler_list, link, chandler2) {
1608                 if (chandler->namlen != len)
1609                         continue;
1610                 if (memcmp(chandler->name, name, len))
1611                         continue;
1612
1613                 TAILQ_REMOVE(&devfs_chandler_list, chandler, link);
1614                 kfree(chandler->name, M_DEVFS);
1615                 kfree(chandler, M_DEVFS);
1616                 break;
1617         }
1618
1619         return 0;
1620 }
1621
1622 /*
1623  * Worker function that finds a given device name and changes
1624  * the message received accordingly so that when replied to,
1625  * the answer is returned to the caller.
1626  */
1627 static int
1628 devfs_find_device_by_name_worker(devfs_msg_t devfs_msg)
1629 {
1630         struct devfs_alias *alias;
1631         cdev_t dev;
1632         cdev_t found = NULL;
1633
1634         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1635                 if (strcmp(devfs_msg->mdv_name, dev->si_name) == 0) {
1636                         found = dev;
1637                         break;
1638                 }
1639         }
1640         if (found == NULL) {
1641                 TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1642                         if (strcmp(devfs_msg->mdv_name, alias->name) == 0) {
1643                                 found = alias->dev_target;
1644                                 break;
1645                         }
1646                 }
1647         }
1648         devfs_msg->mdv_cdev = found;
1649
1650         return 0;
1651 }
1652
1653 /*
1654  * Worker function that finds a given device udev and changes
1655  * the message received accordingly so that when replied to,
1656  * the answer is returned to the caller.
1657  */
1658 static int
1659 devfs_find_device_by_devid_worker(devfs_msg_t devfs_msg)
1660 {
1661         cdev_t dev, dev1;
1662         cdev_t found = NULL;
1663
1664         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1665                 if (((dev_t)dev->si_inode) == devfs_msg->mdv_udev) {
1666                         found = dev;
1667                         break;
1668                 }
1669         }
1670         devfs_msg->mdv_cdev = found;
1671
1672         return 0;
1673 }
1674
1675 /*
1676  * Worker function that inserts a given alias into the
1677  * alias list, and propagates the alias to all mount
1678  * points.
1679  */
1680 static int
1681 devfs_make_alias_worker(struct devfs_alias *alias)
1682 {
1683         struct devfs_alias *alias2;
1684         size_t len = strlen(alias->name);
1685         int found = 0;
1686
1687         TAILQ_FOREACH(alias2, &devfs_alias_list, link) {
1688                 if (len != alias2->namlen)
1689                         continue;
1690
1691                 if (!memcmp(alias->name, alias2->name, len)) {
1692                         found = 1;
1693                         break;
1694                 }
1695         }
1696
1697         if (!found) {
1698                 /*
1699                  * The alias doesn't exist yet, so we add it to the alias list
1700                  */
1701                 TAILQ_INSERT_TAIL(&devfs_alias_list, alias, link);
1702                 devfs_alias_propagate(alias, 0);
1703                 udev_event_attach(alias->dev_target, alias->name, 1);
1704         } else {
1705                 devfs_debug(DEVFS_DEBUG_WARNING,
1706                             "Warning: duplicate devfs_make_alias for %s\n",
1707                             alias->name);
1708                 kfree(alias->name, M_DEVFS);
1709                 kfree(alias, M_DEVFS);
1710         }
1711
1712         return 0;
1713 }
1714
1715 /*
1716  * Worker function that delete a given alias from the
1717  * alias list, and propagates the removal to all mount
1718  * points.
1719  */
1720 static int
1721 devfs_destroy_alias_worker(struct devfs_alias *alias)
1722 {
1723         struct devfs_alias *alias2;
1724         int found = 0;
1725
1726         TAILQ_FOREACH(alias2, &devfs_alias_list, link) {
1727                 if (alias->dev_target != alias2->dev_target)
1728                         continue;
1729
1730                 if (devfs_WildCmp(alias->name, alias2->name) == 0) {
1731                         found = 1;
1732                         break;
1733                 }
1734         }
1735
1736         if (!found) {
1737                 devfs_debug(DEVFS_DEBUG_WARNING,
1738                     "Warning: devfs_destroy_alias for inexistant alias: %s\n",
1739                     alias->name);
1740                 kfree(alias->name, M_DEVFS);
1741                 kfree(alias, M_DEVFS);
1742         } else {
1743                 /*
1744                  * The alias exists, so we delete it from the alias list
1745                  */
1746                 TAILQ_REMOVE(&devfs_alias_list, alias2, link);
1747                 devfs_alias_propagate(alias2, 1);
1748                 udev_event_detach(alias2->dev_target, alias2->name, 1);
1749                 kfree(alias->name, M_DEVFS);
1750                 kfree(alias, M_DEVFS);
1751                 kfree(alias2->name, M_DEVFS);
1752                 kfree(alias2, M_DEVFS);
1753         }
1754
1755         return 0;
1756 }
1757
1758 /*
1759  * Function that removes and frees all aliases.
1760  */
1761 static int
1762 devfs_alias_reap(void)
1763 {
1764         struct devfs_alias *alias, *alias2;
1765
1766         TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1767                 TAILQ_REMOVE(&devfs_alias_list, alias, link);
1768                 kfree(alias->name, M_DEVFS);
1769                 kfree(alias, M_DEVFS);
1770         }
1771         return 0;
1772 }
1773
1774 /*
1775  * Function that removes an alias matching a specific cdev and frees
1776  * it accordingly.
1777  */
1778 static int
1779 devfs_alias_remove(cdev_t dev)
1780 {
1781         struct devfs_alias *alias, *alias2;
1782
1783         TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1784                 if (alias->dev_target == dev) {
1785                         TAILQ_REMOVE(&devfs_alias_list, alias, link);
1786                         udev_event_detach(alias->dev_target, alias->name, 1);
1787                         kfree(alias->name, M_DEVFS);
1788                         kfree(alias, M_DEVFS);
1789                 }
1790         }
1791         return 0;
1792 }
1793
1794 /*
1795  * This function propagates an alias addition or removal to
1796  * all mount points.
1797  */
1798 static int
1799 devfs_alias_propagate(struct devfs_alias *alias, int remove)
1800 {
1801         struct devfs_mnt_data *mnt;
1802
1803         TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1804                 if (remove) {
1805                         devfs_destroy_node(mnt->root_node, alias->name);
1806                 } else {
1807                         devfs_alias_apply(mnt->root_node, alias);
1808                 }
1809         }
1810         return 0;
1811 }
1812
1813 /*
1814  * This function is a recursive function iterating through
1815  * all device nodes in the topology and, if applicable,
1816  * creating the relevant alias for a device node.
1817  */
1818 static int
1819 devfs_alias_apply(struct devfs_node *node, struct devfs_alias *alias)
1820 {
1821         struct devfs_node *node1, *node2;
1822
1823         KKASSERT(alias != NULL);
1824
1825         if ((node->node_type == Nroot) || (node->node_type == Ndir)) {
1826                 if (node->nchildren > 2) {
1827                         TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node), link, node2) {
1828                                 devfs_alias_apply(node1, alias);
1829                         }
1830                 }
1831         } else {
1832                 if (node->d_dev == alias->dev_target)
1833                         devfs_alias_create(alias->name, node, 0);
1834         }
1835         return 0;
1836 }
1837
1838 /*
1839  * This function checks if any alias possibly is applicable
1840  * to the given node. If so, the alias is created.
1841  */
1842 static int
1843 devfs_alias_check_create(struct devfs_node *node)
1844 {
1845         struct devfs_alias *alias;
1846
1847         TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1848                 if (node->d_dev == alias->dev_target)
1849                         devfs_alias_create(alias->name, node, 0);
1850         }
1851         return 0;
1852 }
1853
1854 /*
1855  * This function creates an alias with a given name
1856  * linking to a given devfs node. It also increments
1857  * the link count on the target node.
1858  */
1859 int
1860 devfs_alias_create(char *name_orig, struct devfs_node *target, int rule_based)
1861 {
1862         struct mount *mp = target->mp;
1863         struct devfs_node *parent = DEVFS_MNTDATA(mp)->root_node;
1864         struct devfs_node *linknode;
1865         char *create_path = NULL;
1866         char *name;
1867         char *name_buf;
1868         int result = 0;
1869
1870         KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1871
1872         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1873         devfs_resolve_name_path(name_orig, name_buf, &create_path, &name);
1874
1875         if (create_path)
1876                 parent = devfs_resolve_or_create_path(parent, create_path, 1);
1877
1878
1879         if (devfs_find_device_node_by_name(parent, name)) {
1880                 devfs_debug(DEVFS_DEBUG_WARNING,
1881                             "Node already exists: %s "
1882                             "(devfs_make_alias_worker)!\n",
1883                             name);
1884                 result = 1;
1885                 goto done;
1886         }
1887
1888         linknode = devfs_allocp(Nlink, name, parent, mp, NULL);
1889         if (linknode == NULL) {
1890                 result = 1;
1891                 goto done;
1892         }
1893
1894         linknode->link_target = target;
1895         target->nlinks++;
1896
1897         if (rule_based)
1898                 linknode->flags |= DEVFS_RULE_CREATED;
1899
1900 done:
1901         kfree(name_buf, M_TEMP);
1902         return (result);
1903 }
1904
1905 /*
1906  * This function is called by the core and handles mount point
1907  * strings. It either calls the relevant worker (devfs_apply_
1908  * reset_rules_worker) on all mountpoints or only a specific
1909  * one.
1910  */
1911 static int
1912 devfs_apply_reset_rules_caller(char *mountto, int apply)
1913 {
1914         struct devfs_mnt_data *mnt;
1915
1916         if (mountto[0] == '*') {
1917                 TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1918                         devfs_iterate_topology(mnt->root_node,
1919                                         (apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1920                                         NULL);
1921                 }
1922         } else {
1923                 TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1924                         if (!strcmp(mnt->mp->mnt_stat.f_mntonname, mountto)) {
1925                                 devfs_iterate_topology(mnt->root_node,
1926                                         (apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1927                                         NULL);
1928                                 break;
1929                         }
1930                 }
1931         }
1932
1933         kfree(mountto, M_DEVFS);
1934         return 0;
1935 }
1936
1937 /*
1938  * This function calls a given callback function for
1939  * every dev node in the devfs dev list.
1940  */
1941 static int
1942 devfs_scan_callback_worker(devfs_scan_t *callback, void *arg)
1943 {
1944         cdev_t dev, dev1;
1945         struct devfs_alias *alias, *alias1;
1946
1947         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1948                 callback(dev->si_name, dev, false, arg);
1949         }
1950         TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias1) {
1951                 callback(alias->name, alias->dev_target, true, arg);
1952         }
1953
1954         return 0;
1955 }
1956
1957 /*
1958  * This function tries to resolve a given directory, or if not
1959  * found and creation requested, creates the given directory.
1960  */
1961 static struct devfs_node *
1962 devfs_resolve_or_create_dir(struct devfs_node *parent, char *dir_name,
1963                             size_t name_len, int create)
1964 {
1965         struct devfs_node *node, *found = NULL;
1966
1967         TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
1968                 if (name_len != node->d_dir.d_namlen)
1969                         continue;
1970
1971                 if (!memcmp(dir_name, node->d_dir.d_name, name_len)) {
1972                         found = node;
1973                         break;
1974                 }
1975         }
1976
1977         if ((found == NULL) && (create)) {
1978                 found = devfs_allocp(Ndir, dir_name, parent, parent->mp, NULL);
1979         }
1980
1981         return found;
1982 }
1983
1984 /*
1985  * This function tries to resolve a complete path. If creation is requested,
1986  * if a given part of the path cannot be resolved (because it doesn't exist),
1987  * it is created.
1988  */
1989 struct devfs_node *
1990 devfs_resolve_or_create_path(struct devfs_node *parent, char *path, int create)
1991 {
1992         struct devfs_node *node = parent;
1993         char *buf;
1994         size_t idx = 0;
1995
1996         if (path == NULL)
1997                 return parent;
1998
1999         buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
2000
2001         while (*path && idx < PATH_MAX - 1) {
2002                 if (*path != '/') {
2003                         buf[idx++] = *path;
2004                 } else {
2005                         buf[idx] = '\0';
2006                         node = devfs_resolve_or_create_dir(node, buf, idx, create);
2007                         if (node == NULL) {
2008                                 kfree(buf, M_TEMP);
2009                                 return NULL;
2010                         }
2011                         idx = 0;
2012                 }
2013                 ++path;
2014         }
2015         buf[idx] = '\0';
2016         node = devfs_resolve_or_create_dir(node, buf, idx, create);
2017         kfree (buf, M_TEMP);
2018         return (node);
2019 }
2020
2021 /*
2022  * Takes a full path and strips it into a directory path and a name.
2023  * For a/b/c/foo, it returns foo in namep and a/b/c in pathp. It
2024  * requires a working buffer with enough size to keep the whole
2025  * fullpath.
2026  */
2027 int
2028 devfs_resolve_name_path(char *fullpath, char *buf, char **pathp, char **namep)
2029 {
2030         char *name = NULL;
2031         char *path = NULL;
2032         size_t len = strlen(fullpath) + 1;
2033         int i;
2034
2035         KKASSERT((fullpath != NULL) && (buf != NULL));
2036         KKASSERT((pathp != NULL) && (namep != NULL));
2037
2038         memcpy(buf, fullpath, len);
2039
2040         for (i = len-1; i>= 0; i--) {
2041                 if (buf[i] == '/') {
2042                         buf[i] = '\0';
2043                         name = &(buf[i+1]);
2044                         path = buf;
2045                         break;
2046                 }
2047         }
2048
2049         *pathp = path;
2050
2051         if (name) {
2052                 *namep = name;
2053         } else {
2054                 *namep = buf;
2055         }
2056
2057         return 0;
2058 }
2059
2060 /*
2061  * This function creates a new devfs node for a given device.  It can
2062  * handle a complete path as device name, and accordingly creates
2063  * the path and the final device node.
2064  *
2065  * The reference count on the passed dev remains unchanged.
2066  */
2067 struct devfs_node *
2068 devfs_create_device_node(struct devfs_node *root, cdev_t dev,
2069                          int *existsp, char *dev_name, char *path_fmt, ...)
2070 {
2071         struct devfs_node *parent, *node = NULL;
2072         char *path = NULL;
2073         char *name;
2074         char *name_buf;
2075         __va_list ap;
2076         int i, found;
2077         char *create_path = NULL;
2078         char *names = "pqrsPQRS";
2079
2080         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
2081
2082         if (existsp)
2083                 *existsp = 0;
2084
2085         if (path_fmt != NULL) {
2086                 __va_start(ap, path_fmt);
2087                 kvasnprintf(&path, PATH_MAX, path_fmt, ap);
2088                 __va_end(ap);
2089         }
2090
2091         parent = devfs_resolve_or_create_path(root, path, 1);
2092         KKASSERT(parent);
2093
2094         devfs_resolve_name_path(
2095                         ((dev_name == NULL) && (dev))?(dev->si_name):(dev_name),
2096                         name_buf, &create_path, &name);
2097
2098         if (create_path)
2099                 parent = devfs_resolve_or_create_path(parent, create_path, 1);
2100
2101
2102         node = devfs_find_device_node_by_name(parent, name);
2103         if (node) {
2104                 if (node->d_dev == dev) {
2105                         /*
2106                          * Allow case where device caches dev after the
2107                          * close and might desire to reuse it.
2108                          */
2109                         if (existsp)
2110                                 *existsp = 1;
2111                 } else {
2112                         devfs_debug(DEVFS_DEBUG_WARNING,
2113                                     "devfs_create_device_node: "
2114                                     "DEVICE %s ALREADY EXISTS!!! "
2115                                     "Ignoring creation request.\n",
2116                                     name);
2117                         node = NULL;
2118                 }
2119                 goto out;
2120         }
2121
2122         node = devfs_allocp(Ndev, name, parent, parent->mp, dev);
2123         vfs_timestamp(&parent->mtime);
2124
2125         /*
2126          * Ugly unix98 pty magic, to hide pty master (ptm) devices and their
2127          * directory
2128          */
2129         if ((dev) && (strlen(dev->si_name) >= 4) &&
2130                         (!memcmp(dev->si_name, "ptm/", 4))) {
2131                 node->parent->flags |= DEVFS_HIDDEN;
2132                 node->flags |= DEVFS_HIDDEN;
2133         }
2134
2135         /*
2136          * Ugly pty magic, to tag pty devices as such and hide them if needed.
2137          */
2138         if ((strlen(name) >= 3) && (!memcmp(name, "pty", 3)))
2139                 node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
2140
2141         if ((strlen(name) >= 3) && (!memcmp(name, "tty", 3))) {
2142                 found = 0;
2143                 for (i = 0; i < strlen(names); i++) {
2144                         if (name[3] == names[i]) {
2145                                 found = 1;
2146                                 break;
2147                         }
2148                 }
2149                 if (found)
2150                         node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
2151         }
2152
2153 out:
2154         kfree(name_buf, M_TEMP);
2155         kvasfree(&path);
2156         return node;
2157 }
2158
2159 /*
2160  * This function finds a given device node in the topology with a given
2161  * cdev.
2162  */
2163 void *
2164 devfs_find_device_node_callback(struct devfs_node *node, cdev_t target)
2165 {
2166         if ((node->node_type == Ndev) && (node->d_dev == target)) {
2167                 return node;
2168         }
2169
2170         return NULL;
2171 }
2172
2173 /*
2174  * This function finds a device node in the given parent directory by its
2175  * name and returns it.
2176  */
2177 struct devfs_node *
2178 devfs_find_device_node_by_name(struct devfs_node *parent, char *target)
2179 {
2180         struct devfs_node *node, *found = NULL;
2181         size_t len = strlen(target);
2182
2183         TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
2184                 if (len != node->d_dir.d_namlen)
2185                         continue;
2186
2187                 if (!memcmp(node->d_dir.d_name, target, len)) {
2188                         found = node;
2189                         break;
2190                 }
2191         }
2192
2193         return found;
2194 }
2195
2196 static void *
2197 devfs_inode_to_vnode_worker_callback(struct devfs_node *node, ino_t *inop)
2198 {
2199         struct vnode *vp = NULL;
2200         ino_t target = *inop;
2201
2202         if (node->d_dir.d_ino == target) {
2203                 if (node->v_node) {
2204                         vp = node->v_node;
2205                         vget(vp, LK_EXCLUSIVE | LK_RETRY);
2206                         vn_unlock(vp);
2207                 } else {
2208                         devfs_allocv(&vp, node);
2209                         vn_unlock(vp);
2210                 }
2211         }
2212
2213         return vp;
2214 }
2215
2216 /*
2217  * This function takes a cdev and removes its devfs node in the
2218  * given topology.  The cdev remains intact.
2219  */
2220 int
2221 devfs_destroy_device_node(struct devfs_node *root, cdev_t target)
2222 {
2223         KKASSERT(target != NULL);
2224         return devfs_destroy_node(root, target->si_name);
2225 }
2226
2227 /*
2228  * This function takes a path to a devfs node, resolves it and
2229  * removes the devfs node from the given topology.
2230  */
2231 int
2232 devfs_destroy_node(struct devfs_node *root, char *target)
2233 {
2234         struct devfs_node *node, *parent;
2235         char *name;
2236         char *name_buf;
2237         char *create_path = NULL;
2238
2239         KKASSERT(target);
2240
2241         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
2242         ksnprintf(name_buf, PATH_MAX, "%s", target);
2243
2244         devfs_resolve_name_path(target, name_buf, &create_path, &name);
2245
2246         if (create_path)
2247                 parent = devfs_resolve_or_create_path(root, create_path, 0);
2248         else
2249                 parent = root;
2250
2251         if (parent == NULL) {
2252                 kfree(name_buf, M_TEMP);
2253                 return 1;
2254         }
2255
2256         node = devfs_find_device_node_by_name(parent, name);
2257
2258         if (node) {
2259                 vfs_timestamp(&node->parent->mtime);
2260                 devfs_gc(node);
2261         }
2262
2263         kfree(name_buf, M_TEMP);
2264
2265         return 0;
2266 }
2267
2268 /*
2269  * Just set perms and ownership for given node.
2270  */
2271 int
2272 devfs_set_perms(struct devfs_node *node, uid_t uid, gid_t gid,
2273                 u_short mode, u_long flags)
2274 {
2275         node->mode = mode;
2276         node->uid = uid;
2277         node->gid = gid;
2278
2279         return 0;
2280 }
2281
2282 /*
2283  * Propagates a device attach/detach to all mount
2284  * points. Also takes care of automatic alias removal
2285  * for a deleted cdev.
2286  */
2287 static int
2288 devfs_propagate_dev(cdev_t dev, int attach)
2289 {
2290         struct devfs_mnt_data *mnt;
2291
2292         TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
2293                 if (attach) {
2294                         /* Device is being attached */
2295                         devfs_create_device_node(mnt->root_node, dev,
2296                                                  NULL, NULL, NULL);
2297                 } else {
2298                         /* Device is being detached */
2299                         devfs_alias_remove(dev);
2300                         devfs_destroy_device_node(mnt->root_node, dev);
2301                 }
2302         }
2303         return 0;
2304 }
2305
2306 /*
2307  * devfs_clone either returns a basename from a complete name by
2308  * returning the length of the name without trailing digits, or,
2309  * if clone != 0, calls the device's clone handler to get a new
2310  * device, which in turn is returned in devp.
2311  *
2312  * Caller must hold a shared devfs_lock
2313  */
2314 cdev_t
2315 devfs_clone(cdev_t dev, const char *name, size_t len, int mode,
2316                 struct ucred *cred)
2317 {
2318         int error;
2319         struct devfs_clone_handler *chandler;
2320         struct dev_clone_args ap;
2321
2322         TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
2323                 if (chandler->namlen != len)
2324                         continue;
2325                 if ((!memcmp(chandler->name, name, len)) &&
2326                     (chandler->nhandler)) {
2327                         /*
2328                          * We have to unlock across the config and the
2329                          * callback to avoid deadlocking.  The device is
2330                          * likely to obtain its own lock in the callback
2331                          * and might then call into devfs.
2332                          */
2333                         lockmgr(&devfs_lock, LK_RELEASE);
2334                         devfs_config();
2335                         ap.a_head.a_dev = dev;
2336                         ap.a_dev = NULL;
2337                         ap.a_name = name;
2338                         ap.a_namelen = len;
2339                         ap.a_mode = mode;
2340                         ap.a_cred = cred;
2341                         error = (chandler->nhandler)(&ap);
2342                         lockmgr(&devfs_lock, LK_SHARED);
2343                         if (error)
2344                                 continue;
2345
2346                         return ap.a_dev;
2347                 }
2348         }
2349
2350         return NULL;
2351 }
2352
2353
2354 /*
2355  * Registers a new orphan in the orphan list.
2356  */
2357 void
2358 devfs_tracer_add_orphan(struct devfs_node *node)
2359 {
2360         struct devfs_orphan *orphan;
2361
2362         KKASSERT(node);
2363         orphan = kmalloc(sizeof(struct devfs_orphan), M_DEVFS, M_WAITOK);
2364         orphan->node = node;
2365
2366         KKASSERT((node->flags & DEVFS_ORPHANED) == 0);
2367         node->flags |= DEVFS_ORPHANED;
2368         TAILQ_INSERT_TAIL(DEVFS_ORPHANLIST(node->mp), orphan, link);
2369 }
2370
2371 /*
2372  * Removes an orphan from the orphan list.
2373  */
2374 void
2375 devfs_tracer_del_orphan(struct devfs_node *node)
2376 {
2377         struct devfs_orphan *orphan;
2378
2379         KKASSERT(node);
2380
2381         TAILQ_FOREACH(orphan, DEVFS_ORPHANLIST(node->mp), link) {
2382                 if (orphan->node == node) {
2383                         node->flags &= ~DEVFS_ORPHANED;
2384                         TAILQ_REMOVE(DEVFS_ORPHANLIST(node->mp), orphan, link);
2385                         kfree(orphan, M_DEVFS);
2386                         break;
2387                 }
2388         }
2389 }
2390
2391 /*
2392  * Counts the orphans in the orphan list, and if cleanup
2393  * is specified, also frees the orphan and removes it from
2394  * the list.
2395  */
2396 size_t
2397 devfs_tracer_orphan_count(struct mount *mp, int cleanup)
2398 {
2399         struct devfs_orphan *orphan, *orphan2;
2400         size_t count = 0;
2401
2402         TAILQ_FOREACH_MUTABLE(orphan, DEVFS_ORPHANLIST(mp), link, orphan2)      {
2403                 count++;
2404                 /*
2405                  * If we are instructed to clean up, we do so.
2406                  */
2407                 if (cleanup) {
2408                         TAILQ_REMOVE(DEVFS_ORPHANLIST(mp), orphan, link);
2409                         orphan->node->flags &= ~DEVFS_ORPHANED;
2410                         devfs_freep(orphan->node);
2411                         kfree(orphan, M_DEVFS);
2412                 }
2413         }
2414
2415         return count;
2416 }
2417
2418 /*
2419  * Fetch an ino_t from the global d_ino by increasing it
2420  * while spinlocked.
2421  */
2422 static ino_t
2423 devfs_fetch_ino(void)
2424 {
2425         ino_t   ret;
2426
2427         spin_lock(&ino_lock);
2428         ret = d_ino++;
2429         spin_unlock(&ino_lock);
2430
2431         return ret;
2432 }
2433
2434 /*
2435  * Allocates a new cdev and initializes it's most basic
2436  * fields.
2437  */
2438 cdev_t
2439 devfs_new_cdev(struct dev_ops *ops, int minor, struct dev_ops *bops)
2440 {
2441         cdev_t dev = sysref_alloc(&cdev_sysref_class);
2442
2443         sysref_activate(&dev->si_sysref);
2444         reference_dev(dev);
2445         bzero(dev, offsetof(struct cdev, si_sysref));
2446
2447         lockmgr(&devfs_lock, LK_EXCLUSIVE);
2448
2449         dev->si_uid = 0;
2450         dev->si_gid = 0;
2451         dev->si_perms = 0;
2452         dev->si_drv1 = NULL;
2453         dev->si_drv2 = NULL;
2454         dev->si_lastread = 0;           /* time_uptime */
2455         dev->si_lastwrite = 0;          /* time_uptime */
2456
2457         dev->si_dict = NULL;
2458         dev->si_parent = NULL;
2459         dev->si_ops = ops;
2460         dev->si_flags = 0;
2461         dev->si_uminor = minor;
2462         dev->si_bops = bops;
2463
2464         /*
2465          * Since the disk subsystem is in the way, we need to
2466          * propagate the D_CANFREE from bops (and ops) to
2467          * si_flags.
2468          */
2469         if (bops && (bops->head.flags & D_CANFREE)) {
2470                 dev->si_flags |= SI_CANFREE;
2471         } else if (ops->head.flags & D_CANFREE) {
2472                 dev->si_flags |= SI_CANFREE;
2473         }
2474
2475         /* If there is a backing device, we reference its ops */
2476         if (bops == NULL)
2477                 bops = ops;
2478         dev->si_inode = makeudev(devfs_reference_ops(bops), minor);
2479         dev->si_umajor = umajor(dev->si_inode);
2480
2481         lockmgr(&devfs_lock, LK_RELEASE);
2482
2483         return dev;
2484 }
2485
2486 static void
2487 devfs_cdev_terminate(cdev_t dev)
2488 {
2489         /*
2490          * Make sure the node isn't linked anymore. Otherwise we've screwed
2491          * up somewhere, since normal devs are unlinked on the call to
2492          * destroy_dev and only-cdevs that have not been used for cloning
2493          * are not linked in the first place. only-cdevs used for cloning
2494          * will be linked in, too, and should only be destroyed via
2495          * destroy_dev, not destroy_only_dev, so we catch that problem, too.
2496          */
2497         KKASSERT((dev->si_flags & SI_DEVFS_LINKED) == 0);
2498
2499         /* If there is a backing device, we release the backing device's ops */
2500         devfs_release_ops((dev->si_bops)?(dev->si_bops):(dev->si_ops));
2501
2502         /* devfs_cdev_unlock() is not called, unlock ourselves */
2503         lockmgr(&devfs_lock, LK_RELEASE);
2504
2505         /* Finally destroy the device */
2506         sysref_put(&dev->si_sysref);
2507 }
2508
2509 /*
2510  * Dummies for now (individual locks for MPSAFE)
2511  */
2512 static void
2513 devfs_cdev_lock(cdev_t dev)
2514 {
2515         lockmgr(&devfs_lock, LK_EXCLUSIVE);
2516 }
2517
2518 static void
2519 devfs_cdev_unlock(cdev_t dev)
2520 {
2521         lockmgr(&devfs_lock, LK_RELEASE);
2522 }
2523
2524 static int
2525 devfs_detached_filter_eof(struct knote *kn, long hint)
2526 {
2527         kn->kn_flags |= (EV_EOF | EV_NODATA);
2528         return (1);
2529 }
2530
2531 static void
2532 devfs_detached_filter_detach(struct knote *kn)
2533 {
2534         cdev_t dev = (cdev_t)kn->kn_hook;
2535
2536         knote_remove(&dev->si_kqinfo.ki_note, kn);
2537 }
2538
2539 static struct filterops devfs_detached_filterops =
2540         { FILTEROP_ISFD, NULL,
2541           devfs_detached_filter_detach,
2542           devfs_detached_filter_eof };
2543
2544 /*
2545  * Delegates knote filter handling responsibility to devfs
2546  *
2547  * Any device that implements kqfilter event handling and could be detached
2548  * or shut down out from under the kevent subsystem must allow devfs to
2549  * assume responsibility for any knotes it may hold.
2550  */
2551 void
2552 devfs_assume_knotes(cdev_t dev, struct kqinfo *kqi)
2553 {
2554         /*
2555          * Let kern/kern_event.c do the heavy lifting.
2556          */
2557         knote_assume_knotes(kqi, &dev->si_kqinfo,
2558                             &devfs_detached_filterops, (void *)dev);
2559
2560         /*
2561          * These should probably be activated individually, but doing so
2562          * would require refactoring kq's public in-kernel interface.
2563          */
2564         KNOTE(&dev->si_kqinfo.ki_note, 0);
2565 }
2566
2567 /*
2568  * Links a given cdev into the dev list.
2569  */
2570 int
2571 devfs_link_dev(cdev_t dev)
2572 {
2573         KKASSERT((dev->si_flags & SI_DEVFS_LINKED) == 0);
2574         dev->si_flags |= SI_DEVFS_LINKED;
2575         TAILQ_INSERT_TAIL(&devfs_dev_list, dev, link);
2576
2577         return 0;
2578 }
2579
2580 /*
2581  * Removes a given cdev from the dev list.  The caller is responsible for
2582  * releasing the reference on the device associated with the linkage.
2583  *
2584  * Returns EALREADY if the dev has already been unlinked.
2585  */
2586 static int
2587 devfs_unlink_dev(cdev_t dev)
2588 {
2589         if ((dev->si_flags & SI_DEVFS_LINKED)) {
2590                 TAILQ_REMOVE(&devfs_dev_list, dev, link);
2591                 dev->si_flags &= ~SI_DEVFS_LINKED;
2592                 return (0);
2593         }
2594         return (EALREADY);
2595 }
2596
2597 int
2598 devfs_node_is_accessible(struct devfs_node *node)
2599 {
2600         if ((node) && (!(node->flags & DEVFS_HIDDEN)))
2601                 return 1;
2602         else
2603                 return 0;
2604 }
2605
2606 /*
2607  * devfs must be locked
2608  */
2609 static int
2610 devfs_reference_ops(struct dev_ops *ops)
2611 {
2612         int unit;
2613         struct devfs_dev_ops *found = NULL;
2614         struct devfs_dev_ops *devops;
2615
2616         TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2617                 if (devops->ops == ops) {
2618                         found = devops;
2619                         break;
2620                 }
2621         }
2622
2623         if (!found) {
2624                 found = kmalloc(sizeof(struct devfs_dev_ops),
2625                                 M_DEVFS, M_WAITOK);
2626                 found->ops = ops;
2627                 found->ref_count = 0;
2628                 TAILQ_INSERT_TAIL(&devfs_dev_ops_list, found, link);
2629         }
2630
2631         KKASSERT(found);
2632
2633         if (found->ref_count == 0) {
2634                 found->id =
2635                     devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(ops_id), 255);
2636                 if (found->id == -1) {
2637                         /* Ran out of unique ids */
2638                         devfs_debug(DEVFS_DEBUG_WARNING,
2639                                     "devfs_reference_ops: WARNING: ran "
2640                                     "out of unique ids\n");
2641                 }
2642         }
2643         unit = found->id;
2644         ++found->ref_count;
2645
2646         return unit;
2647 }
2648
2649 /*
2650  * devfs must be locked
2651  */
2652 static void
2653 devfs_release_ops(struct dev_ops *ops)
2654 {
2655         struct devfs_dev_ops *found = NULL;
2656         struct devfs_dev_ops *devops;
2657
2658         TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2659                 if (devops->ops == ops) {
2660                         found = devops;
2661                         break;
2662                 }
2663         }
2664
2665         KKASSERT(found);
2666
2667         --found->ref_count;
2668
2669         if (found->ref_count == 0) {
2670                 TAILQ_REMOVE(&devfs_dev_ops_list, found, link);
2671                 lockmgr(&devfs_lock, LK_RELEASE);
2672                 devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(ops_id), found->id);
2673                 lockmgr(&devfs_lock, LK_EXCLUSIVE);
2674                 kfree(found, M_DEVFS);
2675         }
2676 }
2677
2678 /*
2679  * Wait for asynchronous messages to complete in the devfs helper
2680  * thread, then return.  Do nothing if the helper thread is dead
2681  * or we are being indirectly called from the helper thread itself.
2682  */
2683 void
2684 devfs_config(void)
2685 {
2686         devfs_msg_t msg;
2687
2688         if (devfs_run && curthread != td_core) {
2689                 msg = devfs_msg_get();
2690                 devfs_msg_send_sync(DEVFS_SYNC, msg);
2691                 devfs_msg_put(msg);
2692         }
2693 }
2694
2695 /*
2696  * Called on init of devfs; creates the objcaches and
2697  * spawns off the devfs core thread. Also initializes
2698  * locks.
2699  */
2700 static void
2701 devfs_init(void)
2702 {
2703         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init() called\n");
2704         /* Create objcaches for nodes, msgs and devs */
2705         devfs_node_cache = objcache_create("devfs-node-cache", 0, 0,
2706                                            NULL, NULL, NULL,
2707                                            objcache_malloc_alloc,
2708                                            objcache_malloc_free,
2709                                            &devfs_node_malloc_args );
2710
2711         devfs_msg_cache = objcache_create("devfs-msg-cache", 0, 0,
2712                                           NULL, NULL, NULL,
2713                                           objcache_malloc_alloc,
2714                                           objcache_malloc_free,
2715                                           &devfs_msg_malloc_args );
2716
2717         devfs_dev_cache = objcache_create("devfs-dev-cache", 0, 0,
2718                                           NULL, NULL, NULL,
2719                                           objcache_malloc_alloc,
2720                                           objcache_malloc_free,
2721                                           &devfs_dev_malloc_args );
2722
2723         devfs_clone_bitmap_init(&DEVFS_CLONE_BITMAP(ops_id));
2724
2725         /* Initialize the reply-only port which acts as a message drain */
2726         lwkt_initport_replyonly(&devfs_dispose_port, devfs_msg_autofree_reply);
2727
2728         /* Initialize *THE* devfs lock */
2729         lockinit(&devfs_lock, "devfs_core lock", 0, LK_CANRECURSE);
2730         lwkt_token_init(&devfs_token, "devfs_core");
2731
2732         lockmgr(&devfs_lock, LK_EXCLUSIVE);
2733         lwkt_create(devfs_msg_core, /*args*/NULL, &td_core, NULL,
2734                     0, -1, "devfs_msg_core");
2735         while (devfs_run == 0)
2736                 lksleep(td_core, &devfs_lock, 0, "devfsc", 0);
2737         lockmgr(&devfs_lock, LK_RELEASE);
2738
2739         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init finished\n");
2740 }
2741
2742 /*
2743  * Called on unload of devfs; takes care of destroying the core
2744  * and the objcaches. Also removes aliases that are no longer needed.
2745  */
2746 static void
2747 devfs_uninit(void)
2748 {
2749         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_uninit() called\n");
2750
2751         devfs_msg_send(DEVFS_TERMINATE_CORE, NULL);
2752         while (devfs_run)
2753                 tsleep(td_core, 0, "devfsc", hz*10);
2754         tsleep(td_core, 0, "devfsc", hz);
2755
2756         devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(ops_id));
2757
2758         /* Destroy the objcaches */
2759         objcache_destroy(devfs_msg_cache);
2760         objcache_destroy(devfs_node_cache);
2761         objcache_destroy(devfs_dev_cache);
2762
2763         devfs_alias_reap();
2764 }
2765
2766 /*
2767  * This is a sysctl handler to assist userland devname(3) to
2768  * find the device name for a given udev.
2769  */
2770 static int
2771 devfs_sysctl_devname_helper(SYSCTL_HANDLER_ARGS)
2772 {
2773         dev_t   udev;
2774         cdev_t  found;
2775         int             error;
2776
2777         if ((error = SYSCTL_IN(req, &udev, sizeof(dev_t))))
2778                 return (error);
2779
2780         devfs_debug(DEVFS_DEBUG_DEBUG,
2781                     "devfs sysctl, received udev: %d\n", udev);
2782
2783         if (udev == NOUDEV)
2784                 return(EINVAL);
2785
2786         if ((found = devfs_find_device_by_devid(udev)) == NULL)
2787                 return(ENOENT);
2788
2789         return(SYSCTL_OUT(req, found->si_name, strlen(found->si_name) + 1));
2790 }
2791
2792
2793 SYSCTL_PROC(_kern, OID_AUTO, devname,
2794             CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK,
2795             NULL, 0, devfs_sysctl_devname_helper, "",
2796             "helper for devname(3)");
2797
2798 SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2799 TUNABLE_INT("vfs.devfs.debug", &devfs_debug_enable);
2800 SYSCTL_INT(_vfs_devfs, OID_AUTO, debug, CTLFLAG_RW, &devfs_debug_enable,
2801                 0, "Enable DevFS debugging");
2802
2803 SYSINIT(vfs_devfs_register, SI_SUB_DEVFS_CORE, SI_ORDER_FIRST,
2804                 devfs_init, NULL);
2805 SYSUNINIT(vfs_devfs_register, SI_SUB_DEVFS_CORE, SI_ORDER_ANY,
2806                 devfs_uninit, NULL);
2807
2808 /*
2809  * WildCmp() - compare wild string to sane string
2810  *
2811  *      Returns 0 on success, -1 on failure.
2812  */
2813 static int
2814 wildCmp(const char **mary, int d, const char *w, const char *s)
2815 {
2816     int i;
2817
2818     /*
2819      * skip fixed portion
2820      */
2821     for (;;) {
2822         switch(*w) {
2823         case '*':
2824             /*
2825              * optimize terminator
2826              */
2827             if (w[1] == 0)
2828                 return(0);
2829             if (w[1] != '?' && w[1] != '*') {
2830                 /*
2831                  * optimize * followed by non-wild
2832                  */
2833                 for (i = 0; s + i < mary[d]; ++i) {
2834                     if (s[i] == w[1] && wildCmp(mary, d + 1, w + 1, s + i) == 0)
2835                         return(0);
2836                 }
2837             } else {
2838                 /*
2839                  * less-optimal
2840                  */
2841                 for (i = 0; s + i < mary[d]; ++i) {
2842                     if (wildCmp(mary, d + 1, w + 1, s + i) == 0)
2843                         return(0);
2844                 }
2845             }
2846             mary[d] = s;
2847             return(-1);
2848         case '?':
2849             if (*s == 0)
2850                 return(-1);
2851             ++w;
2852             ++s;
2853             break;
2854         default:
2855             if (*w != *s)
2856                 return(-1);
2857             if (*w == 0)        /* terminator */
2858                 return(0);
2859             ++w;
2860             ++s;
2861             break;
2862         }
2863     }
2864     /* not reached */
2865     return(-1);
2866 }
2867
2868
2869 /*
2870  * WildCaseCmp() - compare wild string to sane string, case insensitive
2871  *
2872  *      Returns 0 on success, -1 on failure.
2873  */
2874 static int
2875 wildCaseCmp(const char **mary, int d, const char *w, const char *s)
2876 {
2877     int i;
2878
2879     /*
2880      * skip fixed portion
2881      */
2882     for (;;) {
2883         switch(*w) {
2884         case '*':
2885             /*
2886              * optimize terminator
2887              */
2888             if (w[1] == 0)
2889                 return(0);
2890             if (w[1] != '?' && w[1] != '*') {
2891                 /*
2892                  * optimize * followed by non-wild
2893                  */
2894                 for (i = 0; s + i < mary[d]; ++i) {
2895                     if (s[i] == w[1] && wildCaseCmp(mary, d + 1, w + 1, s + i) == 0)
2896                         return(0);
2897                 }
2898             } else {
2899                 /*
2900                  * less-optimal
2901                  */
2902                 for (i = 0; s + i < mary[d]; ++i) {
2903                     if (wildCaseCmp(mary, d + 1, w + 1, s + i) == 0)
2904                         return(0);
2905                 }
2906             }
2907             mary[d] = s;
2908             return(-1);
2909         case '?':
2910             if (*s == 0)
2911                 return(-1);
2912             ++w;
2913             ++s;
2914             break;
2915         default:
2916             if (*w != *s) {
2917 #define tolower(x)      ((x >= 'A' && x <= 'Z')?(x+('a'-'A')):(x))
2918                 if (tolower(*w) != tolower(*s))
2919                     return(-1);
2920             }
2921             if (*w == 0)        /* terminator */
2922                 return(0);
2923             ++w;
2924             ++s;
2925             break;
2926         }
2927     }
2928     /* not reached */
2929     return(-1);
2930 }
2931
2932 struct cdev_privdata {
2933         void            *cdpd_data;
2934         d_priv_dtor_t   *cdpd_dtr;
2935 };
2936
2937 int
2938 devfs_get_cdevpriv(struct file *fp, void **datap)
2939 {
2940         int error;
2941
2942         if (fp == NULL)
2943                 return(EBADF);
2944
2945         spin_lock_shared(&fp->f_spin);
2946         if (fp->f_data1 == NULL) {
2947                 *datap = NULL;
2948                 error = ENOENT;
2949         } else {
2950                 struct cdev_privdata *p = fp->f_data1;
2951
2952                 *datap = p->cdpd_data;
2953                 error = 0;
2954         }
2955         spin_unlock_shared(&fp->f_spin);
2956
2957         return (error);
2958 }
2959
2960 int
2961 devfs_set_cdevpriv(struct file *fp, void *priv, d_priv_dtor_t *dtr)
2962 {
2963         struct cdev_privdata *p;
2964         int error;
2965
2966         if (fp == NULL)
2967                 return (ENOENT);
2968
2969         p = kmalloc(sizeof(struct cdev_privdata), M_DEVFS, M_WAITOK);
2970         p->cdpd_data = priv;
2971         p->cdpd_dtr = dtr;
2972
2973         spin_lock(&fp->f_spin);
2974         if (fp->f_data1 == NULL) {
2975                 fp->f_data1 = p;
2976                 error = 0;
2977         } else {
2978                 error = EBUSY;
2979         }
2980         spin_unlock(&fp->f_spin);
2981
2982         if (error)
2983                 kfree(p, M_DEVFS);
2984
2985         return error;
2986 }
2987
2988 void
2989 devfs_clear_cdevpriv(struct file *fp)
2990 {
2991         struct cdev_privdata *p;
2992
2993         if (fp == NULL)
2994                 return;
2995
2996         spin_lock(&fp->f_spin);
2997         p = fp->f_data1;
2998         fp->f_data1 = NULL;
2999         spin_unlock(&fp->f_spin);
3000
3001         if (p != NULL) {
3002                 p->cdpd_dtr(p->cdpd_data);
3003                 kfree(p, M_DEVFS);
3004         }
3005 }
3006
3007 int
3008 devfs_WildCmp(const char *w, const char *s)
3009 {
3010     int i;
3011     int c;
3012     int slen = strlen(s);
3013     const char **mary;
3014
3015     for (i = c = 0; w[i]; ++i) {
3016         if (w[i] == '*')
3017             ++c;
3018     }
3019     mary = kmalloc(sizeof(char *) * (c + 1), M_DEVFS, M_WAITOK);
3020     for (i = 0; i < c; ++i)
3021         mary[i] = s + slen;
3022     i = wildCmp(mary, 0, w, s);
3023     kfree(mary, M_DEVFS);
3024     return(i);
3025 }
3026
3027 int
3028 devfs_WildCaseCmp(const char *w, const char *s)
3029 {
3030     int i;
3031     int c;
3032     int slen = strlen(s);
3033     const char **mary;
3034
3035     for (i = c = 0; w[i]; ++i) {
3036         if (w[i] == '*')
3037             ++c;
3038     }
3039     mary = kmalloc(sizeof(char *) * (c + 1), M_DEVFS, M_WAITOK);
3040     for (i = 0; i < c; ++i)
3041         mary[i] = s + slen;
3042     i = wildCaseCmp(mary, 0, w, s);
3043     kfree(mary, M_DEVFS);
3044     return(i);
3045 }
3046