sys/vfs/devfs/devfs_core.c

   1 /*
   2  * Copyright (c) 2009 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/mount.h>
  38 #include <sys/vnode.h>
  39 #include <sys/types.h>
  40 #include <sys/lock.h>
  41 #include <sys/msgport.h>
  42 #include <sys/msgport2.h>
  43 #include <sys/spinlock2.h>
  44 #include <sys/sysctl.h>
  45 #include <sys/ucred.h>
  46 #include <sys/param.h>
  47 #include <sys/sysref2.h>
  48 #include <sys/systm.h>
  49 #include <sys/devfs.h>
  50 #include <sys/devfs_rules.h>
  51 #include <sys/hotplug.h>
  52
  53 MALLOC_DEFINE(M_DEVFS, "devfs", "Device File System (devfs) allocations");
  54 DEVFS_DECLARE_CLONE_BITMAP(ops_id);
  55 /*
  56  * SYSREF Integration - reference counting, allocation,
  57  * sysid and syslink integration.
  58  */
  59 static void devfs_cdev_terminate(cdev_t dev);
  60 static void devfs_cdev_lock(cdev_t dev);
  61 static void devfs_cdev_unlock(cdev_t dev);
  62 static struct sysref_class     cdev_sysref_class = {
  63         .name =         "cdev",
  64         .mtype =        M_DEVFS,
  65         .proto =        SYSREF_PROTO_DEV,
  66         .offset =       offsetof(struct cdev, si_sysref),
  67         .objsize =      sizeof(struct cdev),
  68         .mag_capacity = 32,
  69         .flags =        0,
  70         .ops =  {
  71                 .terminate = (sysref_terminate_func_t)devfs_cdev_terminate,
  72                 .lock = (sysref_lock_func_t)devfs_cdev_lock,
  73                 .unlock = (sysref_unlock_func_t)devfs_cdev_unlock
  74         }
  75 };
  76
  77 static struct objcache  *devfs_node_cache;
  78 static struct objcache  *devfs_msg_cache;
  79 static struct objcache  *devfs_dev_cache;
  80
  81 static struct objcache_malloc_args devfs_node_malloc_args = {
  82         sizeof(struct devfs_node), M_DEVFS };
  83 struct objcache_malloc_args devfs_msg_malloc_args = {
  84         sizeof(struct devfs_msg), M_DEVFS };
  85 struct objcache_malloc_args devfs_dev_malloc_args = {
  86         sizeof(struct cdev), M_DEVFS };
  87
  88 static struct devfs_dev_head devfs_dev_list =
  89                 TAILQ_HEAD_INITIALIZER(devfs_dev_list);
  90 static struct devfs_mnt_head devfs_mnt_list =
  91                 TAILQ_HEAD_INITIALIZER(devfs_mnt_list);
  92 static struct devfs_chandler_head devfs_chandler_list =
  93                 TAILQ_HEAD_INITIALIZER(devfs_chandler_list);
  94 static struct devfs_alias_head devfs_alias_list =
  95                 TAILQ_HEAD_INITIALIZER(devfs_alias_list);
  96 static struct devfs_dev_ops_head devfs_dev_ops_list =
  97                 TAILQ_HEAD_INITIALIZER(devfs_dev_ops_list);
  98
  99 struct lock             devfs_lock;
 100 static struct lwkt_port devfs_dispose_port;
 101 static struct lwkt_port devfs_msg_port;
 102 static struct thread    *td_core;
 103
 104 static struct spinlock  ino_lock;
 105 static ino_t    d_ino;
 106 static int      devfs_debug_enable;
 107 static int      devfs_run;
 108
 109 static ino_t devfs_fetch_ino(void);
 110 static int devfs_create_all_dev_worker(struct devfs_node *);
 111 static int devfs_create_dev_worker(cdev_t, uid_t, gid_t, int);
 112 static int devfs_destroy_dev_worker(cdev_t);
 113 static int devfs_destroy_subnames_worker(char *);
 114 static int devfs_destroy_dev_by_ops_worker(struct dev_ops *, int);
 115 static int devfs_propagate_dev(cdev_t, int);
 116 static int devfs_unlink_dev(cdev_t dev);
 117 static void devfs_msg_exec(devfs_msg_t msg);
 118
 119 static int devfs_chandler_add_worker(const char *, d_clone_t *);
 120 static int devfs_chandler_del_worker(const char *);
 121
 122 static void devfs_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
 123 static void devfs_msg_core(void *);
 124
 125 static int devfs_find_device_by_name_worker(devfs_msg_t);
 126 static int devfs_find_device_by_udev_worker(devfs_msg_t);
 127
 128 static int devfs_apply_reset_rules_caller(char *, int);
 129
 130 static int devfs_scan_callback_worker(devfs_scan_t *);
 131
 132 static struct devfs_node *devfs_resolve_or_create_dir(struct devfs_node *,
 133                 char *, size_t, int);
 134
 135 static int devfs_make_alias_worker(struct devfs_alias *);
 136 static int devfs_alias_remove(cdev_t);
 137 static int devfs_alias_reap(void);
 138 static int devfs_alias_propagate(struct devfs_alias *);
 139 static int devfs_alias_apply(struct devfs_node *, struct devfs_alias *);
 140 static int devfs_alias_check_create(struct devfs_node *);
 141
 142 static int devfs_clr_subnames_flag_worker(char *, uint32_t);
 143 static int devfs_destroy_subnames_without_flag_worker(char *, uint32_t);
 144
 145 static void *devfs_reaperp_callback(struct devfs_node *, void *);
 146 static void *devfs_gc_dirs_callback(struct devfs_node *, void *);
 147 static void *devfs_gc_links_callback(struct devfs_node *, struct devfs_node *);
 148 static void *
 149 devfs_inode_to_vnode_worker_callback(struct devfs_node *, ino_t *);
 150
 151 /* hotplug */
 152 void (*devfs_node_added)(struct hotplug_device*) = NULL;
 153 void (*devfs_node_removed)(struct hotplug_device*) = NULL;
 154
 155 /*
 156  * devfs_debug() is a SYSCTL and TUNABLE controlled debug output function
 157  * using kvprintf
 158  */
 159 int
 160 devfs_debug(int level, char *fmt, ...)
 161 {
 162         __va_list ap;
 163
 164         __va_start(ap, fmt);
 165         if (level <= devfs_debug_enable)
 166                 kvprintf(fmt, ap);
 167         __va_end(ap);
 168
 169         return 0;
 170 }
 171
 172 /*
 173  * devfs_allocp() Allocates a new devfs node with the specified
 174  * parameters. The node is also automatically linked into the topology
 175  * if a parent is specified. It also calls the rule and alias stuff to
 176  * be applied on the new node
 177  */
 178 struct devfs_node *
 179 devfs_allocp(devfs_nodetype devfsnodetype, char *name,
 180              struct devfs_node *parent, struct mount *mp, cdev_t dev)
 181 {
 182         struct devfs_node *node = NULL;
 183         size_t namlen = strlen(name);
 184
 185         node = objcache_get(devfs_node_cache, M_WAITOK);
 186         bzero(node, sizeof(*node));
 187
 188         atomic_add_long(&(DEVFS_MNTDATA(mp)->leak_count), 1);
 189
 190         node->d_dev = NULL;
 191         node->nchildren = 1;
 192         node->mp = mp;
 193         node->d_dir.d_ino = devfs_fetch_ino();
 194
 195         /*
 196          * Cookie jar for children. Leave 0 and 1 for '.' and '..' entries
 197          * respectively.
 198          */
 199         node->cookie_jar = 2;
 200
 201         /*
 202          * Access Control members
 203          */
 204         node->mode = DEVFS_DEFAULT_MODE;
 205         node->uid = DEVFS_DEFAULT_UID;
 206         node->gid = DEVFS_DEFAULT_GID;
 207
 208         switch (devfsnodetype) {
 209         case Proot:
 210                 /*
 211                  * Ensure that we don't recycle the root vnode by marking it as
 212                  * linked into the topology.
 213                  */
 214                 node->flags |= DEVFS_NODE_LINKED;
 215         case Pdir:
 216                 TAILQ_INIT(DEVFS_DENODE_HEAD(node));
 217                 node->d_dir.d_type = DT_DIR;
 218                 node->nchildren = 2;
 219                 break;
 220
 221         case Plink:
 222                 node->d_dir.d_type = DT_LNK;
 223                 break;
 224
 225         case Preg:
 226                 node->d_dir.d_type = DT_REG;
 227                 break;
 228
 229         case Pdev:
 230                 if (dev != NULL) {
 231                         node->d_dir.d_type = DT_CHR;
 232                         node->d_dev = dev;
 233
 234                         node->mode = dev->si_perms;
 235                         node->uid = dev->si_uid;
 236                         node->gid = dev->si_gid;
 237
 238                         devfs_alias_check_create(node);
 239                 }
 240                 break;
 241
 242         default:
 243                 panic("devfs_allocp: unknown node type");
 244         }
 245
 246         node->v_node = NULL;
 247         node->node_type = devfsnodetype;
 248
 249         /* Initialize the dirent structure of each devfs vnode */
 250         KKASSERT(namlen < 256);
 251         node->d_dir.d_namlen = namlen;
 252         node->d_dir.d_name = kmalloc(namlen+1, M_DEVFS, M_WAITOK);
 253         memcpy(node->d_dir.d_name, name, namlen);
 254         node->d_dir.d_name[namlen] = '\0';
 255
 256         /* Initialize the parent node element */
 257         node->parent = parent;
 258
 259         /* Apply rules */
 260         devfs_rule_check_apply(node, NULL);
 261
 262         /* Initialize *time members */
 263         nanotime(&node->atime);
 264         node->mtime = node->ctime = node->atime;
 265
 266         /*
 267          * Associate with parent as last step, clean out namecache
 268          * reference.
 269          */
 270         if ((parent != NULL) &&
 271             ((parent->node_type == Proot) || (parent->node_type == Pdir))) {
 272                 parent->nchildren++;
 273                 node->cookie = parent->cookie_jar++;
 274                 node->flags |= DEVFS_NODE_LINKED;
 275                 TAILQ_INSERT_TAIL(DEVFS_DENODE_HEAD(parent), node, link);
 276
 277                 /* This forces negative namecache lookups to clear */
 278                 ++mp->mnt_namecache_gen;
 279         }
 280
 281         ++DEVFS_MNTDATA(mp)->file_count;
 282
 283         return node;
 284 }
 285
 286 /*
 287  * devfs_allocv() allocates a new vnode based on a devfs node.
 288  */
 289 int
 290 devfs_allocv(struct vnode **vpp, struct devfs_node *node)
 291 {
 292         struct vnode *vp;
 293         int error = 0;
 294
 295         KKASSERT(node);
 296
 297 try_again:
 298         while ((vp = node->v_node) != NULL) {
 299                 error = vget(vp, LK_EXCLUSIVE);
 300                 if (error != ENOENT) {
 301                         *vpp = vp;
 302                         goto out;
 303                 }
 304         }
 305
 306         if ((error = getnewvnode(VT_DEVFS, node->mp, vpp, 0, 0)) != 0)
 307                 goto out;
 308
 309         vp = *vpp;
 310
 311         if (node->v_node != NULL) {
 312                 vp->v_type = VBAD;
 313                 vx_put(vp);
 314                 goto try_again;
 315         }
 316
 317         vp->v_data = node;
 318         node->v_node = vp;
 319
 320         switch (node->node_type) {
 321         case Proot:
 322                 vsetflags(vp, VROOT);
 323                 /* fall through */
 324         case Pdir:
 325                 vp->v_type = VDIR;
 326                 break;
 327
 328         case Plink:
 329                 vp->v_type = VLNK;
 330                 break;
 331
 332         case Preg:
 333                 vp->v_type = VREG;
 334                 break;
 335
 336         case Pdev:
 337                 vp->v_type = VCHR;
 338                 KKASSERT(node->d_dev);
 339
 340                 vp->v_uminor = node->d_dev->si_uminor;
 341                 vp->v_umajor = 0;
 342
 343                 v_associate_rdev(vp, node->d_dev);
 344                 vp->v_ops = &node->mp->mnt_vn_spec_ops;
 345                 break;
 346
 347         default:
 348                 panic("devfs_allocv: unknown node type");
 349         }
 350
 351 out:
 352         return error;
 353 }
 354
 355 /*
 356  * devfs_allocvp allocates both a devfs node (with the given settings) and a vnode
 357  * based on the newly created devfs node.
 358  */
 359 int
 360 devfs_allocvp(struct mount *mp, struct vnode **vpp, devfs_nodetype devfsnodetype,
 361                 char *name, struct devfs_node *parent, cdev_t dev)
 362 {
 363         struct devfs_node *node;
 364
 365         node = devfs_allocp(devfsnodetype, name, parent, mp, dev);
 366
 367         if (node != NULL)
 368                 devfs_allocv(vpp, node);
 369         else
 370                 *vpp = NULL;
 371
 372         return 0;
 373 }
 374
 375 /*
 376  * Destroy the devfs_node.  The node must be unlinked from the topology.
 377  *
 378  * This function will also destroy any vnode association with the node
 379  * and device.
 380  *
 381  * The cdev_t itself remains intact.
 382  */
 383 int
 384 devfs_freep(struct devfs_node *node)
 385 {
 386         struct vnode *vp;
 387
 388         KKASSERT(node);
 389         KKASSERT(((node->flags & DEVFS_NODE_LINKED) == 0) ||
 390                  (node->node_type == Proot));
 391         KKASSERT((node->flags & DEVFS_DESTROYED) == 0);
 392
 393         atomic_subtract_long(&(DEVFS_MNTDATA(node->mp)->leak_count), 1);
 394         if (node->symlink_name) {
 395                 kfree(node->symlink_name, M_DEVFS);
 396                 node->symlink_name = NULL;
 397         }
 398
 399         /*
 400          * Remove the node from the orphan list if it is still on it.
 401          */
 402         if (node->flags & DEVFS_ORPHANED)
 403                 devfs_tracer_del_orphan(node);
 404
 405         /*
 406          * Disassociate the vnode from the node.  This also prevents the
 407          * vnode's reclaim code from double-freeing the node.
 408          *
 409          * The vget is needed to safely modify the vp.  It also serves
 410          * to cycle the refs and terminate the vnode if it happens to
 411          * be inactive, otherwise namecache references may not get cleared.
 412          */
 413         while ((vp = node->v_node) != NULL) {
 414                 if (vget(vp, LK_EXCLUSIVE | LK_RETRY) != 0)
 415                         break;
 416                 v_release_rdev(vp);
 417                 vp->v_data = NULL;
 418                 node->v_node = NULL;
 419                 cache_inval_vp(vp, CINV_DESTROY);
 420                 vput(vp);
 421         }
 422         if (node->d_dir.d_name) {
 423                 kfree(node->d_dir.d_name, M_DEVFS);
 424                 node->d_dir.d_name = NULL;
 425         }
 426         node->flags |= DEVFS_DESTROYED;
 427
 428         --DEVFS_MNTDATA(node->mp)->file_count;
 429
 430         objcache_put(devfs_node_cache, node);
 431
 432         return 0;
 433 }
 434
 435 /*
 436  * Unlink the devfs node from the topology and add it to the orphan list.
 437  * The node will later be destroyed by freep.
 438  *
 439  * Any vnode association, including the v_rdev and v_data, remains intact
 440  * until the freep.
 441  */
 442 int
 443 devfs_unlinkp(struct devfs_node *node)
 444 {
 445         struct devfs_node *parent;
 446         struct hotplug_device *hpdev;
 447         KKASSERT(node);
 448
 449         /*
 450          * Add the node to the orphan list, so it is referenced somewhere, to
 451          * so we don't leak it.
 452          */
 453         devfs_tracer_add_orphan(node);
 454
 455         parent = node->parent;
 456
 457         /*
 458          * If the parent is known we can unlink the node out of the topology
 459          */
 460         if (parent)     {
 461                 TAILQ_REMOVE(DEVFS_DENODE_HEAD(parent), node, link);
 462                 parent->nchildren--;
 463                 KKASSERT((parent->nchildren >= 0));
 464                 node->flags &= ~DEVFS_NODE_LINKED;
 465         }
 466         /* hotplug handler */
 467         if(devfs_node_removed) {
 468                 hpdev = kmalloc(sizeof(struct hotplug_device), M_TEMP, M_WAITOK);
 469                 hpdev->dev = node->d_dev;
 470                 if(hpdev->dev)
 471                         hpdev->name = node->d_dev->si_name;
 472                 devfs_node_removed(hpdev);
 473                 kfree(hpdev, M_TEMP);
 474         }
 475         node->parent = NULL;
 476         return 0;
 477 }
 478
 479 void *
 480 devfs_iterate_topology(struct devfs_node *node,
 481                 devfs_iterate_callback_t *callback, void *arg1)
 482 {
 483         struct devfs_node *node1, *node2;
 484         void *ret = NULL;
 485
 486         if ((node->node_type == Proot) || (node->node_type == Pdir)) {
 487                 if (node->nchildren > 2) {
 488                         TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node),
 489                                                         link, node2) {
 490                                 if ((ret = devfs_iterate_topology(node1, callback, arg1)))
 491                                         return ret;
 492                         }
 493                 }
 494         }
 495
 496         ret = callback(node, arg1);
 497         return ret;
 498 }
 499
 500 /*
 501  * devfs_reaperp() is a recursive function that iterates through all the
 502  * topology, unlinking and freeing all devfs nodes.
 503  */
 504 static void *
 505 devfs_reaperp_callback(struct devfs_node *node, void *unused)
 506 {
 507         devfs_unlinkp(node);
 508         devfs_freep(node);
 509
 510         return NULL;
 511 }
 512
 513 static void *
 514 devfs_gc_dirs_callback(struct devfs_node *node, void *unused)
 515 {
 516         if (node->node_type == Pdir) {
 517                 if (node->nchildren == 2) {
 518                         devfs_unlinkp(node);
 519                         devfs_freep(node);
 520                 }
 521         }
 522
 523         return NULL;
 524 }
 525
 526 static void *
 527 devfs_gc_links_callback(struct devfs_node *node, struct devfs_node *target)
 528 {
 529         if ((node->node_type == Plink) && (node->link_target == target)) {
 530                 devfs_unlinkp(node);
 531                 devfs_freep(node);
 532         }
 533
 534         return NULL;
 535 }
 536
 537 /*
 538  * devfs_gc() is devfs garbage collector. It takes care of unlinking and
 539  * freeing a node, but also removes empty directories and links that link
 540  * via devfs auto-link mechanism to the node being deleted.
 541  */
 542 int
 543 devfs_gc(struct devfs_node *node)
 544 {
 545         struct devfs_node *root_node = DEVFS_MNTDATA(node->mp)->root_node;
 546
 547         if (node->nlinks > 0)
 548                 devfs_iterate_topology(root_node,
 549                                 (devfs_iterate_callback_t *)devfs_gc_links_callback, node);
 550
 551         devfs_unlinkp(node);
 552         devfs_iterate_topology(root_node,
 553                         (devfs_iterate_callback_t *)devfs_gc_dirs_callback, NULL);
 554
 555         devfs_freep(node);
 556
 557         return 0;
 558 }
 559
 560 /*
 561  * devfs_create_dev() is the asynchronous entry point for device creation.
 562  * It just sends a message with the relevant details to the devfs core.
 563  *
 564  * This function will reference the passed device.  The reference is owned
 565  * by devfs and represents all of the device's node associations.
 566  */
 567 int
 568 devfs_create_dev(cdev_t dev, uid_t uid, gid_t gid, int perms)
 569 {
 570         reference_dev(dev);
 571         devfs_msg_send_dev(DEVFS_DEVICE_CREATE, dev, uid, gid, perms);
 572
 573         return 0;
 574 }
 575
 576 /*
 577  * devfs_destroy_dev() is the asynchronous entry point for device destruction.
 578  * It just sends a message with the relevant details to the devfs core.
 579  */
 580 int
 581 devfs_destroy_dev(cdev_t dev)
 582 {
 583         devfs_msg_send_dev(DEVFS_DEVICE_DESTROY, dev, 0, 0, 0);
 584         return 0;
 585 }
 586
 587 /*
 588  * devfs_mount_add() is the synchronous entry point for adding a new devfs
 589  * mount.  It sends a synchronous message with the relevant details to the
 590  * devfs core.
 591  */
 592 int
 593 devfs_mount_add(struct devfs_mnt_data *mnt)
 594 {
 595         devfs_msg_t msg;
 596
 597         msg = devfs_msg_get();
 598         msg->mdv_mnt = mnt;
 599         msg = devfs_msg_send_sync(DEVFS_MOUNT_ADD, msg);
 600         devfs_msg_put(msg);
 601
 602         return 0;
 603 }
 604
 605 /*
 606  * devfs_mount_del() is the synchronous entry point for removing a devfs mount.
 607  * It sends a synchronous message with the relevant details to the devfs core.
 608  */
 609 int
 610 devfs_mount_del(struct devfs_mnt_data *mnt)
 611 {
 612         devfs_msg_t msg;
 613
 614         msg = devfs_msg_get();
 615         msg->mdv_mnt = mnt;
 616         msg = devfs_msg_send_sync(DEVFS_MOUNT_DEL, msg);
 617         devfs_msg_put(msg);
 618
 619         return 0;
 620 }
 621
 622 /*
 623  * devfs_destroy_subnames() is the synchronous entry point for device
 624  * destruction by subname. It just sends a message with the relevant details to
 625  * the devfs core.
 626  */
 627 int
 628 devfs_destroy_subnames(char *name)
 629 {
 630         devfs_msg_t msg;
 631
 632         msg = devfs_msg_get();
 633         msg->mdv_load = name;
 634         msg = devfs_msg_send_sync(DEVFS_DESTROY_SUBNAMES, msg);
 635         devfs_msg_put(msg);
 636         return 0;
 637 }
 638
 639 int
 640 devfs_clr_subnames_flag(char *name, uint32_t flag)
 641 {
 642         devfs_msg_t msg;
 643
 644         msg = devfs_msg_get();
 645         msg->mdv_flags.name = name;
 646         msg->mdv_flags.flag = flag;
 647         msg = devfs_msg_send_sync(DEVFS_CLR_SUBNAMES_FLAG, msg);
 648         devfs_msg_put(msg);
 649
 650         return 0;
 651 }
 652
 653 int
 654 devfs_destroy_subnames_without_flag(char *name, uint32_t flag)
 655 {
 656         devfs_msg_t msg;
 657
 658         msg = devfs_msg_get();
 659         msg->mdv_flags.name = name;
 660         msg->mdv_flags.flag = flag;
 661         msg = devfs_msg_send_sync(DEVFS_DESTROY_SUBNAMES_WO_FLAG, msg);
 662         devfs_msg_put(msg);
 663
 664         return 0;
 665 }
 666
 667 /*
 668  * devfs_create_all_dev is the asynchronous entry point to trigger device
 669  * node creation.  It just sends a message with the relevant details to
 670  * the devfs core.
 671  */
 672 int
 673 devfs_create_all_dev(struct devfs_node *root)
 674 {
 675         devfs_msg_send_generic(DEVFS_CREATE_ALL_DEV, root);
 676         return 0;
 677 }
 678
 679 /*
 680  * devfs_destroy_dev_by_ops is the asynchronous entry point to destroy all
 681  * devices with a specific set of dev_ops and minor.  It just sends a
 682  * message with the relevant details to the devfs core.
 683  */
 684 int
 685 devfs_destroy_dev_by_ops(struct dev_ops *ops, int minor)
 686 {
 687         devfs_msg_send_ops(DEVFS_DESTROY_DEV_BY_OPS, ops, minor);
 688         return 0;
 689 }
 690
 691 /*
 692  * devfs_clone_handler_add is the synchronous entry point to add a new
 693  * clone handler.  It just sends a message with the relevant details to
 694  * the devfs core.
 695  */
 696 int
 697 devfs_clone_handler_add(const char *name, d_clone_t *nhandler)
 698 {
 699         devfs_msg_t msg;
 700
 701         msg = devfs_msg_get();
 702         msg->mdv_chandler.name = name;
 703         msg->mdv_chandler.nhandler = nhandler;
 704         msg = devfs_msg_send_sync(DEVFS_CHANDLER_ADD, msg);
 705         devfs_msg_put(msg);
 706         return 0;
 707 }
 708
 709 /*
 710  * devfs_clone_handler_del is the synchronous entry point to remove a
 711  * clone handler.  It just sends a message with the relevant details to
 712  * the devfs core.
 713  */
 714 int
 715 devfs_clone_handler_del(const char *name)
 716 {
 717         devfs_msg_t msg;
 718
 719         msg = devfs_msg_get();
 720         msg->mdv_chandler.name = name;
 721         msg->mdv_chandler.nhandler = NULL;
 722         msg = devfs_msg_send_sync(DEVFS_CHANDLER_DEL, msg);
 723         devfs_msg_put(msg);
 724         return 0;
 725 }
 726
 727 /*
 728  * devfs_find_device_by_name is the synchronous entry point to find a
 729  * device given its name.  It sends a synchronous message with the
 730  * relevant details to the devfs core and returns the answer.
 731  */
 732 cdev_t
 733 devfs_find_device_by_name(const char *fmt, ...)
 734 {
 735         cdev_t found = NULL;
 736         devfs_msg_t msg;
 737         char *target;
 738         __va_list ap;
 739
 740         if (fmt == NULL)
 741                 return NULL;
 742
 743         __va_start(ap, fmt);
 744         kvasnrprintf(&target, PATH_MAX, 10, fmt, ap);
 745         __va_end(ap);
 746
 747         msg = devfs_msg_get();
 748         msg->mdv_name = target;
 749         msg = devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_NAME, msg);
 750         found = msg->mdv_cdev;
 751         devfs_msg_put(msg);
 752         kvasfree(&target);
 753
 754         return found;
 755 }
 756
 757 /*
 758  * devfs_find_device_by_udev is the synchronous entry point to find a
 759  * device given its udev number.  It sends a synchronous message with
 760  * the relevant details to the devfs core and returns the answer.
 761  */
 762 cdev_t
 763 devfs_find_device_by_udev(udev_t udev)
 764 {
 765         cdev_t found = NULL;
 766         devfs_msg_t msg;
 767
 768         msg = devfs_msg_get();
 769         msg->mdv_udev = udev;
 770         msg = devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_UDEV, msg);
 771         found = msg->mdv_cdev;
 772         devfs_msg_put(msg);
 773
 774         devfs_debug(DEVFS_DEBUG_DEBUG,
 775                     "devfs_find_device_by_udev found? %s  -end:3-\n",
 776                     ((found) ? found->si_name:"NO"));
 777         return found;
 778 }
 779
 780 struct vnode *
 781 devfs_inode_to_vnode(struct mount *mp, ino_t target)
 782 {
 783         struct vnode *vp = NULL;
 784         devfs_msg_t msg;
 785
 786         if (mp == NULL)
 787                 return NULL;
 788
 789         msg = devfs_msg_get();
 790         msg->mdv_ino.mp = mp;
 791         msg->mdv_ino.ino = target;
 792         msg = devfs_msg_send_sync(DEVFS_INODE_TO_VNODE, msg);
 793         vp = msg->mdv_ino.vp;
 794         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 795         devfs_msg_put(msg);
 796
 797         return vp;
 798 }
 799
 800 /*
 801  * devfs_make_alias is the asynchronous entry point to register an alias
 802  * for a device.  It just sends a message with the relevant details to the
 803  * devfs core.
 804  */
 805 int
 806 devfs_make_alias(const char *name, cdev_t dev_target)
 807 {
 808         struct devfs_alias *alias;
 809         size_t len;
 810
 811         len = strlen(name);
 812
 813         alias = kmalloc(sizeof(struct devfs_alias), M_DEVFS, M_WAITOK);
 814         alias->name = kstrdup(name, M_DEVFS);
 815         alias->namlen = len;
 816         alias->dev_target = dev_target;
 817
 818         devfs_msg_send_generic(DEVFS_MAKE_ALIAS, alias);
 819         return 0;
 820 }
 821
 822 /*
 823  * devfs_apply_rules is the asynchronous entry point to trigger application
 824  * of all rules.  It just sends a message with the relevant details to the
 825  * devfs core.
 826  */
 827 int
 828 devfs_apply_rules(char *mntto)
 829 {
 830         char *new_name;
 831
 832         new_name = kstrdup(mntto, M_DEVFS);
 833         devfs_msg_send_name(DEVFS_APPLY_RULES, new_name);
 834
 835         return 0;
 836 }
 837
 838 /*
 839  * devfs_reset_rules is the asynchronous entry point to trigger reset of all
 840  * rules. It just sends a message with the relevant details to the devfs core.
 841  */
 842 int
 843 devfs_reset_rules(char *mntto)
 844 {
 845         char *new_name;
 846
 847         new_name = kstrdup(mntto, M_DEVFS);
 848         devfs_msg_send_name(DEVFS_RESET_RULES, new_name);
 849
 850         return 0;
 851 }
 852
 853
 854 /*
 855  * devfs_scan_callback is the asynchronous entry point to call a callback
 856  * on all cdevs.
 857  * It just sends a message with the relevant details to the devfs core.
 858  */
 859 int
 860 devfs_scan_callback(devfs_scan_t *callback)
 861 {
 862         devfs_msg_t msg;
 863
 864         KKASSERT(sizeof(callback) == sizeof(void *));
 865
 866         msg = devfs_msg_get();
 867         msg->mdv_load = callback;
 868         msg = devfs_msg_send_sync(DEVFS_SCAN_CALLBACK, msg);
 869         devfs_msg_put(msg);
 870
 871         return 0;
 872 }
 873
 874
 875 /*
 876  * Acts as a message drain. Any message that is replied to here gets destroyed
 877  * and the memory freed.
 878  */
 879 static void
 880 devfs_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
 881 {
 882         devfs_msg_put((devfs_msg_t)msg);
 883 }
 884
 885 /*
 886  * devfs_msg_get allocates a new devfs msg and returns it.
 887  */
 888 devfs_msg_t
 889 devfs_msg_get(void)
 890 {
 891         return objcache_get(devfs_msg_cache, M_WAITOK);
 892 }
 893
 894 /*
 895  * devfs_msg_put deallocates a given devfs msg.
 896  */
 897 int
 898 devfs_msg_put(devfs_msg_t msg)
 899 {
 900         objcache_put(devfs_msg_cache, msg);
 901         return 0;
 902 }
 903
 904 /*
 905  * devfs_msg_send is the generic asynchronous message sending facility
 906  * for devfs. By default the reply port is the automatic disposal port.
 907  *
 908  * If the current thread is the devfs_msg_port thread we execute the
 909  * operation synchronously.
 910  */
 911 void
 912 devfs_msg_send(uint32_t cmd, devfs_msg_t devfs_msg)
 913 {
 914         lwkt_port_t port = &devfs_msg_port;
 915
 916         lwkt_initmsg(&devfs_msg->hdr, &devfs_dispose_port, 0);
 917
 918         devfs_msg->hdr.u.ms_result = cmd;
 919
 920         if (port->mpu_td == curthread) {
 921                 devfs_msg_exec(devfs_msg);
 922                 lwkt_replymsg(&devfs_msg->hdr, 0);
 923         } else {
 924                 lwkt_sendmsg(port, (lwkt_msg_t)devfs_msg);
 925         }
 926 }
 927
 928 /*
 929  * devfs_msg_send_sync is the generic synchronous message sending
 930  * facility for devfs. It initializes a local reply port and waits
 931  * for the core's answer. This answer is then returned.
 932  */
 933 devfs_msg_t
 934 devfs_msg_send_sync(uint32_t cmd, devfs_msg_t devfs_msg)
 935 {
 936         struct lwkt_port rep_port;
 937         devfs_msg_t     msg_incoming;
 938         lwkt_port_t port = &devfs_msg_port;
 939
 940         lwkt_initport_thread(&rep_port, curthread);
 941         lwkt_initmsg(&devfs_msg->hdr, &rep_port, 0);
 942
 943         devfs_msg->hdr.u.ms_result = cmd;
 944
 945         lwkt_sendmsg(port, (lwkt_msg_t)devfs_msg);
 946         msg_incoming = lwkt_waitport(&rep_port, 0);
 947
 948         return msg_incoming;
 949 }
 950
 951 /*
 952  * sends a message with a generic argument.
 953  */
 954 void
 955 devfs_msg_send_generic(uint32_t cmd, void *load)
 956 {
 957         devfs_msg_t devfs_msg = devfs_msg_get();
 958
 959         devfs_msg->mdv_load = load;
 960         devfs_msg_send(cmd, devfs_msg);
 961 }
 962
 963 /*
 964  * sends a message with a name argument.
 965  */
 966 void
 967 devfs_msg_send_name(uint32_t cmd, char *name)
 968 {
 969         devfs_msg_t devfs_msg = devfs_msg_get();
 970
 971         devfs_msg->mdv_name = name;
 972         devfs_msg_send(cmd, devfs_msg);
 973 }
 974
 975 /*
 976  * sends a message with a mount argument.
 977  */
 978 void
 979 devfs_msg_send_mount(uint32_t cmd, struct devfs_mnt_data *mnt)
 980 {
 981         devfs_msg_t devfs_msg = devfs_msg_get();
 982
 983         devfs_msg->mdv_mnt = mnt;
 984         devfs_msg_send(cmd, devfs_msg);
 985 }
 986
 987 /*
 988  * sends a message with an ops argument.
 989  */
 990 void
 991 devfs_msg_send_ops(uint32_t cmd, struct dev_ops *ops, int minor)
 992 {
 993         devfs_msg_t devfs_msg = devfs_msg_get();
 994
 995         devfs_msg->mdv_ops.ops = ops;
 996         devfs_msg->mdv_ops.minor = minor;
 997         devfs_msg_send(cmd, devfs_msg);
 998 }
 999
1000 /*
1001  * sends a message with a clone handler argument.
1002  */
1003 void
1004 devfs_msg_send_chandler(uint32_t cmd, char *name, d_clone_t handler)
1005 {
1006         devfs_msg_t devfs_msg = devfs_msg_get();
1007
1008         devfs_msg->mdv_chandler.name = name;
1009         devfs_msg->mdv_chandler.nhandler = handler;
1010         devfs_msg_send(cmd, devfs_msg);
1011 }
1012
1013 /*
1014  * sends a message with a device argument.
1015  */
1016 void
1017 devfs_msg_send_dev(uint32_t cmd, cdev_t dev, uid_t uid, gid_t gid, int perms)
1018 {
1019         devfs_msg_t devfs_msg = devfs_msg_get();
1020
1021         devfs_msg->mdv_dev.dev = dev;
1022         devfs_msg->mdv_dev.uid = uid;
1023         devfs_msg->mdv_dev.gid = gid;
1024         devfs_msg->mdv_dev.perms = perms;
1025
1026         devfs_msg_send(cmd, devfs_msg);
1027 }
1028
1029 /*
1030  * sends a message with a link argument.
1031  */
1032 void
1033 devfs_msg_send_link(uint32_t cmd, char *name, char *target, struct mount *mp)
1034 {
1035         devfs_msg_t devfs_msg = devfs_msg_get();
1036
1037         devfs_msg->mdv_link.name = name;
1038         devfs_msg->mdv_link.target = target;
1039         devfs_msg->mdv_link.mp = mp;
1040         devfs_msg_send(cmd, devfs_msg);
1041 }
1042
1043 /*
1044  * devfs_msg_core is the main devfs thread. It handles all incoming messages
1045  * and calls the relevant worker functions. By using messages it's assured
1046  * that events occur in the correct order.
1047  */
1048 static void
1049 devfs_msg_core(void *arg)
1050 {
1051         devfs_msg_t msg;
1052
1053         devfs_run = 1;
1054         lwkt_initport_thread(&devfs_msg_port, curthread);
1055         wakeup(td_core);
1056
1057         while (devfs_run) {
1058                 msg = (devfs_msg_t)lwkt_waitport(&devfs_msg_port, 0);
1059                 devfs_debug(DEVFS_DEBUG_DEBUG,
1060                                 "devfs_msg_core, new msg: %x\n",
1061                                 (unsigned int)msg->hdr.u.ms_result);
1062                 devfs_msg_exec(msg);
1063                 lwkt_replymsg(&msg->hdr, 0);
1064         }
1065         wakeup(td_core);
1066         lwkt_exit();
1067 }
1068
1069 static void
1070 devfs_msg_exec(devfs_msg_t msg)
1071 {
1072         struct devfs_mnt_data *mnt;
1073         struct devfs_node *node;
1074         cdev_t  dev;
1075
1076         /*
1077          * Acquire the devfs lock to ensure safety of all called functions
1078          */
1079         lockmgr(&devfs_lock, LK_EXCLUSIVE);
1080
1081         switch (msg->hdr.u.ms_result) {
1082         case DEVFS_DEVICE_CREATE:
1083                 dev = msg->mdv_dev.dev;
1084                 devfs_create_dev_worker(dev,
1085                                         msg->mdv_dev.uid,
1086                                         msg->mdv_dev.gid,
1087                                         msg->mdv_dev.perms);
1088                 break;
1089         case DEVFS_DEVICE_DESTROY:
1090                 dev = msg->mdv_dev.dev;
1091                 devfs_destroy_dev_worker(dev);
1092                 break;
1093         case DEVFS_DESTROY_SUBNAMES:
1094                 devfs_destroy_subnames_worker(msg->mdv_load);
1095                 break;
1096         case DEVFS_DESTROY_DEV_BY_OPS:
1097                 devfs_destroy_dev_by_ops_worker(msg->mdv_ops.ops,
1098                                                 msg->mdv_ops.minor);
1099                 break;
1100         case DEVFS_CREATE_ALL_DEV:
1101                 node = (struct devfs_node *)msg->mdv_load;
1102                 devfs_create_all_dev_worker(node);
1103                 break;
1104         case DEVFS_MOUNT_ADD:
1105                 mnt = msg->mdv_mnt;
1106                 TAILQ_INSERT_TAIL(&devfs_mnt_list, mnt, link);
1107                 devfs_create_all_dev_worker(mnt->root_node);
1108                 break;
1109         case DEVFS_MOUNT_DEL:
1110                 mnt = msg->mdv_mnt;
1111                 TAILQ_REMOVE(&devfs_mnt_list, mnt, link);
1112                 devfs_iterate_topology(mnt->root_node, devfs_reaperp_callback,
1113                                        NULL);
1114                 if (mnt->leak_count) {
1115                         devfs_debug(DEVFS_DEBUG_SHOW,
1116                                     "Leaked %ld devfs_node elements!\n",
1117                                     mnt->leak_count);
1118                 }
1119                 break;
1120         case DEVFS_CHANDLER_ADD:
1121                 devfs_chandler_add_worker(msg->mdv_chandler.name,
1122                                 msg->mdv_chandler.nhandler);
1123                 break;
1124         case DEVFS_CHANDLER_DEL:
1125                 devfs_chandler_del_worker(msg->mdv_chandler.name);
1126                 break;
1127         case DEVFS_FIND_DEVICE_BY_NAME:
1128                 devfs_find_device_by_name_worker(msg);
1129                 break;
1130         case DEVFS_FIND_DEVICE_BY_UDEV:
1131                 devfs_find_device_by_udev_worker(msg);
1132                 break;
1133         case DEVFS_MAKE_ALIAS:
1134                 devfs_make_alias_worker((struct devfs_alias *)msg->mdv_load);
1135                 break;
1136         case DEVFS_APPLY_RULES:
1137                 devfs_apply_reset_rules_caller(msg->mdv_name, 1);
1138                 break;
1139         case DEVFS_RESET_RULES:
1140                 devfs_apply_reset_rules_caller(msg->mdv_name, 0);
1141                 break;
1142         case DEVFS_SCAN_CALLBACK:
1143                 devfs_scan_callback_worker((devfs_scan_t *)msg->mdv_load);
1144                 break;
1145         case DEVFS_CLR_SUBNAMES_FLAG:
1146                 devfs_clr_subnames_flag_worker(msg->mdv_flags.name,
1147                                 msg->mdv_flags.flag);
1148                 break;
1149         case DEVFS_DESTROY_SUBNAMES_WO_FLAG:
1150                 devfs_destroy_subnames_without_flag_worker(msg->mdv_flags.name,
1151                                 msg->mdv_flags.flag);
1152                 break;
1153         case DEVFS_INODE_TO_VNODE:
1154                 msg->mdv_ino.vp = devfs_iterate_topology(
1155                         DEVFS_MNTDATA(msg->mdv_ino.mp)->root_node,
1156                         (devfs_iterate_callback_t *)devfs_inode_to_vnode_worker_callback,
1157                         &msg->mdv_ino.ino);
1158                 break;
1159         case DEVFS_TERMINATE_CORE:
1160                 devfs_run = 0;
1161                 break;
1162         case DEVFS_SYNC:
1163                 break;
1164         default:
1165                 devfs_debug(DEVFS_DEBUG_WARNING,
1166                             "devfs_msg_core: unknown message "
1167                             "received at core\n");
1168                 break;
1169         }
1170         lockmgr(&devfs_lock, LK_RELEASE);
1171 }
1172
1173 /*
1174  * Worker function to insert a new dev into the dev list and initialize its
1175  * permissions. It also calls devfs_propagate_dev which in turn propagates
1176  * the change to all mount points.
1177  *
1178  * The passed dev is already referenced.  This reference is eaten by this
1179  * function and represents the dev's linkage into devfs_dev_list.
1180  */
1181 static int
1182 devfs_create_dev_worker(cdev_t dev, uid_t uid, gid_t gid, int perms)
1183 {
1184         KKASSERT(dev);
1185
1186         dev->si_uid = uid;
1187         dev->si_gid = gid;
1188         dev->si_perms = perms;
1189
1190         devfs_link_dev(dev);
1191         devfs_propagate_dev(dev, 1);
1192
1193         return 0;
1194 }
1195
1196 /*
1197  * Worker function to delete a dev from the dev list and free the cdev.
1198  * It also calls devfs_propagate_dev which in turn propagates the change
1199  * to all mount points.
1200  */
1201 static int
1202 devfs_destroy_dev_worker(cdev_t dev)
1203 {
1204         int error;
1205
1206         KKASSERT(dev);
1207         KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1208
1209         error = devfs_unlink_dev(dev);
1210         devfs_propagate_dev(dev, 0);
1211         if (error == 0)
1212                 release_dev(dev);       /* link ref */
1213         release_dev(dev);
1214         release_dev(dev);
1215
1216         return 0;
1217 }
1218
1219 /*
1220  * Worker function to destroy all devices with a certain basename.
1221  * Calls devfs_destroy_dev_worker for the actual destruction.
1222  */
1223 static int
1224 devfs_destroy_subnames_worker(char *name)
1225 {
1226         cdev_t dev, dev1;
1227         size_t len = strlen(name);
1228
1229         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1230                 if ((!strncmp(dev->si_name, name, len)) &&
1231                                 (dev->si_name[len] != '\0')) {
1232                         devfs_destroy_dev_worker(dev);
1233                 }
1234         }
1235         return 0;
1236 }
1237
1238 static int
1239 devfs_clr_subnames_flag_worker(char *name, uint32_t flag)
1240 {
1241         cdev_t dev, dev1;
1242         size_t len = strlen(name);
1243
1244         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1245                 if ((!strncmp(dev->si_name, name, len)) &&
1246                                 (dev->si_name[len] != '\0')) {
1247                         dev->si_flags &= ~flag;
1248                 }
1249         }
1250
1251         return 0;
1252 }
1253
1254 static int
1255 devfs_destroy_subnames_without_flag_worker(char *name, uint32_t flag)
1256 {
1257         cdev_t dev, dev1;
1258         size_t len = strlen(name);
1259
1260         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1261                 if ((!strncmp(dev->si_name, name, len)) &&
1262                                 (dev->si_name[len] != '\0')) {
1263                         if (!(dev->si_flags & flag)) {
1264                                 devfs_destroy_dev_worker(dev);
1265                         }
1266                 }
1267         }
1268
1269         return 0;
1270 }
1271
1272 /*
1273  * Worker function that creates all device nodes on top of a devfs
1274  * root node.
1275  */
1276 static int
1277 devfs_create_all_dev_worker(struct devfs_node *root)
1278 {
1279         cdev_t dev;
1280
1281         KKASSERT(root);
1282
1283         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1284                 devfs_create_device_node(root, dev, NULL, NULL);
1285         }
1286
1287         return 0;
1288 }
1289
1290 /*
1291  * Worker function that destroys all devices that match a specific
1292  * dev_ops and/or minor. If minor is less than 0, it is not matched
1293  * against. It also propagates all changes.
1294  */
1295 static int
1296 devfs_destroy_dev_by_ops_worker(struct dev_ops *ops, int minor)
1297 {
1298         cdev_t dev, dev1;
1299
1300         KKASSERT(ops);
1301
1302         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1303                 if (dev->si_ops != ops)
1304                         continue;
1305                 if ((minor < 0) || (dev->si_uminor == minor)) {
1306                         devfs_destroy_dev_worker(dev);
1307                 }
1308         }
1309
1310         return 0;
1311 }
1312
1313 /*
1314  * Worker function that registers a new clone handler in devfs.
1315  */
1316 static int
1317 devfs_chandler_add_worker(const char *name, d_clone_t *nhandler)
1318 {
1319         struct devfs_clone_handler *chandler = NULL;
1320         u_char len = strlen(name);
1321
1322         if (len == 0)
1323                 return 1;
1324
1325         TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
1326                 if (chandler->namlen != len)
1327                         continue;
1328
1329                 if (!memcmp(chandler->name, name, len)) {
1330                         /* Clonable basename already exists */
1331                         return 1;
1332                 }
1333         }
1334
1335         chandler = kmalloc(sizeof(*chandler), M_DEVFS, M_WAITOK | M_ZERO);
1336         chandler->name = kstrdup(name, M_DEVFS);
1337         chandler->namlen = len;
1338         chandler->nhandler = nhandler;
1339
1340         TAILQ_INSERT_TAIL(&devfs_chandler_list, chandler, link);
1341         return 0;
1342 }
1343
1344 /*
1345  * Worker function that removes a given clone handler from the
1346  * clone handler list.
1347  */
1348 static int
1349 devfs_chandler_del_worker(const char *name)
1350 {
1351         struct devfs_clone_handler *chandler, *chandler2;
1352         u_char len = strlen(name);
1353
1354         if (len == 0)
1355                 return 1;
1356
1357         TAILQ_FOREACH_MUTABLE(chandler, &devfs_chandler_list, link, chandler2) {
1358                 if (chandler->namlen != len)
1359                         continue;
1360                 if (memcmp(chandler->name, name, len))
1361                         continue;
1362
1363                 TAILQ_REMOVE(&devfs_chandler_list, chandler, link);
1364                 kfree(chandler->name, M_DEVFS);
1365                 kfree(chandler, M_DEVFS);
1366                 break;
1367         }
1368
1369         return 0;
1370 }
1371
1372 /*
1373  * Worker function that finds a given device name and changes
1374  * the message received accordingly so that when replied to,
1375  * the answer is returned to the caller.
1376  */
1377 static int
1378 devfs_find_device_by_name_worker(devfs_msg_t devfs_msg)
1379 {
1380         struct devfs_alias *alias;
1381         cdev_t dev;
1382         cdev_t found = NULL;
1383
1384         TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1385                 if (strcmp(devfs_msg->mdv_name, dev->si_name) == 0) {
1386                         found = dev;
1387                         break;
1388                 }
1389         }
1390         if (found == NULL) {
1391                 TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1392                         if (strcmp(devfs_msg->mdv_name, alias->name) == 0) {
1393                                 found = alias->dev_target;
1394                                 break;
1395                         }
1396                 }
1397         }
1398         devfs_msg->mdv_cdev = found;
1399
1400         return 0;
1401 }
1402
1403 /*
1404  * Worker function that finds a given device udev and changes
1405  * the message received accordingly so that when replied to,
1406  * the answer is returned to the caller.
1407  */
1408 static int
1409 devfs_find_device_by_udev_worker(devfs_msg_t devfs_msg)
1410 {
1411         cdev_t dev, dev1;
1412         cdev_t found = NULL;
1413
1414         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1415                 if (((udev_t)dev->si_inode) == devfs_msg->mdv_udev) {
1416                         found = dev;
1417                         break;
1418                 }
1419         }
1420         devfs_msg->mdv_cdev = found;
1421
1422         return 0;
1423 }
1424
1425 /*
1426  * Worker function that inserts a given alias into the
1427  * alias list, and propagates the alias to all mount
1428  * points.
1429  */
1430 static int
1431 devfs_make_alias_worker(struct devfs_alias *alias)
1432 {
1433         struct devfs_alias *alias2;
1434         size_t len = strlen(alias->name);
1435         int found = 0;
1436
1437         TAILQ_FOREACH(alias2, &devfs_alias_list, link) {
1438                 if (len != alias2->namlen)
1439                         continue;
1440
1441                 if (!memcmp(alias->name, alias2->name, len)) {
1442                         found = 1;
1443                         break;
1444                 }
1445         }
1446
1447         if (!found) {
1448                 /*
1449                  * The alias doesn't exist yet, so we add it to the alias list
1450                  */
1451                 TAILQ_INSERT_TAIL(&devfs_alias_list, alias, link);
1452                 devfs_alias_propagate(alias);
1453         } else {
1454                 devfs_debug(DEVFS_DEBUG_WARNING,
1455                             "Warning: duplicate devfs_make_alias for %s\n",
1456                             alias->name);
1457                 kfree(alias->name, M_DEVFS);
1458                 kfree(alias, M_DEVFS);
1459         }
1460
1461         return 0;
1462 }
1463
1464 /*
1465  * Function that removes and frees all aliases.
1466  */
1467 static int
1468 devfs_alias_reap(void)
1469 {
1470         struct devfs_alias *alias, *alias2;
1471
1472         TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1473                 TAILQ_REMOVE(&devfs_alias_list, alias, link);
1474                 kfree(alias, M_DEVFS);
1475         }
1476         return 0;
1477 }
1478
1479 /*
1480  * Function that removes an alias matching a specific cdev and frees
1481  * it accordingly.
1482  */
1483 static int
1484 devfs_alias_remove(cdev_t dev)
1485 {
1486         struct devfs_alias *alias, *alias2;
1487
1488         TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1489                 if (alias->dev_target == dev) {
1490                         TAILQ_REMOVE(&devfs_alias_list, alias, link);
1491                         kfree(alias, M_DEVFS);
1492                 }
1493         }
1494         return 0;
1495 }
1496
1497 /*
1498  * This function propagates a new alias to all mount points.
1499  */
1500 static int
1501 devfs_alias_propagate(struct devfs_alias *alias)
1502 {
1503         struct devfs_mnt_data *mnt;
1504
1505         TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1506                 devfs_alias_apply(mnt->root_node, alias);
1507         }
1508         return 0;
1509 }
1510
1511 /*
1512  * This function is a recursive function iterating through
1513  * all device nodes in the topology and, if applicable,
1514  * creating the relevant alias for a device node.
1515  */
1516 static int
1517 devfs_alias_apply(struct devfs_node *node, struct devfs_alias *alias)
1518 {
1519         struct devfs_node *node1, *node2;
1520
1521         KKASSERT(alias != NULL);
1522
1523         if ((node->node_type == Proot) || (node->node_type == Pdir)) {
1524                 if (node->nchildren > 2) {
1525                         TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node), link, node2) {
1526                                 devfs_alias_apply(node1, alias);
1527                         }
1528                 }
1529         } else {
1530                 if (node->d_dev == alias->dev_target)
1531                         devfs_alias_create(alias->name, node, 0);
1532         }
1533         return 0;
1534 }
1535
1536 /*
1537  * This function checks if any alias possibly is applicable
1538  * to the given node. If so, the alias is created.
1539  */
1540 static int
1541 devfs_alias_check_create(struct devfs_node *node)
1542 {
1543         struct devfs_alias *alias;
1544
1545         TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1546                 if (node->d_dev == alias->dev_target)
1547                         devfs_alias_create(alias->name, node, 0);
1548         }
1549         return 0;
1550 }
1551
1552 /*
1553  * This function creates an alias with a given name
1554  * linking to a given devfs node. It also increments
1555  * the link count on the target node.
1556  */
1557 int
1558 devfs_alias_create(char *name_orig, struct devfs_node *target, int rule_based)
1559 {
1560         struct mount *mp = target->mp;
1561         struct devfs_node *parent = DEVFS_MNTDATA(mp)->root_node;
1562         struct devfs_node *linknode;
1563         struct hotplug_device *hpdev;
1564         char *create_path = NULL;
1565         char *name;
1566         char *name_buf;
1567         int result = 0;
1568
1569         KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1570
1571         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1572         devfs_resolve_name_path(name_orig, name_buf, &create_path, &name);
1573
1574         if (create_path)
1575                 parent = devfs_resolve_or_create_path(parent, create_path, 1);
1576
1577
1578         if (devfs_find_device_node_by_name(parent, name)) {
1579                 devfs_debug(DEVFS_DEBUG_WARNING,
1580                             "Node already exists: %s "
1581                             "(devfs_make_alias_worker)!\n",
1582                             name);
1583                 result = 1;
1584                 goto done;
1585         }
1586
1587         linknode = devfs_allocp(Plink, name, parent, mp, NULL);
1588         if (linknode == NULL) {
1589                 result = 1;
1590                 goto done;
1591         }
1592
1593         linknode->link_target = target;
1594         target->nlinks++;
1595
1596         if (rule_based)
1597                 linknode->flags |= DEVFS_RULE_CREATED;
1598
1599 done:
1600         /* hotplug handler */
1601         if(devfs_node_added) {
1602                 hpdev = kmalloc(sizeof(struct hotplug_device), M_TEMP, M_WAITOK);
1603                 hpdev->dev = target->d_dev;
1604                 hpdev->name = name_orig;
1605                 devfs_node_added(hpdev);
1606                 kfree(hpdev, M_TEMP);
1607         }
1608         kfree(name_buf, M_TEMP);
1609         return (result);
1610 }
1611
1612 /*
1613  * This function is called by the core and handles mount point
1614  * strings. It either calls the relevant worker (devfs_apply_
1615  * reset_rules_worker) on all mountpoints or only a specific
1616  * one.
1617  */
1618 static int
1619 devfs_apply_reset_rules_caller(char *mountto, int apply)
1620 {
1621         struct devfs_mnt_data *mnt;
1622
1623         if (mountto[0] == '*') {
1624                 TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1625                         devfs_iterate_topology(mnt->root_node,
1626                                         (apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1627                                         NULL);
1628                 }
1629         } else {
1630                 TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1631                         if (!strcmp(mnt->mp->mnt_stat.f_mntonname, mountto)) {
1632                                 devfs_iterate_topology(mnt->root_node,
1633                                         (apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1634                                         NULL);
1635                                 break;
1636                         }
1637                 }
1638         }
1639
1640         kfree(mountto, M_DEVFS);
1641         return 0;
1642 }
1643
1644 /*
1645  * This function calls a given callback function for
1646  * every dev node in the devfs dev list.
1647  */
1648 static int
1649 devfs_scan_callback_worker(devfs_scan_t *callback)
1650 {
1651         cdev_t dev, dev1;
1652
1653         TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1654                 callback(dev);
1655         }
1656
1657         return 0;
1658 }
1659
1660 /*
1661  * This function tries to resolve a given directory, or if not
1662  * found and creation requested, creates the given directory.
1663  */
1664 static struct devfs_node *
1665 devfs_resolve_or_create_dir(struct devfs_node *parent, char *dir_name,
1666                             size_t name_len, int create)
1667 {
1668         struct devfs_node *node, *found = NULL;
1669
1670         TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
1671                 if (name_len != node->d_dir.d_namlen)
1672                         continue;
1673
1674                 if (!memcmp(dir_name, node->d_dir.d_name, name_len)) {
1675                         found = node;
1676                         break;
1677                 }
1678         }
1679
1680         if ((found == NULL) && (create)) {
1681                 found = devfs_allocp(Pdir, dir_name, parent, parent->mp, NULL);
1682         }
1683
1684         return found;
1685 }
1686
1687 /*
1688  * This function tries to resolve a complete path. If creation is requested,
1689  * if a given part of the path cannot be resolved (because it doesn't exist),
1690  * it is created.
1691  */
1692 struct devfs_node *
1693 devfs_resolve_or_create_path(struct devfs_node *parent, char *path, int create)
1694 {
1695         struct devfs_node *node = parent;
1696         char *buf;
1697         size_t idx = 0;
1698
1699         if (path == NULL)
1700                 return parent;
1701
1702         buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1703
1704         while (*path && idx < PATH_MAX - 1) {
1705                 if (*path != '/') {
1706                         buf[idx++] = *path;
1707                 } else {
1708                         buf[idx] = '\0';
1709                         node = devfs_resolve_or_create_dir(node, buf, idx, create);
1710                         if (node == NULL) {
1711                                 kfree(buf, M_TEMP);
1712                                 return NULL;
1713                         }
1714                         idx = 0;
1715                 }
1716                 ++path;
1717         }
1718         buf[idx] = '\0';
1719         node = devfs_resolve_or_create_dir(node, buf, idx, create);
1720         kfree (buf, M_TEMP);
1721         return (node);
1722 }
1723
1724 /*
1725  * Takes a full path and strips it into a directory path and a name.
1726  * For a/b/c/foo, it returns foo in namep and a/b/c in pathp. It
1727  * requires a working buffer with enough size to keep the whole
1728  * fullpath.
1729  */
1730 int
1731 devfs_resolve_name_path(char *fullpath, char *buf, char **pathp, char **namep)
1732 {
1733         char *name = NULL;
1734         char *path = NULL;
1735         size_t len = strlen(fullpath) + 1;
1736         int i;
1737
1738         KKASSERT((fullpath != NULL) && (buf != NULL));
1739         KKASSERT((pathp != NULL) && (namep != NULL));
1740
1741         memcpy(buf, fullpath, len);
1742
1743         for (i = len-1; i>= 0; i--) {
1744                 if (buf[i] == '/') {
1745                         buf[i] = '\0';
1746                         name = &(buf[i+1]);
1747                         path = buf;
1748                         break;
1749                 }
1750         }
1751
1752         *pathp = path;
1753
1754         if (name) {
1755                 *namep = name;
1756         } else {
1757                 *namep = buf;
1758         }
1759
1760         return 0;
1761 }
1762
1763 /*
1764  * This function creates a new devfs node for a given device.  It can
1765  * handle a complete path as device name, and accordingly creates
1766  * the path and the final device node.
1767  *
1768  * The reference count on the passed dev remains unchanged.
1769  */
1770 struct devfs_node *
1771 devfs_create_device_node(struct devfs_node *root, cdev_t dev,
1772                          char *dev_name, char *path_fmt, ...)
1773 {
1774         struct devfs_node *parent, *node = NULL;
1775         struct hotplug_device *hpdev;
1776         char *path = NULL;
1777         char *name;
1778         char *name_buf;
1779         __va_list ap;
1780         int i, found;
1781         char *create_path = NULL;
1782         char *names = "pqrsPQRS";
1783
1784         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1785
1786         if (path_fmt != NULL) {
1787                 __va_start(ap, path_fmt);
1788                 kvasnrprintf(&path, PATH_MAX, 10, path_fmt, ap);
1789                 __va_end(ap);
1790         }
1791
1792         parent = devfs_resolve_or_create_path(root, path, 1);
1793         KKASSERT(parent);
1794
1795         devfs_resolve_name_path(
1796                         ((dev_name == NULL) && (dev))?(dev->si_name):(dev_name),
1797                         name_buf, &create_path, &name);
1798
1799         if (create_path)
1800                 parent = devfs_resolve_or_create_path(parent, create_path, 1);
1801
1802
1803         if (devfs_find_device_node_by_name(parent, name)) {
1804                 devfs_debug(DEVFS_DEBUG_WARNING, "devfs_create_device_node: "
1805                         "DEVICE %s ALREADY EXISTS!!! Ignoring creation request.\n", name);
1806                 goto out;
1807         }
1808
1809         node = devfs_allocp(Pdev, name, parent, parent->mp, dev);
1810         nanotime(&parent->mtime);
1811
1812         /*
1813          * Ugly unix98 pty magic, to hide pty master (ptm) devices and their
1814          * directory
1815          */
1816         if ((dev) && (strlen(dev->si_name) >= 4) &&
1817                         (!memcmp(dev->si_name, "ptm/", 4))) {
1818                 node->parent->flags |= DEVFS_HIDDEN;
1819                 node->flags |= DEVFS_HIDDEN;
1820         }
1821
1822         /*
1823          * Ugly pty magic, to tag pty devices as such and hide them if needed.
1824          */
1825         if ((strlen(name) >= 3) && (!memcmp(name, "pty", 3)))
1826                 node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
1827
1828         if ((strlen(name) >= 3) && (!memcmp(name, "tty", 3))) {
1829                 found = 0;
1830                 for (i = 0; i < strlen(names); i++) {
1831                         if (name[3] == names[i]) {
1832                                 found = 1;
1833                                 break;
1834                         }
1835                 }
1836                 if (found)
1837                         node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
1838         }
1839         /* hotplug handler */
1840         if(devfs_node_added) {
1841                 hpdev = kmalloc(sizeof(struct hotplug_device), M_TEMP, M_WAITOK);
1842                 hpdev->dev = node->d_dev;
1843                 hpdev->name = node->d_dev->si_name;
1844                 devfs_node_added(hpdev);
1845                 kfree(hpdev, M_TEMP);
1846         }
1847
1848 out:
1849         kfree(name_buf, M_TEMP);
1850         kvasfree(&path);
1851         return node;
1852 }
1853
1854 /*
1855  * This function finds a given device node in the topology with a given
1856  * cdev.
1857  */
1858 void *
1859 devfs_find_device_node_callback(struct devfs_node *node, cdev_t target)
1860 {
1861         if ((node->node_type == Pdev) && (node->d_dev == target)) {
1862                 return node;
1863         }
1864
1865         return NULL;
1866 }
1867
1868 /*
1869  * This function finds a device node in the given parent directory by its
1870  * name and returns it.
1871  */
1872 struct devfs_node *
1873 devfs_find_device_node_by_name(struct devfs_node *parent, char *target)
1874 {
1875         struct devfs_node *node, *found = NULL;
1876         size_t len = strlen(target);
1877
1878         TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
1879                 if (len != node->d_dir.d_namlen)
1880                         continue;
1881
1882                 if (!memcmp(node->d_dir.d_name, target, len)) {
1883                         found = node;
1884                         break;
1885                 }
1886         }
1887
1888         return found;
1889 }
1890
1891 static void *
1892 devfs_inode_to_vnode_worker_callback(struct devfs_node *node, ino_t *inop)
1893 {
1894         struct vnode *vp = NULL;
1895         ino_t target = *inop;
1896
1897         if (node->d_dir.d_ino == target) {
1898                 if (node->v_node) {
1899                         vp = node->v_node;
1900                         vget(vp, LK_EXCLUSIVE | LK_RETRY);
1901                         vn_unlock(vp);
1902                 } else {
1903                         devfs_allocv(&vp, node);
1904                         vn_unlock(vp);
1905                 }
1906         }
1907
1908         return vp;
1909 }
1910
1911 /*
1912  * This function takes a cdev and removes its devfs node in the
1913  * given topology.  The cdev remains intact.
1914  */
1915 int
1916 devfs_destroy_device_node(struct devfs_node *root, cdev_t target)
1917 {
1918         struct devfs_node *node, *parent;
1919         char *name;
1920         char *name_buf;
1921         char *create_path = NULL;
1922
1923         KKASSERT(target);
1924
1925         name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1926         ksnprintf(name_buf, PATH_MAX, "%s", target->si_name);
1927
1928         devfs_resolve_name_path(target->si_name, name_buf, &create_path, &name);
1929
1930         if (create_path)
1931                 parent = devfs_resolve_or_create_path(root, create_path, 0);
1932         else
1933                 parent = root;
1934
1935         if (parent == NULL)
1936                 return 1;
1937
1938         node = devfs_find_device_node_by_name(parent, name);
1939
1940         if (node) {
1941                 nanotime(&node->parent->mtime);
1942                 devfs_gc(node);
1943         }
1944
1945         kfree(name_buf, M_TEMP);
1946
1947         return 0;
1948 }
1949
1950 /*
1951  * Just set perms and ownership for given node.
1952  */
1953 int
1954 devfs_set_perms(struct devfs_node *node, uid_t uid, gid_t gid,
1955                 u_short mode, u_long flags)
1956 {
1957         node->mode = mode;
1958         node->uid = uid;
1959         node->gid = gid;
1960
1961         return 0;
1962 }
1963
1964 /*
1965  * Propagates a device attach/detach to all mount
1966  * points. Also takes care of automatic alias removal
1967  * for a deleted cdev.
1968  */
1969 static int
1970 devfs_propagate_dev(cdev_t dev, int attach)
1971 {
1972         struct devfs_mnt_data *mnt;
1973
1974         TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1975                 if (attach) {
1976                         /* Device is being attached */
1977                         devfs_create_device_node(mnt->root_node, dev,
1978                                                  NULL, NULL );
1979                 } else {
1980                         /* Device is being detached */
1981                         devfs_alias_remove(dev);
1982                         devfs_destroy_device_node(mnt->root_node, dev);
1983                 }
1984         }
1985         return 0;
1986 }
1987
1988 /*
1989  * devfs_clone either returns a basename from a complete name by
1990  * returning the length of the name without trailing digits, or,
1991  * if clone != 0, calls the device's clone handler to get a new
1992  * device, which in turn is returned in devp.
1993  */
1994 cdev_t
1995 devfs_clone(cdev_t dev, const char *name, size_t len, int mode,
1996                 struct ucred *cred)
1997 {
1998         int error;
1999         struct devfs_clone_handler *chandler;
2000         struct dev_clone_args ap;
2001
2002         TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
2003                 if (chandler->namlen != len)
2004                         continue;
2005                 if ((!memcmp(chandler->name, name, len)) && (chandler->nhandler)) {
2006                         lockmgr(&devfs_lock, LK_RELEASE);
2007                         devfs_config();
2008                         lockmgr(&devfs_lock, LK_EXCLUSIVE);
2009
2010                         ap.a_head.a_dev = dev;
2011                         ap.a_dev = NULL;
2012                         ap.a_name = name;
2013                         ap.a_namelen = len;
2014                         ap.a_mode = mode;
2015                         ap.a_cred = cred;
2016                         error = (chandler->nhandler)(&ap);
2017                         if (error)
2018                                 continue;
2019
2020                         return ap.a_dev;
2021                 }
2022         }
2023
2024         return NULL;
2025 }
2026
2027
2028 /*
2029  * Registers a new orphan in the orphan list.
2030  */
2031 void
2032 devfs_tracer_add_orphan(struct devfs_node *node)
2033 {
2034         struct devfs_orphan *orphan;
2035
2036         KKASSERT(node);
2037         orphan = kmalloc(sizeof(struct devfs_orphan), M_DEVFS, M_WAITOK);
2038         orphan->node = node;
2039
2040         KKASSERT((node->flags & DEVFS_ORPHANED) == 0);
2041         node->flags |= DEVFS_ORPHANED;
2042         TAILQ_INSERT_TAIL(DEVFS_ORPHANLIST(node->mp), orphan, link);
2043 }
2044
2045 /*
2046  * Removes an orphan from the orphan list.
2047  */
2048 void
2049 devfs_tracer_del_orphan(struct devfs_node *node)
2050 {
2051         struct devfs_orphan *orphan;
2052
2053         KKASSERT(node);
2054
2055         TAILQ_FOREACH(orphan, DEVFS_ORPHANLIST(node->mp), link) {
2056                 if (orphan->node == node) {
2057                         node->flags &= ~DEVFS_ORPHANED;
2058                         TAILQ_REMOVE(DEVFS_ORPHANLIST(node->mp), orphan, link);
2059                         kfree(orphan, M_DEVFS);
2060                         break;
2061                 }
2062         }
2063 }
2064
2065 /*
2066  * Counts the orphans in the orphan list, and if cleanup
2067  * is specified, also frees the orphan and removes it from
2068  * the list.
2069  */
2070 size_t
2071 devfs_tracer_orphan_count(struct mount *mp, int cleanup)
2072 {
2073         struct devfs_orphan *orphan, *orphan2;
2074         size_t count = 0;
2075
2076         TAILQ_FOREACH_MUTABLE(orphan, DEVFS_ORPHANLIST(mp), link, orphan2)      {
2077                 count++;
2078                 /*
2079                  * If we are instructed to clean up, we do so.
2080                  */
2081                 if (cleanup) {
2082                         TAILQ_REMOVE(DEVFS_ORPHANLIST(mp), orphan, link);
2083                         orphan->node->flags &= ~DEVFS_ORPHANED;
2084                         devfs_freep(orphan->node);
2085                         kfree(orphan, M_DEVFS);
2086                 }
2087         }
2088
2089         return count;
2090 }
2091
2092 /*
2093  * Fetch an ino_t from the global d_ino by increasing it
2094  * while spinlocked.
2095  */
2096 static ino_t
2097 devfs_fetch_ino(void)
2098 {
2099         ino_t   ret;
2100
2101         spin_lock_wr(&ino_lock);
2102         ret = d_ino++;
2103         spin_unlock_wr(&ino_lock);
2104
2105         return ret;
2106 }
2107
2108 /*
2109  * Allocates a new cdev and initializes it's most basic
2110  * fields.
2111  */
2112 cdev_t
2113 devfs_new_cdev(struct dev_ops *ops, int minor, struct dev_ops *bops)
2114 {
2115         cdev_t dev = sysref_alloc(&cdev_sysref_class);
2116
2117         sysref_activate(&dev->si_sysref);
2118         reference_dev(dev);
2119         bzero(dev, offsetof(struct cdev, si_sysref));
2120
2121         dev->si_uid = 0;
2122         dev->si_gid = 0;
2123         dev->si_perms = 0;
2124         dev->si_drv1 = NULL;
2125         dev->si_drv2 = NULL;
2126         dev->si_lastread = 0;           /* time_second */
2127         dev->si_lastwrite = 0;          /* time_second */
2128
2129         dev->si_ops = ops;
2130         dev->si_flags = 0;
2131         dev->si_umajor = 0;
2132         dev->si_uminor = minor;
2133         dev->si_bops = bops;
2134         /* If there is a backing device, we reference its ops */
2135         dev->si_inode = makeudev(
2136                     devfs_reference_ops((bops)?(bops):(ops)),
2137                     minor );
2138
2139         return dev;
2140 }
2141
2142 static void
2143 devfs_cdev_terminate(cdev_t dev)
2144 {
2145         int locked = 0;
2146
2147         /* Check if it is locked already. if not, we acquire the devfs lock */
2148         if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
2149                 lockmgr(&devfs_lock, LK_EXCLUSIVE);
2150                 locked = 1;
2151         }
2152
2153         /* Propagate destruction, just in case */
2154         devfs_propagate_dev(dev, 0);
2155
2156         /* If we acquired the lock, we also get rid of it */
2157         if (locked)
2158                 lockmgr(&devfs_lock, LK_RELEASE);
2159
2160         /* If there is a backing device, we release the backing device's ops */
2161         devfs_release_ops((dev->si_bops)?(dev->si_bops):(dev->si_ops));
2162
2163         /* Finally destroy the device */
2164         sysref_put(&dev->si_sysref);
2165 }
2166
2167 /*
2168  * Dummies for now (individual locks for MPSAFE)
2169  */
2170 static void
2171 devfs_cdev_lock(cdev_t dev)
2172 {
2173 }
2174
2175 static void
2176 devfs_cdev_unlock(cdev_t dev)
2177 {
2178 }
2179
2180 /*
2181  * Links a given cdev into the dev list.
2182  */
2183 int
2184 devfs_link_dev(cdev_t dev)
2185 {
2186         KKASSERT((dev->si_flags & SI_DEVFS_LINKED) == 0);
2187         dev->si_flags |= SI_DEVFS_LINKED;
2188         TAILQ_INSERT_TAIL(&devfs_dev_list, dev, link);
2189
2190         return 0;
2191 }
2192
2193 /*
2194  * Removes a given cdev from the dev list.  The caller is responsible for
2195  * releasing the reference on the device associated with the linkage.
2196  *
2197  * Returns EALREADY if the dev has already been unlinked.
2198  */
2199 static int
2200 devfs_unlink_dev(cdev_t dev)
2201 {
2202         if ((dev->si_flags & SI_DEVFS_LINKED)) {
2203                 TAILQ_REMOVE(&devfs_dev_list, dev, link);
2204                 dev->si_flags &= ~SI_DEVFS_LINKED;
2205                 return (0);
2206         }
2207         return (EALREADY);
2208 }
2209
2210 int
2211 devfs_node_is_accessible(struct devfs_node *node)
2212 {
2213         if ((node) && (!(node->flags & DEVFS_HIDDEN)))
2214                 return 1;
2215         else
2216                 return 0;
2217 }
2218
2219 int
2220 devfs_reference_ops(struct dev_ops *ops)
2221 {
2222         int unit;
2223         struct devfs_dev_ops *found = NULL;
2224         struct devfs_dev_ops *devops;
2225
2226         TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2227                 if (devops->ops == ops) {
2228                         found = devops;
2229                         break;
2230                 }
2231         }
2232
2233         if (!found) {
2234                 found = kmalloc(sizeof(struct devfs_dev_ops), M_DEVFS, M_WAITOK);
2235                 found->ops = ops;
2236                 found->ref_count = 0;
2237                 TAILQ_INSERT_TAIL(&devfs_dev_ops_list, found, link);
2238         }
2239
2240         KKASSERT(found);
2241
2242         if (found->ref_count == 0) {
2243                 found->id = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(ops_id), 255);
2244                 if (found->id == -1) {
2245                         /* Ran out of unique ids */
2246                         devfs_debug(DEVFS_DEBUG_WARNING,
2247                                         "devfs_reference_ops: WARNING: ran out of unique ids\n");
2248                 }
2249         }
2250         unit = found->id;
2251         ++found->ref_count;
2252
2253         return unit;
2254 }
2255
2256 void
2257 devfs_release_ops(struct dev_ops *ops)
2258 {
2259         struct devfs_dev_ops *found = NULL;
2260         struct devfs_dev_ops *devops;
2261
2262         TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2263                 if (devops->ops == ops) {
2264                         found = devops;
2265                         break;
2266                 }
2267         }
2268
2269         KKASSERT(found);
2270
2271         --found->ref_count;
2272
2273         if (found->ref_count == 0) {
2274                 TAILQ_REMOVE(&devfs_dev_ops_list, found, link);
2275                 devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(ops_id), found->id);
2276                 kfree(found, M_DEVFS);
2277         }
2278 }
2279
2280 void
2281 devfs_config(void)
2282 {
2283         devfs_msg_t msg;
2284
2285         msg = devfs_msg_get();
2286         msg = devfs_msg_send_sync(DEVFS_SYNC, msg);
2287         devfs_msg_put(msg);
2288 }
2289
2290 /*
2291  * Called on init of devfs; creates the objcaches and
2292  * spawns off the devfs core thread. Also initializes
2293  * locks.
2294  */
2295 static void
2296 devfs_init(void)
2297 {
2298         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init() called\n");
2299         /* Create objcaches for nodes, msgs and devs */
2300         devfs_node_cache = objcache_create("devfs-node-cache", 0, 0,
2301                                            NULL, NULL, NULL,
2302                                            objcache_malloc_alloc,
2303                                            objcache_malloc_free,
2304                                            &devfs_node_malloc_args );
2305
2306         devfs_msg_cache = objcache_create("devfs-msg-cache", 0, 0,
2307                                           NULL, NULL, NULL,
2308                                           objcache_malloc_alloc,
2309                                           objcache_malloc_free,
2310                                           &devfs_msg_malloc_args );
2311
2312         devfs_dev_cache = objcache_create("devfs-dev-cache", 0, 0,
2313                                           NULL, NULL, NULL,
2314                                           objcache_malloc_alloc,
2315                                           objcache_malloc_free,
2316                                           &devfs_dev_malloc_args );
2317
2318         devfs_clone_bitmap_init(&DEVFS_CLONE_BITMAP(ops_id));
2319
2320         /* Initialize the reply-only port which acts as a message drain */
2321         lwkt_initport_replyonly(&devfs_dispose_port, devfs_msg_autofree_reply);
2322
2323         /* Initialize *THE* devfs lock */
2324         lockinit(&devfs_lock, "devfs_core lock", 0, 0);
2325
2326
2327         lwkt_create(devfs_msg_core, /*args*/NULL, &td_core, NULL,
2328                     0, 0, "devfs_msg_core");
2329
2330         tsleep(td_core/*devfs_id*/, 0, "devfsc", 0);
2331
2332         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init finished\n");
2333 }
2334
2335 /*
2336  * Called on unload of devfs; takes care of destroying the core
2337  * and the objcaches. Also removes aliases that are no longer needed.
2338  */
2339 static void
2340 devfs_uninit(void)
2341 {
2342         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_uninit() called\n");
2343
2344         devfs_msg_send(DEVFS_TERMINATE_CORE, NULL);
2345
2346         tsleep(td_core/*devfs_id*/, 0, "devfsc", 0);
2347         tsleep(td_core/*devfs_id*/, 0, "devfsc", 10000);
2348
2349         devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(ops_id));
2350
2351         /* Destroy the objcaches */
2352         objcache_destroy(devfs_msg_cache);
2353         objcache_destroy(devfs_node_cache);
2354         objcache_destroy(devfs_dev_cache);
2355
2356         devfs_alias_reap();
2357 }
2358
2359 /*
2360  * This is a sysctl handler to assist userland devname(3) to
2361  * find the device name for a given udev.
2362  */
2363 static int
2364 devfs_sysctl_devname_helper(SYSCTL_HANDLER_ARGS)
2365 {
2366         udev_t  udev;
2367         cdev_t  found;
2368         int             error;
2369
2370
2371         if ((error = SYSCTL_IN(req, &udev, sizeof(udev_t))))
2372                 return (error);
2373
2374         devfs_debug(DEVFS_DEBUG_DEBUG, "devfs sysctl, received udev: %d\n", udev);
2375
2376         if (udev == NOUDEV)
2377                 return(EINVAL);
2378
2379         if ((found = devfs_find_device_by_udev(udev)) == NULL)
2380                 return(ENOENT);
2381
2382         return(SYSCTL_OUT(req, found->si_name, strlen(found->si_name) + 1));
2383 }
2384
2385
2386 SYSCTL_PROC(_kern, OID_AUTO, devname, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY,
2387                         NULL, 0, devfs_sysctl_devname_helper, "", "helper for devname(3)");
2388
2389 SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2390 TUNABLE_INT("vfs.devfs.debug", &devfs_debug_enable);
2391 SYSCTL_INT(_vfs_devfs, OID_AUTO, debug, CTLFLAG_RW, &devfs_debug_enable,
2392                 0, "Enable DevFS debugging");
2393
2394 SYSINIT(vfs_devfs_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST,
2395                 devfs_init, NULL);
2396 SYSUNINIT(vfs_devfs_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY,
2397                 devfs_uninit, NULL);