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