xref: /xnu-8019.80.24/bsd/miscfs/devfs/devfs_tree.c (revision a325d9c4a84054e40bbe985afedcb50ab80993ea)
1 /*
2  * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 /*
30  * Copyright 1997,1998 Julian Elischer.  All rights reserved.
31  * [email protected]
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions are
35  * met:
36  *  1. Redistributions of source code must retain the above copyright
37  *     notice, this list of conditions and the following disclaimer.
38  *  2. Redistributions in binary form must reproduce the above copyright notice,
39  *     this list of conditions and the following disclaimer in the documentation
40  *     and/or other materials provided with the distribution.
41  *
42  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS
43  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
44  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
45  * DISCLAIMED.  IN NO EVENT SHALL THE HOLDER OR CONTRIBUTORS BE LIABLE FOR
46  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
48  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
49  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52  * SUCH DAMAGE.
53  *
54  * devfs_tree.c
55  */
56 /*
57  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
58  * support for mandatory and extensible security protections.  This notice
59  * is included in support of clause 2.2 (b) of the Apple Public License,
60  * Version 2.0.
61  */
62 
63 /*
64  * HISTORY
65  *  Dieter Siegmund ([email protected]) Thu Apr  8 14:08:19 PDT 1999
66  *  - removed mounting of "hidden" mountpoint
67  *  - fixed problem in which devnode->dn_vn pointer was not
68  *    updated with the vnode returned from checkalias()
69  *  - replaced devfs_vntodn() with a macro VTODN()
70  *  - rewrote dev_finddir() to not use recursion
71  *  - added locking to avoid data structure corruption (DEVFS_(UN)LOCK())
72  *  Dieter Siegmund ([email protected]) Wed Jul 14 13:37:59 PDT 1999
73  *  - fixed problem with devfs_dntovn() checking the v_id against the
74  *    value cached in the device node; a union mount on top of us causes
75  *    the v_id to get incremented thus, we would end up returning a new
76  *    vnode instead of the existing one that has the mounted_here
77  *    field filled in; the net effect was that the filesystem mounted
78  *    on top of us would never show up
79  *  - added devfs_stats to store how many data structures are actually
80  *    allocated
81  */
82 
83 /* SPLIT_DEVS means each devfs uses a different devnode for the same device */
84 /* Otherwise the same device always ends up at the same vnode even if  */
85 /* reached througgh a different devfs instance. The practical difference */
86 /* is that with the same vnode, chmods and chowns show up on all instances of */
87 /* a device. (etc) */
88 
89 #define SPLIT_DEVS 1 /* maybe make this an option */
90 /*#define SPLIT_DEVS 1*/
91 
92 #include <sys/param.h>
93 #include <sys/systm.h>
94 #include <sys/kernel.h>
95 #include <sys/conf.h>
96 #include <sys/malloc.h>
97 #include <sys/mount_internal.h>
98 #include <sys/proc.h>
99 #include <sys/vnode_internal.h>
100 #include <stdarg.h>
101 #include <libkern/OSAtomic.h>
102 #include <os/refcnt.h>
103 #define BSD_KERNEL_PRIVATE      1       /* devfs_make_link() prototype */
104 #include "devfs.h"
105 #include "devfsdefs.h"
106 
107 #if CONFIG_MACF
108 #include <security/mac_framework.h>
109 #endif
110 
111 #if FDESC
112 #include "fdesc.h"
113 #endif
114 
115 typedef struct devfs_vnode_event {
116 	vnode_t                 dve_vp;
117 	uint32_t                dve_vid;
118 	uint32_t                dve_events;
119 } *devfs_vnode_event_t;
120 
121 /*
122  * Size of stack buffer (fast path) for notifications.  If
123  * the number of mounts is small, no need to malloc a buffer.
124  */
125 #define NUM_STACK_ENTRIES 5
126 
127 typedef struct devfs_event_log {
128 	size_t                  del_max;
129 	size_t                  del_used;
130 	devfs_vnode_event_t     del_entries;
131 } *devfs_event_log_t;
132 
133 
134 static void     dev_free_hier(devdirent_t *);
135 static int      devfs_propogate(devdirent_t *, devdirent_t *, devfs_event_log_t);
136 static int      dev_finddir(const char *, devnode_t *, int, devnode_t **, devfs_event_log_t);
137 static int      dev_dup_entry(devnode_t *, devdirent_t *, devdirent_t **, struct devfsmount *);
138 void            devfs_ref_node(devnode_t *);
139 void            devfs_rele_node(devnode_t *);
140 static void     devfs_record_event(devfs_event_log_t, devnode_t*, uint32_t);
141 static int      devfs_init_event_log(devfs_event_log_t, uint32_t, devfs_vnode_event_t);
142 static void     devfs_release_event_log(devfs_event_log_t, int);
143 static void     devfs_bulk_notify(devfs_event_log_t);
144 static devdirent_t *devfs_make_node_internal(dev_t, devfstype_t type, uid_t, gid_t, int,
145     int (*clone)(dev_t dev, int action), const char *fmt, va_list ap);
146 
147 
148 static LCK_GRP_DECLARE(devfs_lck_grp, "devfs_lock");
149 LCK_MTX_DECLARE(devfs_mutex, &devfs_lck_grp);
150 LCK_MTX_DECLARE(devfs_attr_mutex, &devfs_lck_grp);
151 
152 os_refgrp_decl(static, devfs_refgrp, "devfs", NULL);
153 
154 devdirent_t *           dev_root = NULL;        /* root of backing tree */
155 struct devfs_stats      devfs_stats;            /* hold stats */
156 
157 static ino_t            devfs_unique_fileno = 0;
158 
159 #ifdef HIDDEN_MOUNTPOINT
160 static struct mount *devfs_hidden_mount;
161 #endif /* HIDDEN_MOINTPOINT */
162 
163 static int devfs_ready = 0;
164 static uint32_t devfs_nmountplanes = 0; /* The first plane is not used for a mount */
165 
166 #define DEVFS_NOCREATE  FALSE
167 #define DEVFS_CREATE    TRUE
168 
169 /*
170  * Set up the root directory node in the backing plane
171  * This is happenning before the vfs system has been
172  * set up yet, so be careful about what we reference..
173  * Notice that the ops are by indirection.. as they haven't
174  * been set up yet!
175  * DEVFS has a hidden mountpoint that is used as the anchor point
176  * for the internal 'blueprint' version of the dev filesystem tree.
177  */
178 /*proto*/
179 int
devfs_sinit(void)180 devfs_sinit(void)
181 {
182 	int error;
183 
184 	DEVFS_LOCK();
185 	error = dev_add_entry("root", NULL, DEV_DIR, NULL, NULL, NULL, &dev_root);
186 	DEVFS_UNLOCK();
187 
188 	if (error) {
189 		printf("devfs_sinit: dev_add_entry failed ");
190 		return ENOTSUP;
191 	}
192 #ifdef HIDDEN_MOUNTPOINT
193 	devfs_hidden_mount = zalloc_flags(mount_zone, Z_WAITOK | Z_ZERO);
194 	mount_lock_init(devfs_hidden_mount);
195 	TAILQ_INIT(&devfs_hidden_mount->mnt_vnodelist);
196 	TAILQ_INIT(&devfs_hidden_mount->mnt_workerqueue);
197 	TAILQ_INIT(&devfs_hidden_mount->mnt_newvnodes);
198 #if CONFIG_MACF
199 	mac_mount_label_init(devfs_hidden_mount);
200 	mac_mount_label_associate(vfs_context_kernel(), devfs_hidden_mount);
201 #endif
202 
203 	/* Initialize the default IO constraints */
204 	mp->mnt_maxreadcnt = mp->mnt_maxwritecnt = MAXPHYS;
205 	mp->mnt_segreadcnt = mp->mnt_segwritecnt = 32;
206 	mp->mnt_ioflags = 0;
207 	mp->mnt_realrootvp = NULLVP;
208 	mp->mnt_authcache_ttl = CACHED_LOOKUP_RIGHT_TTL;
209 
210 	devfs_mount(devfs_hidden_mount, "dummy", NULL, NULL, NULL);
211 	dev_root->de_dnp->dn_dvm
212 	        = (struct devfsmount *)devfs_hidden_mount->mnt_data;
213 #endif /* HIDDEN_MOUNTPOINT */
214 #if CONFIG_MACF
215 	mac_devfs_label_associate_directory("/", (int) strlen("/"),
216 	    dev_root->de_dnp, "/");
217 #endif
218 	devfs_ready = 1;
219 	return 0;
220 }
221 
222 /***********************************************************************\
223 *************************************************************************
224 *	Routines used to find our way to a point in the tree		*
225 *************************************************************************
226 \***********************************************************************/
227 
228 
229 
230 /***************************************************************
231 * Search down the linked list off a dir to find "name"
232 * return the devnode_t * for that node.
233 *
234 * called with DEVFS_LOCK held
235 ***************************************************************/
236 devdirent_t *
dev_findname(devnode_t * dir,const char * name)237 dev_findname(devnode_t * dir, const char *name)
238 {
239 	devdirent_t * newfp;
240 	if (dir->dn_type != DEV_DIR) {
241 		return 0;                     /*XXX*/ /* printf?*/
242 	}
243 	if (name[0] == '.') {
244 		if (name[1] == 0) {
245 			return dir->dn_typeinfo.Dir.myname;
246 		}
247 		if ((name[1] == '.') && (name[2] == 0)) {
248 			/* for root, .. == . */
249 			return dir->dn_typeinfo.Dir.parent->dn_typeinfo.Dir.myname;
250 		}
251 	}
252 	newfp = dir->dn_typeinfo.Dir.dirlist;
253 
254 	while (newfp) {
255 		if (!(strncmp(name, newfp->de_name, sizeof(newfp->de_name)))) {
256 			return newfp;
257 		}
258 		newfp = newfp->de_next;
259 	}
260 	return NULL;
261 }
262 
263 /***********************************************************************
264 * Given a starting node (0 for root) and a pathname, return the node
265 * for the end item on the path. It MUST BE A DIRECTORY. If the 'DEVFS_CREATE'
266 * option is true, then create any missing nodes in the path and create
267 * and return the final node as well.
268 * This is used to set up a directory, before making nodes in it..
269 *
270 * called with DEVFS_LOCK held
271 ***********************************************************************/
272 static int
dev_finddir(const char * path,devnode_t * dirnode,int create,devnode_t ** dn_pp,devfs_event_log_t delp)273 dev_finddir(const char * path,
274     devnode_t * dirnode,
275     int create,
276     devnode_t * * dn_pp,
277     devfs_event_log_t delp)
278 {
279 	devnode_t *     dnp = NULL;
280 	int             error = 0;
281 	const char *            scan;
282 #if CONFIG_MACF
283 	char            fullpath[DEVMAXPATHSIZE];
284 #endif
285 
286 
287 	if (!dirnode) { /* dirnode == NULL means start at root */
288 		dirnode = dev_root->de_dnp;
289 	}
290 
291 	if (dirnode->dn_type != DEV_DIR) {
292 		return ENOTDIR;
293 	}
294 
295 	if (strlen(path) > (DEVMAXPATHSIZE - 1)) {
296 		return ENAMETOOLONG;
297 	}
298 
299 #if CONFIG_MACF
300 	strlcpy(fullpath, path, DEVMAXPATHSIZE);
301 #endif
302 	scan = path;
303 
304 	while (*scan == '/') {
305 		scan++;
306 	}
307 
308 	*dn_pp = NULL;
309 
310 	while (1) {
311 		char                component[DEVMAXPATHSIZE];
312 		devdirent_t *       dirent_p;
313 		const char *        start;
314 
315 		if (*scan == 0) {
316 			/* we hit the end of the string, we're done */
317 			*dn_pp = dirnode;
318 			break;
319 		}
320 		start = scan;
321 		while (*scan != '/' && *scan) {
322 			scan++;
323 		}
324 
325 		strlcpy(component, start, (scan - start) + 1);
326 		if (*scan == '/') {
327 			scan++;
328 		}
329 
330 		dirent_p = dev_findname(dirnode, component);
331 		if (dirent_p) {
332 			dnp = dirent_p->de_dnp;
333 			if (dnp->dn_type != DEV_DIR) {
334 				error = ENOTDIR;
335 				break;
336 			}
337 		} else {
338 			if (!create) {
339 				error = ENOENT;
340 				break;
341 			}
342 			error = dev_add_entry(component, dirnode,
343 			    DEV_DIR, NULL, NULL, NULL, &dirent_p);
344 			if (error) {
345 				break;
346 			}
347 			dnp = dirent_p->de_dnp;
348 #if CONFIG_MACF
349 			mac_devfs_label_associate_directory(
350 				dirnode->dn_typeinfo.Dir.myname->de_name,
351 				(int) strlen(dirnode->dn_typeinfo.Dir.myname->de_name),
352 				dnp, fullpath);
353 #endif
354 			devfs_propogate(dirnode->dn_typeinfo.Dir.myname, dirent_p, delp);
355 		}
356 		dirnode = dnp; /* continue relative to this directory */
357 	}
358 	return error;
359 }
360 
361 
362 /***********************************************************************
363 * Add a new NAME element to the devfs
364 * If we're creating a root node, then dirname is NULL
365 * Basically this creates a new namespace entry for the device node
366 *
367 * Creates a name node, and links it to the supplied node
368 *
369 * called with DEVFS_LOCK held
370 ***********************************************************************/
371 int
dev_add_name(const char * name,devnode_t * dirnode,__unused devdirent_t * back,devnode_t * dnp,devdirent_t ** dirent_pp)372 dev_add_name(const char * name, devnode_t * dirnode, __unused devdirent_t * back,
373     devnode_t * dnp, devdirent_t * *dirent_pp)
374 {
375 	devdirent_t *   dirent_p = NULL;
376 
377 	if (dirnode != NULL) {
378 		if (dirnode->dn_type != DEV_DIR) {
379 			return ENOTDIR;
380 		}
381 
382 		if (dev_findname(dirnode, name)) {
383 			return EEXIST;
384 		}
385 	}
386 	/*
387 	 * make sure the name is legal
388 	 * slightly misleading in the case of NULL
389 	 */
390 	if (!name || (strlen(name) > (DEVMAXNAMESIZE - 1))) {
391 		return ENAMETOOLONG;
392 	}
393 
394 	/*
395 	 * Allocate and fill out a new directory entry
396 	 */
397 	dirent_p = kalloc_type(devdirent_t, Z_WAITOK | Z_ZERO | Z_NOFAIL);
398 
399 	/* inherrit our parent's mount info */ /*XXX*/
400 	/* a kludge but.... */
401 	if (dirnode && (dnp->dn_dvm == NULL)) {
402 		dnp->dn_dvm = dirnode->dn_dvm;
403 		/* if(!dnp->dn_dvm) printf("parent had null dvm "); */
404 	}
405 
406 	/*
407 	 * Link the two together
408 	 * include the implicit link in the count of links to the devnode..
409 	 * this stops it from being accidentally freed later.
410 	 */
411 	dirent_p->de_dnp = dnp;
412 	dnp->dn_links++;  /* implicit from our own name-node */
413 
414 	/*
415 	 * Make sure that we can find all the links that reference a node
416 	 * so that we can get them all if we need to zap the node.
417 	 */
418 	if (dnp->dn_linklist) {
419 		dirent_p->de_nextlink = dnp->dn_linklist;
420 		dirent_p->de_prevlinkp = dirent_p->de_nextlink->de_prevlinkp;
421 		dirent_p->de_nextlink->de_prevlinkp = &(dirent_p->de_nextlink);
422 		*dirent_p->de_prevlinkp = dirent_p;
423 	} else {
424 		dirent_p->de_nextlink = dirent_p;
425 		dirent_p->de_prevlinkp = &(dirent_p->de_nextlink);
426 	}
427 	dnp->dn_linklist = dirent_p;
428 
429 	/*
430 	 * If the node is a directory, then we need to handle the
431 	 * creation of the .. link.
432 	 * A NULL dirnode indicates a root node, so point to ourself.
433 	 */
434 	if (dnp->dn_type == DEV_DIR) {
435 		dnp->dn_typeinfo.Dir.myname = dirent_p;
436 		/*
437 		 * If we are unlinking from an old dir, decrement its links
438 		 * as we point our '..' elsewhere
439 		 * Note: it's up to the calling code to remove the
440 		 * us from the original directory's list
441 		 */
442 		if (dnp->dn_typeinfo.Dir.parent) {
443 			dnp->dn_typeinfo.Dir.parent->dn_links--;
444 		}
445 		if (dirnode) {
446 			dnp->dn_typeinfo.Dir.parent = dirnode;
447 		} else {
448 			dnp->dn_typeinfo.Dir.parent = dnp;
449 		}
450 		dnp->dn_typeinfo.Dir.parent->dn_links++; /* account for the new '..' */
451 	}
452 
453 	/*
454 	 * put the name into the directory entry.
455 	 */
456 	strlcpy(dirent_p->de_name, name, DEVMAXNAMESIZE);
457 
458 
459 	/*
460 	 * Check if we are not making a root node..
461 	 * (i.e. have parent)
462 	 */
463 	if (dirnode) {
464 		/*
465 		 * Put it on the END of the linked list of directory entries
466 		 */
467 		dirent_p->de_parent = dirnode; /* null for root */
468 		dirent_p->de_prevp = dirnode->dn_typeinfo.Dir.dirlast;
469 		dirent_p->de_next = *(dirent_p->de_prevp); /* should be NULL */
470 		                                           /*right?*/
471 		*(dirent_p->de_prevp) = dirent_p;
472 		dirnode->dn_typeinfo.Dir.dirlast = &(dirent_p->de_next);
473 		dirnode->dn_typeinfo.Dir.entrycount++;
474 		dirnode->dn_len += strlen(name) + 8;/*ok, ok?*/
475 	}
476 
477 	*dirent_pp = dirent_p;
478 	DEVFS_INCR_ENTRIES();
479 	return 0;
480 }
481 
482 
483 /***********************************************************************
484 * Add a new element to the devfs plane.
485 *
486 * Creates a new dev_node to go with it if the prototype should not be
487 * reused. (Is a DIR, or we select SPLIT_DEVS at compile time)
488 * typeinfo gives us info to make our node if we don't have a prototype.
489 * If typeinfo is null and proto exists, then the typeinfo field of
490 * the proto is used intead in the DEVFS_CREATE case.
491 * note the 'links' count is 0 (except if a dir)
492 * but it is only cleared on a transition
493 * so this is ok till we link it to something
494 * Even in SPLIT_DEVS mode,
495 * if the node already exists on the wanted plane, just return it
496 *
497 * called with DEVFS_LOCK held
498 ***********************************************************************/
499 int
dev_add_node(int entrytype,devnode_type_t * typeinfo,devnode_t * proto,devnode_t ** dn_pp,struct devfsmount * dvm)500 dev_add_node(int entrytype, devnode_type_t * typeinfo, devnode_t * proto,
501     devnode_t * *dn_pp, struct devfsmount *dvm)
502 {
503 	devnode_t *     dnp = NULL;
504 	int     error = 0;
505 
506 #if defined SPLIT_DEVS
507 	/*
508 	 * If we have a prototype, then check if there is already a sibling
509 	 * on the mount plane we are looking at, if so, just return it.
510 	 */
511 	if (proto) {
512 		dnp = proto->dn_nextsibling;
513 		while (dnp != proto) {
514 			if (dnp->dn_dvm == dvm) {
515 				*dn_pp = dnp;
516 				return 0;
517 			}
518 			dnp = dnp->dn_nextsibling;
519 		}
520 		if (typeinfo == NULL) {
521 			typeinfo = &(proto->dn_typeinfo);
522 		}
523 	}
524 #else   /* SPLIT_DEVS */
525 	if (proto) {
526 		switch (proto->type) {
527 		case DEV_BDEV:
528 		case DEV_CDEV:
529 			*dn_pp = proto;
530 			return 0;
531 		}
532 	}
533 #endif  /* SPLIT_DEVS */
534 	MALLOC(dnp, devnode_t *, sizeof(devnode_t), M_DEVFSNODE, M_WAITOK);
535 	if (!dnp) {
536 		return ENOMEM;
537 	}
538 
539 	/*
540 	 * If we have a proto, that means that we are duplicating some
541 	 * other device, which can only happen if we are not at the back plane
542 	 */
543 	if (proto) {
544 		bcopy(proto, dnp, sizeof(devnode_t));
545 		dnp->dn_links = 0;
546 		dnp->dn_linklist = NULL;
547 		dnp->dn_vn = NULL;
548 		dnp->dn_len = 0;
549 		/* add to END of siblings list */
550 		dnp->dn_prevsiblingp = proto->dn_prevsiblingp;
551 		*(dnp->dn_prevsiblingp) = dnp;
552 		dnp->dn_nextsibling = proto;
553 		proto->dn_prevsiblingp = &(dnp->dn_nextsibling);
554 #if CONFIG_MACF
555 		mac_devfs_label_init(dnp);
556 		mac_devfs_label_copy(mac_devfs_label(proto), mac_devfs_label(dnp));
557 #endif
558 	} else {
559 		struct timeval tv;
560 
561 		/*
562 		 * We have no prototype, so start off with a clean slate
563 		 */
564 		microtime(&tv);
565 		bzero(dnp, sizeof(devnode_t));
566 		dnp->dn_type = entrytype;
567 		dnp->dn_nextsibling = dnp;
568 		dnp->dn_prevsiblingp = &(dnp->dn_nextsibling);
569 		dnp->dn_atime.tv_sec = tv.tv_sec;
570 		dnp->dn_mtime.tv_sec = tv.tv_sec;
571 		dnp->dn_ctime.tv_sec = tv.tv_sec;
572 #if CONFIG_MACF
573 		mac_devfs_label_init(dnp);
574 #endif
575 	}
576 	dnp->dn_dvm = dvm;
577 
578 	/* Note: this inits the reference count to 1, this is considered unreferenced */
579 	os_ref_init_raw(&dnp->dn_refcount, &devfs_refgrp);
580 	dnp->dn_ino = devfs_unique_fileno;
581 	devfs_unique_fileno++;
582 
583 	/*
584 	 * fill out the dev node according to type
585 	 */
586 	switch (entrytype) {
587 	case DEV_DIR:
588 		/*
589 		 * As it's a directory, make sure
590 		 * it has a null entries list
591 		 */
592 		dnp->dn_typeinfo.Dir.dirlast = &(dnp->dn_typeinfo.Dir.dirlist);
593 		dnp->dn_typeinfo.Dir.dirlist = (devdirent_t *)0;
594 		dnp->dn_typeinfo.Dir.entrycount = 0;
595 		/*  until we know better, it has a null parent pointer*/
596 		dnp->dn_typeinfo.Dir.parent = NULL;
597 		dnp->dn_links++; /* for .*/
598 		dnp->dn_typeinfo.Dir.myname = NULL;
599 		/*
600 		 * make sure that the ops associated with it are the ops
601 		 * that we use (by default) for directories
602 		 */
603 		dnp->dn_ops = &devfs_vnodeop_p;
604 		dnp->dn_mode |= 0555;   /* default perms */
605 		break;
606 	case DEV_SLNK:
607 		/*
608 		 * As it's a symlink allocate and store the link info
609 		 * Symlinks should only ever be created by the user,
610 		 * so they are not on the back plane and should not be
611 		 * propogated forward.. a bit like directories in that way..
612 		 * A symlink only exists on one plane and has its own
613 		 * node.. therefore we might be on any random plane.
614 		 */
615 		dnp->dn_typeinfo.Slnk.name = kalloc_data(typeinfo->Slnk.namelen + 1, Z_WAITOK);
616 		if (!dnp->dn_typeinfo.Slnk.name) {
617 			error = ENOMEM;
618 			break;
619 		}
620 		strlcpy(dnp->dn_typeinfo.Slnk.name, typeinfo->Slnk.name,
621 		    typeinfo->Slnk.namelen + 1);
622 		dnp->dn_typeinfo.Slnk.namelen = typeinfo->Slnk.namelen;
623 		DEVFS_INCR_STRINGSPACE(dnp->dn_typeinfo.Slnk.namelen + 1);
624 		dnp->dn_ops = &devfs_vnodeop_p;
625 		dnp->dn_mode |= 0555;   /* default perms */
626 		break;
627 	case DEV_CDEV:
628 	case DEV_BDEV:
629 		/*
630 		 * Make sure it has DEVICE type ops
631 		 * and device specific fields are correct
632 		 */
633 		dnp->dn_ops = &devfs_spec_vnodeop_p;
634 		dnp->dn_typeinfo.dev = typeinfo->dev;
635 		break;
636 
637 	#if FDESC
638 	/* /dev/fd is special */
639 	case DEV_DEVFD:
640 		dnp->dn_ops = &devfs_devfd_vnodeop_p;
641 		dnp->dn_mode |= 0555;   /* default perms */
642 		break;
643 
644 	#endif /* FDESC */
645 	default:
646 		error = EINVAL;
647 	}
648 
649 	if (error) {
650 		FREE(dnp, M_DEVFSNODE);
651 	} else {
652 		*dn_pp = dnp;
653 		DEVFS_INCR_NODES();
654 	}
655 
656 	return error;
657 }
658 
659 
660 /***********************************************************************
661  * called with DEVFS_LOCK held
662  **********************************************************************/
663 void
devnode_free(devnode_t * dnp)664 devnode_free(devnode_t * dnp)
665 {
666 #if CONFIG_MACF
667 	mac_devfs_label_destroy(dnp);
668 #endif
669 	if (dnp->dn_type == DEV_SLNK) {
670 		DEVFS_DECR_STRINGSPACE(dnp->dn_typeinfo.Slnk.namelen + 1);
671 		kfree_data(dnp->dn_typeinfo.Slnk.name, dnp->dn_typeinfo.Slnk.namelen + 1);
672 	}
673 	DEVFS_DECR_NODES();
674 	FREE(dnp, M_DEVFSNODE);
675 }
676 
677 
678 /***********************************************************************
679  * called with DEVFS_LOCK held
680  **********************************************************************/
681 static void
devfs_dn_free(devnode_t * dnp)682 devfs_dn_free(devnode_t * dnp)
683 {
684 	if (--dnp->dn_links <= 0) { /* can be -1 for initial free, on error */
685 		/*probably need to do other cleanups XXX */
686 		if (dnp->dn_nextsibling != dnp) {
687 			devnode_t * *   prevp = dnp->dn_prevsiblingp;
688 			*prevp = dnp->dn_nextsibling;
689 			dnp->dn_nextsibling->dn_prevsiblingp = prevp;
690 		}
691 
692 		/* Can only free if there are no references; otherwise, wait for last vnode to be reclaimed */
693 		os_ref_count_t rc = os_ref_get_count_raw(&dnp->dn_refcount);
694 		if (rc == 1) {
695 			/* release final reference from dev_add_node */
696 			(void) os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
697 			devnode_free(dnp);
698 		} else {
699 			dnp->dn_lflags |= DN_DELETE;
700 		}
701 	}
702 }
703 
704 /***********************************************************************\
705 *	Front Node Operations						*
706 *	Add or delete a chain of front nodes				*
707 \***********************************************************************/
708 
709 
710 /***********************************************************************
711 * Given a directory backing node, and a child backing node, add the
712 * appropriate front nodes to the front nodes of the directory to
713 * represent the child node to the user
714 *
715 * on failure, front nodes will either be correct or not exist for each
716 * front dir, however dirs completed will not be stripped of completed
717 * frontnodes on failure of a later frontnode
718 *
719 * This allows a new node to be propogated through all mounted planes
720 *
721 * called with DEVFS_LOCK held
722 ***********************************************************************/
723 static int
devfs_propogate(devdirent_t * parent,devdirent_t * child,devfs_event_log_t delp)724 devfs_propogate(devdirent_t * parent, devdirent_t * child, devfs_event_log_t delp)
725 {
726 	int     error;
727 	devdirent_t * newnmp;
728 	devnode_t *     dnp = child->de_dnp;
729 	devnode_t *     pdnp = parent->de_dnp;
730 	devnode_t *     adnp = parent->de_dnp;
731 	int type = child->de_dnp->dn_type;
732 	uint32_t events;
733 
734 	events = (dnp->dn_type == DEV_DIR ? VNODE_EVENT_DIR_CREATED : VNODE_EVENT_FILE_CREATED);
735 	if (delp != NULL) {
736 		devfs_record_event(delp, pdnp, events);
737 	}
738 
739 	/***********************************************
740 	* Find the other instances of the parent node
741 	***********************************************/
742 	for (adnp = pdnp->dn_nextsibling;
743 	    adnp != pdnp;
744 	    adnp = adnp->dn_nextsibling) {
745 		/*
746 		 * Make the node, using the original as a prototype)
747 		 * if the node already exists on that plane it won't be
748 		 * re-made..
749 		 */
750 		if ((error = dev_add_entry(child->de_name, adnp, type,
751 		    NULL, dnp, adnp->dn_dvm,
752 		    &newnmp)) != 0) {
753 			printf("duplicating %s failed\n", child->de_name);
754 		} else {
755 			if (delp != NULL) {
756 				devfs_record_event(delp, adnp, events);
757 
758 				/*
759 				 * Slightly subtle.  We're guaranteed that there will
760 				 * only be a vnode hooked into this devnode if we're creating
761 				 * a new link to an existing node; otherwise, the devnode is new
762 				 * and no one can have looked it up yet. If we're making a link,
763 				 * then the buffer is large enough for two nodes in each
764 				 * plane; otherwise, there's no vnode and this call will
765 				 * do nothing.
766 				 */
767 				devfs_record_event(delp, newnmp->de_dnp, VNODE_EVENT_LINK);
768 			}
769 		}
770 	}
771 	return 0;       /* for now always succeed */
772 }
773 
774 static uint32_t
remove_notify_count(devnode_t * dnp)775 remove_notify_count(devnode_t *dnp)
776 {
777 	uint32_t notify_count = 0;
778 	devnode_t *dnp2;
779 
780 	/*
781 	 * Could need to notify for one removed node on each mount and
782 	 * one parent for each such node.
783 	 */
784 	notify_count = devfs_nmountplanes;
785 	notify_count += dnp->dn_links;
786 	for (dnp2 = dnp->dn_nextsibling; dnp2 != dnp; dnp2 = dnp2->dn_nextsibling) {
787 		notify_count += dnp2->dn_links;
788 	}
789 
790 	return notify_count;
791 }
792 
793 /***********************************************************************
794 * remove all instances of this devicename [for backing nodes..]
795 * note.. if there is another link to the node (non dir nodes only)
796 * then the devfs_node will still exist as the ref count will be non-0
797 * removing a directory node will remove all sup-nodes on all planes (ZAP)
798 *
799 * Used by device drivers to remove nodes that are no longer relevant
800 * The argument is the 'cookie' they were given when they created the node
801 * this function is exported.. see devfs.h
802 ***********************************************************************/
803 void
devfs_remove(void * dirent_p)804 devfs_remove(void *dirent_p)
805 {
806 	devnode_t * dnp = ((devdirent_t *)dirent_p)->de_dnp;
807 	devnode_t * dnp2;
808 	boolean_t   lastlink;
809 	struct devfs_event_log event_log;
810 	uint32_t    log_count = 0;
811 	int         do_notify = 0;
812 	int         need_free = 0;
813 	struct devfs_vnode_event stackbuf[NUM_STACK_ENTRIES];
814 
815 	DEVFS_LOCK();
816 
817 	if (!devfs_ready) {
818 		printf("devfs_remove: not ready for devices!\n");
819 		goto out;
820 	}
821 
822 	log_count = remove_notify_count(dnp);
823 
824 	if (log_count > NUM_STACK_ENTRIES) {
825 		uint32_t new_count;
826 wrongsize:
827 		DEVFS_UNLOCK();
828 		if (devfs_init_event_log(&event_log, log_count, NULL) == 0) {
829 			do_notify = 1;
830 			need_free = 1;
831 		}
832 		DEVFS_LOCK();
833 
834 		new_count = remove_notify_count(dnp);
835 		if (need_free && (new_count > log_count)) {
836 			devfs_release_event_log(&event_log, 1);
837 			need_free = 0;
838 			do_notify = 0;
839 			log_count = log_count * 2;
840 			goto wrongsize;
841 		}
842 	} else {
843 		if (devfs_init_event_log(&event_log, NUM_STACK_ENTRIES, &stackbuf[0]) == 0) {
844 			do_notify = 1;
845 		}
846 	}
847 
848 	/* This file has been deleted */
849 	if (do_notify != 0) {
850 		devfs_record_event(&event_log, dnp, VNODE_EVENT_DELETE);
851 	}
852 
853 	/* keep removing the next sibling till only we exist. */
854 	while ((dnp2 = dnp->dn_nextsibling) != dnp) {
855 		/*
856 		 * Keep removing the next front node till no more exist
857 		 */
858 		dnp->dn_nextsibling = dnp2->dn_nextsibling;
859 		dnp->dn_nextsibling->dn_prevsiblingp = &(dnp->dn_nextsibling);
860 		dnp2->dn_nextsibling = dnp2;
861 		dnp2->dn_prevsiblingp = &(dnp2->dn_nextsibling);
862 
863 		/* This file has been deleted in this plane */
864 		if (do_notify != 0) {
865 			devfs_record_event(&event_log, dnp2, VNODE_EVENT_DELETE);
866 		}
867 
868 		if (dnp2->dn_linklist) {
869 			do {
870 				lastlink = (1 == dnp2->dn_links);
871 				/* Each parent of a link to this file has lost a child in this plane */
872 				if (do_notify != 0) {
873 					devfs_record_event(&event_log, dnp2->dn_linklist->de_parent, VNODE_EVENT_FILE_REMOVED);
874 				}
875 				dev_free_name(dnp2->dn_linklist);
876 			} while (!lastlink);
877 		}
878 	}
879 
880 	/*
881 	 * then free the main node
882 	 * If we are not running in SPLIT_DEVS mode, then
883 	 * THIS is what gets rid of the propogated nodes.
884 	 */
885 	if (dnp->dn_linklist) {
886 		do {
887 			lastlink = (1 == dnp->dn_links);
888 			/* Each parent of a link to this file has lost a child */
889 			if (do_notify != 0) {
890 				devfs_record_event(&event_log, dnp->dn_linklist->de_parent, VNODE_EVENT_FILE_REMOVED);
891 			}
892 			dev_free_name(dnp->dn_linklist);
893 		} while (!lastlink);
894 	}
895 out:
896 	DEVFS_UNLOCK();
897 	if (do_notify != 0) {
898 		devfs_bulk_notify(&event_log);
899 		devfs_release_event_log(&event_log, need_free);
900 	}
901 
902 	return;
903 }
904 
905 
906 
907 /***************************************************************
908  * duplicate the backing tree into a tree of nodes hung off the
909  * mount point given as the argument. Do this by
910  * calling dev_dup_entry which recurses all the way
911  * up the tree..
912  *
913  * called with DEVFS_LOCK held
914  **************************************************************/
915 int
dev_dup_plane(struct devfsmount * devfs_mp_p)916 dev_dup_plane(struct devfsmount *devfs_mp_p)
917 {
918 	devdirent_t *   new;
919 	int             error = 0;
920 
921 	if ((error = dev_dup_entry(NULL, dev_root, &new, devfs_mp_p))) {
922 		return error;
923 	}
924 	devfs_mp_p->plane_root = new;
925 	devfs_nmountplanes++;
926 	return error;
927 }
928 
929 
930 
931 /***************************************************************
932 * Free a whole plane
933 *
934 * called with DEVFS_LOCK held
935 ***************************************************************/
936 void
devfs_free_plane(struct devfsmount * devfs_mp_p)937 devfs_free_plane(struct devfsmount *devfs_mp_p)
938 {
939 	devdirent_t * dirent_p;
940 
941 	dirent_p = devfs_mp_p->plane_root;
942 	if (dirent_p) {
943 		dev_free_hier(dirent_p);
944 		dev_free_name(dirent_p);
945 	}
946 	devfs_mp_p->plane_root = NULL;
947 	devfs_nmountplanes--;
948 
949 	if (devfs_nmountplanes > (devfs_nmountplanes + 1)) {
950 		panic("plane count wrapped around.");
951 	}
952 }
953 
954 
955 /***************************************************************
956 * Create and link in a new front element..
957 * Parent can be 0 for a root node
958 * Not presently usable to make a symlink XXX
959 * (Ok, symlinks don't propogate)
960 * recursively will create subnodes corresponding to equivalent
961 * child nodes in the base level
962 *
963 * called with DEVFS_LOCK held
964 ***************************************************************/
965 static int
dev_dup_entry(devnode_t * parent,devdirent_t * back,devdirent_t ** dnm_pp,struct devfsmount * dvm)966 dev_dup_entry(devnode_t * parent, devdirent_t * back, devdirent_t * *dnm_pp,
967     struct devfsmount *dvm)
968 {
969 	devdirent_t *   entry_p = NULL;
970 	devdirent_t *   newback;
971 	devdirent_t *   newfront;
972 	int     error;
973 	devnode_t *     dnp = back->de_dnp;
974 	int type = dnp->dn_type;
975 
976 	/*
977 	 * go get the node made (if we need to)
978 	 * use the back one as a prototype
979 	 */
980 	error = dev_add_entry(back->de_name, parent, type, NULL, dnp,
981 	    parent?parent->dn_dvm:dvm, &entry_p);
982 	if (!error && (entry_p == NULL)) {
983 		error = ENOMEM; /* Really can't happen, but make static analyzer happy */
984 	}
985 	if (error != 0) {
986 		printf("duplicating %s failed\n", back->de_name);
987 		goto out;
988 	}
989 
990 	/*
991 	 * If we have just made the root, then insert the pointer to the
992 	 * mount information
993 	 */
994 	if (dvm) {
995 		entry_p->de_dnp->dn_dvm = dvm;
996 	}
997 
998 	/*
999 	 * If it is a directory, then recurse down all the other
1000 	 * subnodes in it....
1001 	 * note that this time we don't pass on the mount info..
1002 	 */
1003 	if (type == DEV_DIR) {
1004 		for (newback = back->de_dnp->dn_typeinfo.Dir.dirlist;
1005 		    newback; newback = newback->de_next) {
1006 			if ((error = dev_dup_entry(entry_p->de_dnp,
1007 			    newback, &newfront, NULL)) != 0) {
1008 				break; /* back out with an error */
1009 			}
1010 		}
1011 	}
1012 out:
1013 	*dnm_pp = entry_p;
1014 	return error;
1015 }
1016 
1017 
1018 /***************************************************************
1019 * Free a name node
1020 * remember that if there are other names pointing to the
1021 * dev_node then it may not get freed yet
1022 * can handle if there is no dnp
1023 *
1024 * called with DEVFS_LOCK held
1025 ***************************************************************/
1026 
1027 int
dev_free_name(devdirent_t * dirent_p)1028 dev_free_name(devdirent_t * dirent_p)
1029 {
1030 	devnode_t *     parent = dirent_p->de_parent;
1031 	devnode_t *     dnp = dirent_p->de_dnp;
1032 
1033 	if (dnp) {
1034 		if (dnp->dn_type == DEV_DIR) {
1035 			devnode_t * p;
1036 
1037 			if (dnp->dn_typeinfo.Dir.dirlist) {
1038 				return ENOTEMPTY;
1039 			}
1040 			p = dnp->dn_typeinfo.Dir.parent;
1041 			devfs_dn_free(dnp);     /* account for '.' */
1042 			devfs_dn_free(p);       /* '..' */
1043 		}
1044 		/*
1045 		 * unlink us from the list of links for this node
1046 		 * If we are the only link, it's easy!
1047 		 * if we are a DIR of course there should not be any
1048 		 * other links.
1049 		 */
1050 		if (dirent_p->de_nextlink == dirent_p) {
1051 			dnp->dn_linklist = NULL;
1052 		} else {
1053 			if (dnp->dn_linklist == dirent_p) {
1054 				dnp->dn_linklist = dirent_p->de_nextlink;
1055 			}
1056 		}
1057 		devfs_dn_free(dnp);
1058 	}
1059 
1060 	dirent_p->de_nextlink->de_prevlinkp = dirent_p->de_prevlinkp;
1061 	*(dirent_p->de_prevlinkp) = dirent_p->de_nextlink;
1062 
1063 	/*
1064 	 * unlink ourselves from the directory on this plane
1065 	 */
1066 	if (parent) { /* if not fs root */
1067 		if ((*dirent_p->de_prevp = dirent_p->de_next)) {/* yes, assign */
1068 			dirent_p->de_next->de_prevp = dirent_p->de_prevp;
1069 		} else {
1070 			parent->dn_typeinfo.Dir.dirlast
1071 			        = dirent_p->de_prevp;
1072 		}
1073 		parent->dn_typeinfo.Dir.entrycount--;
1074 		parent->dn_len -= strlen(dirent_p->de_name) + 8;
1075 	}
1076 
1077 	DEVFS_DECR_ENTRIES();
1078 	kfree_type(devdirent_t, dirent_p);
1079 	return 0;
1080 }
1081 
1082 
1083 /***************************************************************
1084 * Free a hierarchy starting at a directory node name
1085 * remember that if there are other names pointing to the
1086 * dev_node then it may not get freed yet
1087 * can handle if there is no dnp
1088 * leave the node itself allocated.
1089 *
1090 * called with DEVFS_LOCK held
1091 ***************************************************************/
1092 
1093 static void
dev_free_hier(devdirent_t * dirent_p)1094 dev_free_hier(devdirent_t * dirent_p)
1095 {
1096 	devnode_t *     dnp = dirent_p->de_dnp;
1097 
1098 	if (dnp) {
1099 		if (dnp->dn_type == DEV_DIR) {
1100 			while (dnp->dn_typeinfo.Dir.dirlist) {
1101 				dev_free_hier(dnp->dn_typeinfo.Dir.dirlist);
1102 				dev_free_name(dnp->dn_typeinfo.Dir.dirlist);
1103 			}
1104 		}
1105 	}
1106 }
1107 
1108 
1109 /***************************************************************
1110  * given a dev_node, find the appropriate vnode if one is already
1111  * associated, or get a new one and associate it with the dev_node
1112  *
1113  * called with DEVFS_LOCK held
1114  *
1115  * If an error is returned, then the dnp may have been freed (we
1116  * raced with a delete and lost).  A devnode should not be accessed
1117  * after devfs_dntovn() fails.
1118  ****************************************************************/
1119 int
devfs_dntovn(devnode_t * dnp,struct vnode ** vn_pp,__unused struct proc * p)1120 devfs_dntovn(devnode_t * dnp, struct vnode **vn_pp, __unused struct proc * p)
1121 {
1122 	struct vnode *vn_p;
1123 	int error = 0;
1124 	struct vnode_fsparam vfsp;
1125 	enum vtype vtype = 0;
1126 	int markroot = 0;
1127 	int nretries = 0;
1128 	int n_minor = DEVFS_CLONE_ALLOC; /* new minor number for clone device */
1129 
1130 	/*
1131 	 * We should never come in and find that our devnode has been marked for delete.
1132 	 * The lookup should have held the lock from entry until now; it should not have
1133 	 * been able to find a removed entry. Any other pathway would have just created
1134 	 * the devnode and come here without dropping the devfs lock, so no one would
1135 	 * have a chance to delete.
1136 	 */
1137 	if (dnp->dn_lflags & DN_DELETE) {
1138 		panic("devfs_dntovn: DN_DELETE set on a devnode upon entry.");
1139 	}
1140 
1141 	devfs_ref_node(dnp);
1142 
1143 retry:
1144 	*vn_pp = NULL;
1145 	vn_p = dnp->dn_vn;
1146 
1147 	if (vn_p) { /* already has a vnode */
1148 		uint32_t vid;
1149 
1150 		vid = vnode_vid(vn_p);
1151 
1152 		DEVFS_UNLOCK();
1153 
1154 		/*
1155 		 * We want to use the drainok variant of vnode_getwithvid
1156 		 * because we _don't_ want to get an iocount if the vnode is
1157 		 * is blocked in vnode_drain as it can cause infinite
1158 		 * loops in vn_open_auth. While in use vnodes are typically
1159 		 * only reclaimed on forced unmounts, In use devfs tty vnodes
1160 		 * can  be quite frequently reclaimed by revoke(2) or by the
1161 		 * exit of a controlling process.
1162 		 */
1163 		error = vnode_getwithvid_drainok(vn_p, vid);
1164 
1165 		DEVFS_LOCK();
1166 
1167 		if (dnp->dn_lflags & DN_DELETE) {
1168 			/*
1169 			 * our BUSY node got marked for
1170 			 * deletion while the DEVFS lock
1171 			 * was dropped...
1172 			 */
1173 			if (error == 0) {
1174 				/*
1175 				 * vnode_getwithvid returned a valid ref
1176 				 * which we need to drop
1177 				 */
1178 				vnode_put(vn_p);
1179 			}
1180 
1181 			/*
1182 			 * This entry is no longer in the namespace.  This is only
1183 			 * possible for lookup: no other path would not find an existing
1184 			 * vnode.  Therefore, ENOENT is a valid result.
1185 			 */
1186 			error = ENOENT;
1187 		} else if (error == ENODEV) {
1188 			/*
1189 			 * The Filesystem is getting unmounted.
1190 			 */
1191 			error = ENOENT;
1192 		} else if (error && (nretries < DEV_MAX_VNODE_RETRY)) {
1193 			/*
1194 			 * If we got an error from vnode_getwithvid, it means
1195 			 * we raced with a recycle and lost i.e. we asked for
1196 			 * an iocount only after vnode_drain had been entered
1197 			 * for the vnode and returned with an error only after
1198 			 * devfs_reclaim was called on the vnode.  devfs_reclaim
1199 			 * sets dn_vn to NULL but while we were waiting to
1200 			 * reacquire DEVFS_LOCK, another vnode might have gotten
1201 			 * associated with the dnp. In either case, we need to
1202 			 * retry otherwise we will end up returning an ENOENT
1203 			 * for this lookup but the next lookup will  succeed
1204 			 * because it creates a new vnode (or a racing  lookup
1205 			 * created a new vnode already).
1206 			 */
1207 			error = 0;
1208 			nretries++;
1209 			goto retry;
1210 		}
1211 		if (!error) {
1212 			*vn_pp = vn_p;
1213 		}
1214 
1215 		goto out;
1216 	}
1217 
1218 	/*
1219 	 * If we get here, then we've beaten any deletes;
1220 	 * if someone sets DN_DELETE during a subsequent drop
1221 	 * of the devfs lock, we'll still vend a vnode.
1222 	 */
1223 
1224 	if (dnp->dn_lflags & DN_CREATE) {
1225 		dnp->dn_lflags |= DN_CREATEWAIT;
1226 		msleep(&dnp->dn_lflags, &devfs_mutex, PRIBIO, 0, 0);
1227 		goto retry;
1228 	}
1229 
1230 	dnp->dn_lflags |= DN_CREATE;
1231 
1232 	switch (dnp->dn_type) {
1233 	case    DEV_SLNK:
1234 		vtype = VLNK;
1235 		break;
1236 	case    DEV_DIR:
1237 		if (dnp->dn_typeinfo.Dir.parent == dnp) {
1238 			markroot = 1;
1239 		}
1240 		vtype = VDIR;
1241 		break;
1242 	case    DEV_BDEV:
1243 	case    DEV_CDEV:
1244 		vtype = (dnp->dn_type == DEV_BDEV) ? VBLK : VCHR;
1245 		break;
1246 #if FDESC
1247 	case    DEV_DEVFD:
1248 		vtype = VDIR;
1249 		break;
1250 #endif /* FDESC */
1251 	}
1252 	vfsp.vnfs_mp = dnp->dn_dvm->mount;
1253 	vfsp.vnfs_vtype = vtype;
1254 	vfsp.vnfs_str = "devfs";
1255 	vfsp.vnfs_dvp = 0;
1256 	vfsp.vnfs_fsnode = dnp;
1257 	vfsp.vnfs_cnp = 0;
1258 	vfsp.vnfs_vops = *(dnp->dn_ops);
1259 
1260 	if (vtype == VBLK || vtype == VCHR) {
1261 		/*
1262 		 * Ask the clone minor number function for a new minor number
1263 		 * to use for the next device instance.  If an administative
1264 		 * limit has been reached, this function will return -1.
1265 		 */
1266 		if (dnp->dn_clone != NULL) {
1267 			int     n_major = major(dnp->dn_typeinfo.dev);
1268 
1269 			n_minor = (*dnp->dn_clone)(dnp->dn_typeinfo.dev, DEVFS_CLONE_ALLOC);
1270 			if (n_minor == -1) {
1271 				error = ENOMEM;
1272 				goto out;
1273 			}
1274 
1275 			vfsp.vnfs_rdev = makedev(n_major, n_minor);
1276 		} else {
1277 			vfsp.vnfs_rdev = dnp->dn_typeinfo.dev;
1278 		}
1279 	} else {
1280 		vfsp.vnfs_rdev = 0;
1281 	}
1282 	vfsp.vnfs_filesize = 0;
1283 	vfsp.vnfs_flags = VNFS_NOCACHE | VNFS_CANTCACHE;
1284 	/* Tag system files */
1285 	vfsp.vnfs_marksystem = 0;
1286 	vfsp.vnfs_markroot = markroot;
1287 
1288 	DEVFS_UNLOCK();
1289 
1290 	error = vnode_create(VNCREATE_FLAVOR, VCREATESIZE, &vfsp, &vn_p);
1291 
1292 	/* Do this before grabbing the lock */
1293 	if (error == 0) {
1294 		vnode_setneedinactive(vn_p);
1295 	}
1296 
1297 	DEVFS_LOCK();
1298 
1299 	if (error == 0) {
1300 		vnode_settag(vn_p, VT_DEVFS);
1301 
1302 		if ((dnp->dn_clone != NULL) && (dnp->dn_vn != NULLVP)) {
1303 			panic("devfs_dntovn: cloning device with a vnode?");
1304 		}
1305 
1306 		*vn_pp = vn_p;
1307 
1308 		/*
1309 		 * Another vnode that has this devnode as its v_data.
1310 		 * This reference, unlike the one taken at the start
1311 		 * of the function, persists until a VNOP_RECLAIM
1312 		 * comes through for this vnode.
1313 		 */
1314 		devfs_ref_node(dnp);
1315 
1316 		/*
1317 		 * A cloned vnode is not hooked into the devnode; every lookup
1318 		 * gets a new vnode.
1319 		 */
1320 		if (dnp->dn_clone == NULL) {
1321 			dnp->dn_vn = vn_p;
1322 		}
1323 	} else if (n_minor != DEVFS_CLONE_ALLOC) {
1324 		/*
1325 		 * If we failed the create, we need to release the cloned minor
1326 		 * back to the free list.  In general, this is only useful if
1327 		 * the clone function results in a state change in the cloned
1328 		 * device for which the minor number was obtained.  If we get
1329 		 * past this point withouth falling into this case, it's
1330 		 * assumed that any state to be released will be released when
1331 		 * the vnode is dropped, instead.
1332 		 */
1333 		(void)(*dnp->dn_clone)(dnp->dn_typeinfo.dev, DEVFS_CLONE_FREE);
1334 	}
1335 
1336 	dnp->dn_lflags &= ~DN_CREATE;
1337 	if (dnp->dn_lflags & DN_CREATEWAIT) {
1338 		dnp->dn_lflags &= ~DN_CREATEWAIT;
1339 		wakeup(&dnp->dn_lflags);
1340 	}
1341 
1342 out:
1343 	/*
1344 	 * Release the reference we took to prevent deletion while we weren't holding the lock.
1345 	 * If not returning success, then dropping this reference could delete the devnode;
1346 	 * no one should access a devnode after a call to devfs_dntovn fails.
1347 	 */
1348 	devfs_rele_node(dnp);
1349 
1350 	return error;
1351 }
1352 
1353 /*
1354  * Increment refcount on a devnode; prevents free of the node
1355  * while the devfs lock is not held.
1356  */
1357 void
devfs_ref_node(devnode_t * dnp)1358 devfs_ref_node(devnode_t *dnp)
1359 {
1360 	os_ref_retain_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1361 }
1362 
1363 /*
1364  * Release a reference on a devnode.  If the devnode is marked for
1365  * free and the refcount is dropped to one, do the free.
1366  */
1367 void
devfs_rele_node(devnode_t * dnp)1368 devfs_rele_node(devnode_t *dnp)
1369 {
1370 	os_ref_count_t rc = os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1371 	if (rc < 1) {
1372 		panic("devfs_rele_node: devnode without a refcount!");
1373 	} else if ((rc == 1) && (dnp->dn_lflags & DN_DELETE)) {
1374 		/* release final reference from dev_add_node */
1375 		(void) os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1376 		devnode_free(dnp);
1377 	}
1378 }
1379 
1380 /***********************************************************************
1381 * add a whole device, with no prototype.. make name element and node
1382 * Used for adding the original device entries
1383 *
1384 * called with DEVFS_LOCK held
1385 ***********************************************************************/
1386 int
dev_add_entry(const char * name,devnode_t * parent,int type,devnode_type_t * typeinfo,devnode_t * proto,struct devfsmount * dvm,devdirent_t ** nm_pp)1387 dev_add_entry(const char *name, devnode_t * parent, int type, devnode_type_t * typeinfo,
1388     devnode_t * proto, struct devfsmount *dvm, devdirent_t * *nm_pp)
1389 {
1390 	devnode_t *     dnp;
1391 	int     error = 0;
1392 
1393 	if ((error = dev_add_node(type, typeinfo, proto, &dnp,
1394 	    (parent?parent->dn_dvm:dvm))) != 0) {
1395 		printf("devfs: %s: base node allocation failed (Errno=%d)\n",
1396 		    name, error);
1397 		return error;
1398 	}
1399 	if ((error = dev_add_name(name, parent, NULL, dnp, nm_pp)) != 0) {
1400 		devfs_dn_free(dnp); /* 1->0 for dir, 0->(-1) for other */
1401 		printf("devfs: %s: name slot allocation failed (Errno=%d)\n",
1402 		    name, error);
1403 	}
1404 	return error;
1405 }
1406 
1407 static void
devfs_bulk_notify(devfs_event_log_t delp)1408 devfs_bulk_notify(devfs_event_log_t delp)
1409 {
1410 	uint32_t i;
1411 	for (i = 0; i < delp->del_used; i++) {
1412 		devfs_vnode_event_t dvep = &delp->del_entries[i];
1413 		if (vnode_getwithvid(dvep->dve_vp, dvep->dve_vid) == 0) {
1414 			vnode_notify(dvep->dve_vp, dvep->dve_events, NULL);
1415 			vnode_put(dvep->dve_vp);
1416 		}
1417 	}
1418 }
1419 
1420 static void
devfs_record_event(devfs_event_log_t delp,devnode_t * dnp,uint32_t events)1421 devfs_record_event(devfs_event_log_t delp, devnode_t *dnp, uint32_t events)
1422 {
1423 	if (delp->del_used >= delp->del_max) {
1424 		panic("devfs event log overflowed.");
1425 	}
1426 
1427 	/* Can only notify for nodes that have an associated vnode */
1428 	if (dnp->dn_vn != NULLVP && vnode_ismonitored(dnp->dn_vn)) {
1429 		devfs_vnode_event_t dvep = &delp->del_entries[delp->del_used];
1430 		dvep->dve_vp = dnp->dn_vn;
1431 		dvep->dve_vid = vnode_vid(dnp->dn_vn);
1432 		dvep->dve_events = events;
1433 		delp->del_used++;
1434 	}
1435 }
1436 
1437 static int
devfs_init_event_log(devfs_event_log_t delp,uint32_t count,devfs_vnode_event_t buf)1438 devfs_init_event_log(devfs_event_log_t delp, uint32_t count, devfs_vnode_event_t buf)
1439 {
1440 	devfs_vnode_event_t dvearr;
1441 
1442 	if (buf == NULL) {
1443 		MALLOC(dvearr, devfs_vnode_event_t, count * sizeof(struct devfs_vnode_event), M_TEMP, M_WAITOK | M_ZERO);
1444 		if (dvearr == NULL) {
1445 			return ENOMEM;
1446 		}
1447 	} else {
1448 		dvearr = buf;
1449 	}
1450 
1451 	delp->del_max = count;
1452 	delp->del_used = 0;
1453 	delp->del_entries = dvearr;
1454 	return 0;
1455 }
1456 
1457 static void
devfs_release_event_log(devfs_event_log_t delp,int need_free)1458 devfs_release_event_log(devfs_event_log_t delp, int need_free)
1459 {
1460 	if (delp->del_entries == NULL) {
1461 		panic("Free of devfs notify info that has not been intialized.");
1462 	}
1463 
1464 	if (need_free) {
1465 		FREE(delp->del_entries, M_TEMP);
1466 	}
1467 
1468 	delp->del_entries = NULL;
1469 }
1470 
1471 /*
1472  * Function: devfs_make_node
1473  *
1474  * Purpose
1475  *   Create a device node with the given pathname in the devfs namespace.
1476  *
1477  * Parameters:
1478  *   dev        - the dev_t value to associate
1479  *   chrblk	- block or character device (DEVFS_CHAR or DEVFS_BLOCK)
1480  *   uid, gid	- ownership
1481  *   perms	- permissions
1482  *   clone	- minor number cloning function
1483  *   fmt, ...	- path format string with printf args to format the path name
1484  * Returns:
1485  *   A handle to a device node if successful, NULL otherwise.
1486  */
1487 void *
devfs_make_node_clone(dev_t dev,int chrblk,uid_t uid,gid_t gid,int perms,int (* clone)(dev_t dev,int action),const char * fmt,...)1488 devfs_make_node_clone(dev_t dev, int chrblk, uid_t uid,
1489     gid_t gid, int perms, int (*clone)(dev_t dev, int action),
1490     const char *fmt, ...)
1491 {
1492 	devdirent_t *   new_dev = NULL;
1493 	devfstype_t     type;
1494 	va_list ap;
1495 
1496 	switch (chrblk) {
1497 	case DEVFS_CHAR:
1498 		type = DEV_CDEV;
1499 		break;
1500 	case DEVFS_BLOCK:
1501 		type = DEV_BDEV;
1502 		break;
1503 	default:
1504 		goto out;
1505 	}
1506 
1507 	va_start(ap, fmt);
1508 	new_dev = devfs_make_node_internal(dev, type, uid, gid, perms, clone, fmt, ap);
1509 	va_end(ap);
1510 out:
1511 	return new_dev;
1512 }
1513 
1514 
1515 /*
1516  * Function: devfs_make_node
1517  *
1518  * Purpose
1519  *   Create a device node with the given pathname in the devfs namespace.
1520  *
1521  * Parameters:
1522  *   dev        - the dev_t value to associate
1523  *   chrblk	- block or character device (DEVFS_CHAR or DEVFS_BLOCK)
1524  *   uid, gid	- ownership
1525  *   perms	- permissions
1526  *   fmt, ...	- path format string with printf args to format the path name
1527  * Returns:
1528  *   A handle to a device node if successful, NULL otherwise.
1529  */
1530 void *
devfs_make_node(dev_t dev,int chrblk,uid_t uid,gid_t gid,int perms,const char * fmt,...)1531 devfs_make_node(dev_t dev, int chrblk, uid_t uid,
1532     gid_t gid, int perms, const char *fmt, ...)
1533 {
1534 	devdirent_t *   new_dev = NULL;
1535 	devfstype_t type;
1536 	va_list ap;
1537 
1538 	if (chrblk != DEVFS_CHAR && chrblk != DEVFS_BLOCK) {
1539 		goto out;
1540 	}
1541 
1542 	type = (chrblk == DEVFS_BLOCK ? DEV_BDEV : DEV_CDEV);
1543 
1544 	va_start(ap, fmt);
1545 	new_dev = devfs_make_node_internal(dev, type, uid, gid, perms, NULL, fmt, ap);
1546 	va_end(ap);
1547 
1548 out:
1549 	return new_dev;
1550 }
1551 
1552 static devdirent_t *
devfs_make_node_internal(dev_t dev,devfstype_t type,uid_t uid,gid_t gid,int perms,int (* clone)(dev_t dev,int action),const char * fmt,va_list ap)1553 devfs_make_node_internal(dev_t dev, devfstype_t type, uid_t uid,
1554     gid_t gid, int perms, int (*clone)(dev_t dev, int action), const char *fmt, va_list ap)
1555 {
1556 	devdirent_t *   new_dev = NULL;
1557 	devnode_t * dnp;
1558 	devnode_type_t  typeinfo;
1559 
1560 	char            *name, buf[256]; /* XXX */
1561 	const char      *path;
1562 #if CONFIG_MACF
1563 	char buff[sizeof(buf)];
1564 #endif
1565 	size_t          i;
1566 	uint32_t        log_count;
1567 	struct devfs_event_log event_log;
1568 	struct devfs_vnode_event stackbuf[NUM_STACK_ENTRIES];
1569 	int             need_free = 0;
1570 
1571 	vsnprintf(buf, sizeof(buf), fmt, ap);
1572 
1573 #if CONFIG_MACF
1574 	bcopy(buf, buff, sizeof(buff));
1575 	buff[sizeof(buff) - 1] = 0;
1576 #endif
1577 	name = NULL;
1578 
1579 	for (i = strlen(buf); i > 0; i--) {
1580 		if (buf[i] == '/') {
1581 			name = &buf[i];
1582 			buf[i] = 0;
1583 			break;
1584 		}
1585 	}
1586 
1587 	if (name) {
1588 		*name++ = '\0';
1589 		path = buf;
1590 	} else {
1591 		name = buf;
1592 		path = "/";
1593 	}
1594 
1595 	log_count = devfs_nmountplanes;
1596 	if (log_count > NUM_STACK_ENTRIES) {
1597 wrongsize:
1598 		need_free = 1;
1599 		if (devfs_init_event_log(&event_log, log_count, NULL) != 0) {
1600 			return NULL;
1601 		}
1602 	} else {
1603 		need_free = 0;
1604 		log_count = NUM_STACK_ENTRIES;
1605 		if (devfs_init_event_log(&event_log, log_count, &stackbuf[0]) != 0) {
1606 			return NULL;
1607 		}
1608 	}
1609 
1610 	DEVFS_LOCK();
1611 	if (log_count < devfs_nmountplanes) {
1612 		DEVFS_UNLOCK();
1613 		devfs_release_event_log(&event_log, need_free);
1614 		log_count = log_count * 2;
1615 		goto wrongsize;
1616 	}
1617 
1618 	if (!devfs_ready) {
1619 		printf("devfs_make_node: not ready for devices!\n");
1620 		goto out;
1621 	}
1622 
1623 	/* find/create directory path ie. mkdir -p */
1624 	if (dev_finddir(path, NULL, DEVFS_CREATE, &dnp, &event_log) == 0) {
1625 		typeinfo.dev = dev;
1626 		if (dev_add_entry(name, dnp, type, &typeinfo, NULL, NULL, &new_dev) == 0) {
1627 			new_dev->de_dnp->dn_gid = gid;
1628 			new_dev->de_dnp->dn_uid = uid;
1629 			new_dev->de_dnp->dn_mode |= perms;
1630 			new_dev->de_dnp->dn_clone = clone;
1631 #if CONFIG_MACF
1632 			mac_devfs_label_associate_device(dev, new_dev->de_dnp, buff);
1633 #endif
1634 			devfs_propogate(dnp->dn_typeinfo.Dir.myname, new_dev, &event_log);
1635 		}
1636 	}
1637 
1638 out:
1639 	DEVFS_UNLOCK();
1640 
1641 	devfs_bulk_notify(&event_log);
1642 	devfs_release_event_log(&event_log, need_free);
1643 	return new_dev;
1644 }
1645 
1646 /*
1647  * Function: devfs_make_link
1648  *
1649  * Purpose:
1650  *   Create a link to a previously created device node.
1651  *
1652  * Returns:
1653  *   0 if successful, -1 if failed
1654  */
1655 int
devfs_make_link(void * original,char * fmt,...)1656 devfs_make_link(void *original, char *fmt, ...)
1657 {
1658 	devdirent_t *   new_dev = NULL;
1659 	devdirent_t *   orig = (devdirent_t *) original;
1660 	devnode_t *     dirnode;        /* devnode for parent directory */
1661 	struct devfs_event_log event_log;
1662 	uint32_t        log_count;
1663 
1664 	va_list ap;
1665 	char *p, buf[256]; /* XXX */
1666 	size_t i;
1667 
1668 	DEVFS_LOCK();
1669 
1670 	if (!devfs_ready) {
1671 		DEVFS_UNLOCK();
1672 		printf("devfs_make_link: not ready for devices!\n");
1673 		return -1;
1674 	}
1675 	DEVFS_UNLOCK();
1676 
1677 	va_start(ap, fmt);
1678 	vsnprintf(buf, sizeof(buf), fmt, ap);
1679 	va_end(ap);
1680 
1681 	p = NULL;
1682 
1683 	for (i = strlen(buf); i > 0; i--) {
1684 		if (buf[i] == '/') {
1685 			p = &buf[i];
1686 			buf[i] = 0;
1687 			break;
1688 		}
1689 	}
1690 
1691 	/*
1692 	 * One slot for each directory, one for each devnode
1693 	 * whose link count changes
1694 	 */
1695 	log_count = devfs_nmountplanes * 2;
1696 wrongsize:
1697 	if (devfs_init_event_log(&event_log, log_count, NULL) != 0) {
1698 		/* No lock held, no allocations done, can just return */
1699 		return -1;
1700 	}
1701 
1702 	DEVFS_LOCK();
1703 
1704 	if (log_count < devfs_nmountplanes) {
1705 		DEVFS_UNLOCK();
1706 		devfs_release_event_log(&event_log, 1);
1707 		log_count = log_count * 2;
1708 		goto wrongsize;
1709 	}
1710 
1711 	if (p) {
1712 		*p++ = '\0';
1713 
1714 		if (dev_finddir(buf, NULL, DEVFS_CREATE, &dirnode, &event_log)
1715 		    || dev_add_name(p, dirnode, NULL, orig->de_dnp, &new_dev)) {
1716 			goto fail;
1717 		}
1718 	} else {
1719 		if (dev_finddir("", NULL, DEVFS_CREATE, &dirnode, &event_log)
1720 		    || dev_add_name(buf, dirnode, NULL, orig->de_dnp, &new_dev)) {
1721 			goto fail;
1722 		}
1723 	}
1724 	devfs_propogate(dirnode->dn_typeinfo.Dir.myname, new_dev, &event_log);
1725 fail:
1726 	DEVFS_UNLOCK();
1727 	devfs_bulk_notify(&event_log);
1728 	devfs_release_event_log(&event_log, 1);
1729 
1730 	return (new_dev != NULL) ? 0 : -1;
1731 }
1732