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 dnp = kalloc_type(devnode_t, Z_WAITOK | Z_ZERO | Z_NOFAIL);
535
536 /*
537 * If we have a proto, that means that we are duplicating some
538 * other device, which can only happen if we are not at the back plane
539 */
540 if (proto) {
541 bcopy(proto, dnp, sizeof(devnode_t));
542 dnp->dn_links = 0;
543 dnp->dn_linklist = NULL;
544 dnp->dn_vn = NULL;
545 dnp->dn_len = 0;
546 /* add to END of siblings list */
547 dnp->dn_prevsiblingp = proto->dn_prevsiblingp;
548 *(dnp->dn_prevsiblingp) = dnp;
549 dnp->dn_nextsibling = proto;
550 proto->dn_prevsiblingp = &(dnp->dn_nextsibling);
551 #if CONFIG_MACF
552 mac_devfs_label_init(dnp);
553 mac_devfs_label_copy(mac_devfs_label(proto), mac_devfs_label(dnp));
554 #endif
555 } else {
556 struct timeval tv;
557
558 /*
559 * We have no prototype, so start off with a clean slate
560 */
561 microtime(&tv);
562 dnp->dn_type = entrytype;
563 dnp->dn_nextsibling = dnp;
564 dnp->dn_prevsiblingp = &(dnp->dn_nextsibling);
565 dnp->dn_atime.tv_sec = tv.tv_sec;
566 dnp->dn_mtime.tv_sec = tv.tv_sec;
567 dnp->dn_ctime.tv_sec = tv.tv_sec;
568 #if CONFIG_MACF
569 mac_devfs_label_init(dnp);
570 #endif
571 }
572 dnp->dn_dvm = dvm;
573
574 /* Note: this inits the reference count to 1, this is considered unreferenced */
575 os_ref_init_raw(&dnp->dn_refcount, &devfs_refgrp);
576 dnp->dn_ino = devfs_unique_fileno;
577 devfs_unique_fileno++;
578
579 /*
580 * fill out the dev node according to type
581 */
582 switch (entrytype) {
583 case DEV_DIR:
584 /*
585 * As it's a directory, make sure
586 * it has a null entries list
587 */
588 dnp->dn_typeinfo.Dir.dirlast = &(dnp->dn_typeinfo.Dir.dirlist);
589 dnp->dn_typeinfo.Dir.dirlist = (devdirent_t *)0;
590 dnp->dn_typeinfo.Dir.entrycount = 0;
591 /* until we know better, it has a null parent pointer*/
592 dnp->dn_typeinfo.Dir.parent = NULL;
593 dnp->dn_links++; /* for .*/
594 dnp->dn_typeinfo.Dir.myname = NULL;
595 /*
596 * make sure that the ops associated with it are the ops
597 * that we use (by default) for directories
598 */
599 dnp->dn_ops = &devfs_vnodeop_p;
600 dnp->dn_mode |= 0555; /* default perms */
601 break;
602 case DEV_SLNK:
603 /*
604 * As it's a symlink allocate and store the link info
605 * Symlinks should only ever be created by the user,
606 * so they are not on the back plane and should not be
607 * propogated forward.. a bit like directories in that way..
608 * A symlink only exists on one plane and has its own
609 * node.. therefore we might be on any random plane.
610 */
611 dnp->dn_typeinfo.Slnk.name = kalloc_data(typeinfo->Slnk.namelen + 1, Z_WAITOK);
612 if (!dnp->dn_typeinfo.Slnk.name) {
613 error = ENOMEM;
614 break;
615 }
616 strlcpy(dnp->dn_typeinfo.Slnk.name, typeinfo->Slnk.name,
617 typeinfo->Slnk.namelen + 1);
618 dnp->dn_typeinfo.Slnk.namelen = typeinfo->Slnk.namelen;
619 DEVFS_INCR_STRINGSPACE(dnp->dn_typeinfo.Slnk.namelen + 1);
620 dnp->dn_ops = &devfs_vnodeop_p;
621 dnp->dn_mode |= 0555; /* default perms */
622 break;
623 case DEV_CDEV:
624 case DEV_BDEV:
625 /*
626 * Make sure it has DEVICE type ops
627 * and device specific fields are correct
628 */
629 dnp->dn_ops = &devfs_spec_vnodeop_p;
630 dnp->dn_typeinfo.dev = typeinfo->dev;
631 break;
632
633 #if FDESC
634 /* /dev/fd is special */
635 case DEV_DEVFD:
636 dnp->dn_ops = &devfs_devfd_vnodeop_p;
637 dnp->dn_mode |= 0555; /* default perms */
638 break;
639
640 #endif /* FDESC */
641 default:
642 error = EINVAL;
643 }
644
645 if (error) {
646 kfree_type(devnode_t, dnp);
647 } else {
648 *dn_pp = dnp;
649 DEVFS_INCR_NODES();
650 }
651
652 return error;
653 }
654
655
656 /***********************************************************************
657 * called with DEVFS_LOCK held
658 **********************************************************************/
659 void
devnode_free(devnode_t * dnp)660 devnode_free(devnode_t * dnp)
661 {
662 #if CONFIG_MACF
663 mac_devfs_label_destroy(dnp);
664 #endif
665 if (dnp->dn_type == DEV_SLNK) {
666 DEVFS_DECR_STRINGSPACE(dnp->dn_typeinfo.Slnk.namelen + 1);
667 kfree_data(dnp->dn_typeinfo.Slnk.name, dnp->dn_typeinfo.Slnk.namelen + 1);
668 }
669 DEVFS_DECR_NODES();
670 kfree_type(devnode_t, dnp);
671 }
672
673
674 /***********************************************************************
675 * called with DEVFS_LOCK held
676 **********************************************************************/
677 static void
devfs_dn_free(devnode_t * dnp)678 devfs_dn_free(devnode_t * dnp)
679 {
680 if (--dnp->dn_links <= 0) { /* can be -1 for initial free, on error */
681 /*probably need to do other cleanups XXX */
682 if (dnp->dn_nextsibling != dnp) {
683 devnode_t * * prevp = dnp->dn_prevsiblingp;
684 *prevp = dnp->dn_nextsibling;
685 dnp->dn_nextsibling->dn_prevsiblingp = prevp;
686 }
687
688 /* Can only free if there are no references; otherwise, wait for last vnode to be reclaimed */
689 os_ref_count_t rc = os_ref_get_count_raw(&dnp->dn_refcount);
690 if (rc == 1) {
691 /* release final reference from dev_add_node */
692 (void) os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
693 devnode_free(dnp);
694 } else {
695 dnp->dn_lflags |= DN_DELETE;
696 }
697 }
698 }
699
700 /***********************************************************************\
701 * Front Node Operations *
702 * Add or delete a chain of front nodes *
703 \***********************************************************************/
704
705
706 /***********************************************************************
707 * Given a directory backing node, and a child backing node, add the
708 * appropriate front nodes to the front nodes of the directory to
709 * represent the child node to the user
710 *
711 * on failure, front nodes will either be correct or not exist for each
712 * front dir, however dirs completed will not be stripped of completed
713 * frontnodes on failure of a later frontnode
714 *
715 * This allows a new node to be propogated through all mounted planes
716 *
717 * called with DEVFS_LOCK held
718 ***********************************************************************/
719 static int
devfs_propogate(devdirent_t * parent,devdirent_t * child,devfs_event_log_t delp)720 devfs_propogate(devdirent_t * parent, devdirent_t * child, devfs_event_log_t delp)
721 {
722 int error;
723 devdirent_t * newnmp;
724 devnode_t * dnp = child->de_dnp;
725 devnode_t * pdnp = parent->de_dnp;
726 devnode_t * adnp = parent->de_dnp;
727 int type = child->de_dnp->dn_type;
728 uint32_t events;
729
730 events = (dnp->dn_type == DEV_DIR ? VNODE_EVENT_DIR_CREATED : VNODE_EVENT_FILE_CREATED);
731 if (delp != NULL) {
732 devfs_record_event(delp, pdnp, events);
733 }
734
735 /***********************************************
736 * Find the other instances of the parent node
737 ***********************************************/
738 for (adnp = pdnp->dn_nextsibling;
739 adnp != pdnp;
740 adnp = adnp->dn_nextsibling) {
741 /*
742 * Make the node, using the original as a prototype)
743 * if the node already exists on that plane it won't be
744 * re-made..
745 */
746 if ((error = dev_add_entry(child->de_name, adnp, type,
747 NULL, dnp, adnp->dn_dvm,
748 &newnmp)) != 0) {
749 printf("duplicating %s failed\n", child->de_name);
750 } else {
751 if (delp != NULL) {
752 devfs_record_event(delp, adnp, events);
753
754 /*
755 * Slightly subtle. We're guaranteed that there will
756 * only be a vnode hooked into this devnode if we're creating
757 * a new link to an existing node; otherwise, the devnode is new
758 * and no one can have looked it up yet. If we're making a link,
759 * then the buffer is large enough for two nodes in each
760 * plane; otherwise, there's no vnode and this call will
761 * do nothing.
762 */
763 devfs_record_event(delp, newnmp->de_dnp, VNODE_EVENT_LINK);
764 }
765 }
766 }
767 return 0; /* for now always succeed */
768 }
769
770 static uint32_t
remove_notify_count(devnode_t * dnp)771 remove_notify_count(devnode_t *dnp)
772 {
773 uint32_t notify_count = 0;
774 devnode_t *dnp2;
775
776 /*
777 * Could need to notify for one removed node on each mount and
778 * one parent for each such node.
779 */
780 notify_count = devfs_nmountplanes;
781 notify_count += dnp->dn_links;
782 for (dnp2 = dnp->dn_nextsibling; dnp2 != dnp; dnp2 = dnp2->dn_nextsibling) {
783 notify_count += dnp2->dn_links;
784 }
785
786 return notify_count;
787 }
788
789 /***********************************************************************
790 * remove all instances of this devicename [for backing nodes..]
791 * note.. if there is another link to the node (non dir nodes only)
792 * then the devfs_node will still exist as the ref count will be non-0
793 * removing a directory node will remove all sup-nodes on all planes (ZAP)
794 *
795 * Used by device drivers to remove nodes that are no longer relevant
796 * The argument is the 'cookie' they were given when they created the node
797 * this function is exported.. see devfs.h
798 ***********************************************************************/
799 void
devfs_remove(void * dirent_p)800 devfs_remove(void *dirent_p)
801 {
802 devnode_t * dnp = ((devdirent_t *)dirent_p)->de_dnp;
803 devnode_t * dnp2;
804 boolean_t lastlink;
805 struct devfs_event_log event_log;
806 uint32_t log_count = 0;
807 int do_notify = 0;
808 int need_free = 0;
809 struct devfs_vnode_event stackbuf[NUM_STACK_ENTRIES];
810
811 DEVFS_LOCK();
812
813 if (!devfs_ready) {
814 printf("devfs_remove: not ready for devices!\n");
815 goto out;
816 }
817
818 log_count = remove_notify_count(dnp);
819
820 if (log_count > NUM_STACK_ENTRIES) {
821 uint32_t new_count;
822 wrongsize:
823 DEVFS_UNLOCK();
824 if (devfs_init_event_log(&event_log, log_count, NULL) == 0) {
825 do_notify = 1;
826 need_free = 1;
827 }
828 DEVFS_LOCK();
829
830 new_count = remove_notify_count(dnp);
831 if (need_free && (new_count > log_count)) {
832 devfs_release_event_log(&event_log, 1);
833 need_free = 0;
834 do_notify = 0;
835 log_count = log_count * 2;
836 goto wrongsize;
837 }
838 } else {
839 if (devfs_init_event_log(&event_log, NUM_STACK_ENTRIES, &stackbuf[0]) == 0) {
840 do_notify = 1;
841 }
842 }
843
844 /* This file has been deleted */
845 if (do_notify != 0) {
846 devfs_record_event(&event_log, dnp, VNODE_EVENT_DELETE);
847 }
848
849 /* keep removing the next sibling till only we exist. */
850 while ((dnp2 = dnp->dn_nextsibling) != dnp) {
851 /*
852 * Keep removing the next front node till no more exist
853 */
854 dnp->dn_nextsibling = dnp2->dn_nextsibling;
855 dnp->dn_nextsibling->dn_prevsiblingp = &(dnp->dn_nextsibling);
856 dnp2->dn_nextsibling = dnp2;
857 dnp2->dn_prevsiblingp = &(dnp2->dn_nextsibling);
858
859 /* This file has been deleted in this plane */
860 if (do_notify != 0) {
861 devfs_record_event(&event_log, dnp2, VNODE_EVENT_DELETE);
862 }
863
864 if (dnp2->dn_linklist) {
865 do {
866 lastlink = (1 == dnp2->dn_links);
867 /* Each parent of a link to this file has lost a child in this plane */
868 if (do_notify != 0) {
869 devfs_record_event(&event_log, dnp2->dn_linklist->de_parent, VNODE_EVENT_FILE_REMOVED);
870 }
871 dev_free_name(dnp2->dn_linklist);
872 } while (!lastlink);
873 }
874 }
875
876 /*
877 * then free the main node
878 * If we are not running in SPLIT_DEVS mode, then
879 * THIS is what gets rid of the propogated nodes.
880 */
881 if (dnp->dn_linklist) {
882 do {
883 lastlink = (1 == dnp->dn_links);
884 /* Each parent of a link to this file has lost a child */
885 if (do_notify != 0) {
886 devfs_record_event(&event_log, dnp->dn_linklist->de_parent, VNODE_EVENT_FILE_REMOVED);
887 }
888 dev_free_name(dnp->dn_linklist);
889 } while (!lastlink);
890 }
891 out:
892 DEVFS_UNLOCK();
893 if (do_notify != 0) {
894 devfs_bulk_notify(&event_log);
895 devfs_release_event_log(&event_log, need_free);
896 }
897
898 return;
899 }
900
901
902
903 /***************************************************************
904 * duplicate the backing tree into a tree of nodes hung off the
905 * mount point given as the argument. Do this by
906 * calling dev_dup_entry which recurses all the way
907 * up the tree..
908 *
909 * called with DEVFS_LOCK held
910 **************************************************************/
911 int
dev_dup_plane(struct devfsmount * devfs_mp_p)912 dev_dup_plane(struct devfsmount *devfs_mp_p)
913 {
914 devdirent_t * new;
915 int error = 0;
916
917 if ((error = dev_dup_entry(NULL, dev_root, &new, devfs_mp_p))) {
918 return error;
919 }
920 devfs_mp_p->plane_root = new;
921 devfs_nmountplanes++;
922 return error;
923 }
924
925
926
927 /***************************************************************
928 * Free a whole plane
929 *
930 * called with DEVFS_LOCK held
931 ***************************************************************/
932 void
devfs_free_plane(struct devfsmount * devfs_mp_p)933 devfs_free_plane(struct devfsmount *devfs_mp_p)
934 {
935 devdirent_t * dirent_p;
936
937 dirent_p = devfs_mp_p->plane_root;
938 if (dirent_p) {
939 dev_free_hier(dirent_p);
940 dev_free_name(dirent_p);
941 }
942 devfs_mp_p->plane_root = NULL;
943 devfs_nmountplanes--;
944
945 if (devfs_nmountplanes > (devfs_nmountplanes + 1)) {
946 panic("plane count wrapped around.");
947 }
948 }
949
950
951 /***************************************************************
952 * Create and link in a new front element..
953 * Parent can be 0 for a root node
954 * Not presently usable to make a symlink XXX
955 * (Ok, symlinks don't propogate)
956 * recursively will create subnodes corresponding to equivalent
957 * child nodes in the base level
958 *
959 * called with DEVFS_LOCK held
960 ***************************************************************/
961 static int
dev_dup_entry(devnode_t * parent,devdirent_t * back,devdirent_t ** dnm_pp,struct devfsmount * dvm)962 dev_dup_entry(devnode_t * parent, devdirent_t * back, devdirent_t * *dnm_pp,
963 struct devfsmount *dvm)
964 {
965 devdirent_t * entry_p = NULL;
966 devdirent_t * newback;
967 devdirent_t * newfront;
968 int error;
969 devnode_t * dnp = back->de_dnp;
970 int type = dnp->dn_type;
971
972 /*
973 * go get the node made (if we need to)
974 * use the back one as a prototype
975 */
976 error = dev_add_entry(back->de_name, parent, type, NULL, dnp,
977 parent?parent->dn_dvm:dvm, &entry_p);
978 if (!error && (entry_p == NULL)) {
979 error = ENOMEM; /* Really can't happen, but make static analyzer happy */
980 }
981 if (error != 0) {
982 printf("duplicating %s failed\n", back->de_name);
983 goto out;
984 }
985
986 /*
987 * If we have just made the root, then insert the pointer to the
988 * mount information
989 */
990 if (dvm) {
991 entry_p->de_dnp->dn_dvm = dvm;
992 }
993
994 /*
995 * If it is a directory, then recurse down all the other
996 * subnodes in it....
997 * note that this time we don't pass on the mount info..
998 */
999 if (type == DEV_DIR) {
1000 for (newback = back->de_dnp->dn_typeinfo.Dir.dirlist;
1001 newback; newback = newback->de_next) {
1002 if ((error = dev_dup_entry(entry_p->de_dnp,
1003 newback, &newfront, NULL)) != 0) {
1004 break; /* back out with an error */
1005 }
1006 }
1007 }
1008 out:
1009 *dnm_pp = entry_p;
1010 return error;
1011 }
1012
1013
1014 /***************************************************************
1015 * Free a name node
1016 * remember that if there are other names pointing to the
1017 * dev_node then it may not get freed yet
1018 * can handle if there is no dnp
1019 *
1020 * called with DEVFS_LOCK held
1021 ***************************************************************/
1022
1023 int
dev_free_name(devdirent_t * dirent_p)1024 dev_free_name(devdirent_t * dirent_p)
1025 {
1026 devnode_t * parent = dirent_p->de_parent;
1027 devnode_t * dnp = dirent_p->de_dnp;
1028
1029 if (dnp) {
1030 if (dnp->dn_type == DEV_DIR) {
1031 devnode_t * p;
1032
1033 if (dnp->dn_typeinfo.Dir.dirlist) {
1034 return ENOTEMPTY;
1035 }
1036 p = dnp->dn_typeinfo.Dir.parent;
1037 devfs_dn_free(dnp); /* account for '.' */
1038 devfs_dn_free(p); /* '..' */
1039 }
1040 /*
1041 * unlink us from the list of links for this node
1042 * If we are the only link, it's easy!
1043 * if we are a DIR of course there should not be any
1044 * other links.
1045 */
1046 if (dirent_p->de_nextlink == dirent_p) {
1047 dnp->dn_linklist = NULL;
1048 } else {
1049 if (dnp->dn_linklist == dirent_p) {
1050 dnp->dn_linklist = dirent_p->de_nextlink;
1051 }
1052 }
1053 devfs_dn_free(dnp);
1054 }
1055
1056 dirent_p->de_nextlink->de_prevlinkp = dirent_p->de_prevlinkp;
1057 *(dirent_p->de_prevlinkp) = dirent_p->de_nextlink;
1058
1059 /*
1060 * unlink ourselves from the directory on this plane
1061 */
1062 if (parent) { /* if not fs root */
1063 if ((*dirent_p->de_prevp = dirent_p->de_next)) {/* yes, assign */
1064 dirent_p->de_next->de_prevp = dirent_p->de_prevp;
1065 } else {
1066 parent->dn_typeinfo.Dir.dirlast
1067 = dirent_p->de_prevp;
1068 }
1069 parent->dn_typeinfo.Dir.entrycount--;
1070 parent->dn_len -= strlen(dirent_p->de_name) + 8;
1071 }
1072
1073 DEVFS_DECR_ENTRIES();
1074 kfree_type(devdirent_t, dirent_p);
1075 return 0;
1076 }
1077
1078
1079 /***************************************************************
1080 * Free a hierarchy starting at a directory node name
1081 * remember that if there are other names pointing to the
1082 * dev_node then it may not get freed yet
1083 * can handle if there is no dnp
1084 * leave the node itself allocated.
1085 *
1086 * called with DEVFS_LOCK held
1087 ***************************************************************/
1088
1089 static void
dev_free_hier(devdirent_t * dirent_p)1090 dev_free_hier(devdirent_t * dirent_p)
1091 {
1092 devnode_t * dnp = dirent_p->de_dnp;
1093
1094 if (dnp) {
1095 if (dnp->dn_type == DEV_DIR) {
1096 while (dnp->dn_typeinfo.Dir.dirlist) {
1097 dev_free_hier(dnp->dn_typeinfo.Dir.dirlist);
1098 dev_free_name(dnp->dn_typeinfo.Dir.dirlist);
1099 }
1100 }
1101 }
1102 }
1103
1104
1105 /***************************************************************
1106 * given a dev_node, find the appropriate vnode if one is already
1107 * associated, or get a new one and associate it with the dev_node
1108 *
1109 * called with DEVFS_LOCK held
1110 *
1111 * If an error is returned, then the dnp may have been freed (we
1112 * raced with a delete and lost). A devnode should not be accessed
1113 * after devfs_dntovn() fails.
1114 ****************************************************************/
1115 int
devfs_dntovn(devnode_t * dnp,struct vnode ** vn_pp,__unused struct proc * p)1116 devfs_dntovn(devnode_t * dnp, struct vnode **vn_pp, __unused struct proc * p)
1117 {
1118 struct vnode *vn_p;
1119 int error = 0;
1120 struct vnode_fsparam vfsp;
1121 enum vtype vtype = 0;
1122 int markroot = 0;
1123 int nretries = 0;
1124 int n_minor = DEVFS_CLONE_ALLOC; /* new minor number for clone device */
1125
1126 /*
1127 * We should never come in and find that our devnode has been marked for delete.
1128 * The lookup should have held the lock from entry until now; it should not have
1129 * been able to find a removed entry. Any other pathway would have just created
1130 * the devnode and come here without dropping the devfs lock, so no one would
1131 * have a chance to delete.
1132 */
1133 if (dnp->dn_lflags & DN_DELETE) {
1134 panic("devfs_dntovn: DN_DELETE set on a devnode upon entry.");
1135 }
1136
1137 devfs_ref_node(dnp);
1138
1139 retry:
1140 *vn_pp = NULL;
1141 vn_p = dnp->dn_vn;
1142
1143 if (vn_p) { /* already has a vnode */
1144 uint32_t vid;
1145
1146 vid = vnode_vid(vn_p);
1147
1148 vnode_hold(vn_p);
1149 DEVFS_UNLOCK();
1150
1151 /*
1152 * We want to use the drainok variant of vnode_getwithvid
1153 * because we _don't_ want to get an iocount if the vnode is
1154 * is blocked in vnode_drain as it can cause infinite
1155 * loops in vn_open_auth. While in use vnodes are typically
1156 * only reclaimed on forced unmounts, In use devfs tty vnodes
1157 * can be quite frequently reclaimed by revoke(2) or by the
1158 * exit of a controlling process.
1159 */
1160 error = vnode_getwithvid_drainok(vn_p, vid);
1161
1162 vnode_drop(vn_p);
1163 DEVFS_LOCK();
1164
1165 if (dnp->dn_lflags & DN_DELETE) {
1166 /*
1167 * our BUSY node got marked for
1168 * deletion while the DEVFS lock
1169 * was dropped...
1170 */
1171 if (error == 0) {
1172 /*
1173 * vnode_getwithvid returned a valid ref
1174 * which we need to drop
1175 */
1176 vnode_put(vn_p);
1177 }
1178
1179 /*
1180 * This entry is no longer in the namespace. This is only
1181 * possible for lookup: no other path would not find an existing
1182 * vnode. Therefore, ENOENT is a valid result.
1183 */
1184 error = ENOENT;
1185 } else if (error == ENODEV) {
1186 /*
1187 * The Filesystem is getting unmounted.
1188 */
1189 error = ENOENT;
1190 } else if (error && (nretries < DEV_MAX_VNODE_RETRY)) {
1191 /*
1192 * If we got an error from vnode_getwithvid, it means
1193 * we raced with a recycle and lost i.e. we asked for
1194 * an iocount only after vnode_drain had been entered
1195 * for the vnode and returned with an error only after
1196 * devfs_reclaim was called on the vnode. devfs_reclaim
1197 * sets dn_vn to NULL but while we were waiting to
1198 * reacquire DEVFS_LOCK, another vnode might have gotten
1199 * associated with the dnp. In either case, we need to
1200 * retry otherwise we will end up returning an ENOENT
1201 * for this lookup but the next lookup will succeed
1202 * because it creates a new vnode (or a racing lookup
1203 * created a new vnode already).
1204 */
1205 error = 0;
1206 nretries++;
1207 goto retry;
1208 }
1209 if (!error) {
1210 *vn_pp = vn_p;
1211 }
1212
1213 goto out;
1214 }
1215
1216 /*
1217 * If we get here, then we've beaten any deletes;
1218 * if someone sets DN_DELETE during a subsequent drop
1219 * of the devfs lock, we'll still vend a vnode.
1220 */
1221
1222 if (dnp->dn_lflags & DN_CREATE) {
1223 dnp->dn_lflags |= DN_CREATEWAIT;
1224 msleep(&dnp->dn_lflags, &devfs_mutex, PRIBIO, 0, 0);
1225 goto retry;
1226 }
1227
1228 dnp->dn_lflags |= DN_CREATE;
1229
1230 switch (dnp->dn_type) {
1231 case DEV_SLNK:
1232 vtype = VLNK;
1233 break;
1234 case DEV_DIR:
1235 if (dnp->dn_typeinfo.Dir.parent == dnp) {
1236 markroot = 1;
1237 }
1238 vtype = VDIR;
1239 break;
1240 case DEV_BDEV:
1241 case DEV_CDEV:
1242 vtype = (dnp->dn_type == DEV_BDEV) ? VBLK : VCHR;
1243 break;
1244 #if FDESC
1245 case DEV_DEVFD:
1246 vtype = VDIR;
1247 break;
1248 #endif /* FDESC */
1249 }
1250 vfsp.vnfs_mp = dnp->dn_dvm->mount;
1251 vfsp.vnfs_vtype = vtype;
1252 vfsp.vnfs_str = "devfs";
1253 vfsp.vnfs_dvp = 0;
1254 vfsp.vnfs_fsnode = dnp;
1255 vfsp.vnfs_cnp = 0;
1256 vfsp.vnfs_vops = *(dnp->dn_ops);
1257
1258 if (vtype == VBLK || vtype == VCHR) {
1259 /*
1260 * Ask the clone minor number function for a new minor number
1261 * to use for the next device instance. If an administative
1262 * limit has been reached, this function will return -1.
1263 */
1264 if (dnp->dn_clone != NULL) {
1265 int n_major = major(dnp->dn_typeinfo.dev);
1266
1267 n_minor = (*dnp->dn_clone)(dnp->dn_typeinfo.dev, DEVFS_CLONE_ALLOC);
1268 if (n_minor == -1) {
1269 error = ENOMEM;
1270 goto out;
1271 }
1272
1273 vfsp.vnfs_rdev = makedev(n_major, n_minor);
1274 } else {
1275 vfsp.vnfs_rdev = dnp->dn_typeinfo.dev;
1276 }
1277 } else {
1278 vfsp.vnfs_rdev = 0;
1279 }
1280 vfsp.vnfs_filesize = 0;
1281 vfsp.vnfs_flags = VNFS_NOCACHE | VNFS_CANTCACHE;
1282 /* Tag system files */
1283 vfsp.vnfs_marksystem = 0;
1284 vfsp.vnfs_markroot = markroot;
1285
1286 DEVFS_UNLOCK();
1287
1288 error = vnode_create_ext(VNCREATE_FLAVOR, VCREATESIZE, &vfsp, &vn_p,
1289 VNODE_CREATE_DEFAULT);
1290
1291 /* Do this before grabbing the lock */
1292 if (error == 0) {
1293 vnode_setneedinactive(vn_p);
1294 }
1295
1296 DEVFS_LOCK();
1297
1298 if (error == 0) {
1299 vnode_settag(vn_p, VT_DEVFS);
1300
1301 if ((dnp->dn_clone != NULL) && (dnp->dn_vn != NULLVP)) {
1302 panic("devfs_dntovn: cloning device with a vnode?");
1303 }
1304
1305 *vn_pp = vn_p;
1306
1307 /*
1308 * Another vnode that has this devnode as its v_data.
1309 * This reference, unlike the one taken at the start
1310 * of the function, persists until a VNOP_RECLAIM
1311 * comes through for this vnode.
1312 */
1313 devfs_ref_node(dnp);
1314
1315 /*
1316 * A cloned vnode is not hooked into the devnode; every lookup
1317 * gets a new vnode.
1318 */
1319 if (dnp->dn_clone == NULL) {
1320 dnp->dn_vn = vn_p;
1321 }
1322 } else if (n_minor != DEVFS_CLONE_ALLOC) {
1323 /*
1324 * If we failed the create, we need to release the cloned minor
1325 * back to the free list. In general, this is only useful if
1326 * the clone function results in a state change in the cloned
1327 * device for which the minor number was obtained. If we get
1328 * past this point withouth falling into this case, it's
1329 * assumed that any state to be released will be released when
1330 * the vnode is dropped, instead.
1331 */
1332 (void)(*dnp->dn_clone)(dnp->dn_typeinfo.dev, DEVFS_CLONE_FREE);
1333 }
1334
1335 dnp->dn_lflags &= ~DN_CREATE;
1336 if (dnp->dn_lflags & DN_CREATEWAIT) {
1337 dnp->dn_lflags &= ~DN_CREATEWAIT;
1338 wakeup(&dnp->dn_lflags);
1339 }
1340
1341 out:
1342 /*
1343 * Release the reference we took to prevent deletion while we weren't holding the lock.
1344 * If not returning success, then dropping this reference could delete the devnode;
1345 * no one should access a devnode after a call to devfs_dntovn fails.
1346 */
1347 devfs_rele_node(dnp);
1348
1349 return error;
1350 }
1351
1352 /*
1353 * Increment refcount on a devnode; prevents free of the node
1354 * while the devfs lock is not held.
1355 */
1356 void
devfs_ref_node(devnode_t * dnp)1357 devfs_ref_node(devnode_t *dnp)
1358 {
1359 os_ref_retain_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1360 }
1361
1362 /*
1363 * Release a reference on a devnode. If the devnode is marked for
1364 * free and the refcount is dropped to one, do the free.
1365 */
1366 void
devfs_rele_node(devnode_t * dnp)1367 devfs_rele_node(devnode_t *dnp)
1368 {
1369 os_ref_count_t rc = os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1370 if (rc < 1) {
1371 panic("devfs_rele_node: devnode without a refcount!");
1372 } else if ((rc == 1) && (dnp->dn_lflags & DN_DELETE)) {
1373 /* release final reference from dev_add_node */
1374 (void) os_ref_release_locked_raw(&dnp->dn_refcount, &devfs_refgrp);
1375 devnode_free(dnp);
1376 }
1377 }
1378
1379 /***********************************************************************
1380 * add a whole device, with no prototype.. make name element and node
1381 * Used for adding the original device entries
1382 *
1383 * called with DEVFS_LOCK held
1384 ***********************************************************************/
1385 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)1386 dev_add_entry(const char *name, devnode_t * parent, int type, devnode_type_t * typeinfo,
1387 devnode_t * proto, struct devfsmount *dvm, devdirent_t * *nm_pp)
1388 {
1389 devnode_t * dnp;
1390 int error = 0;
1391
1392 if ((error = dev_add_node(type, typeinfo, proto, &dnp,
1393 (parent?parent->dn_dvm:dvm))) != 0) {
1394 printf("devfs: %s: base node allocation failed (Errno=%d)\n",
1395 name, error);
1396 return error;
1397 }
1398 if ((error = dev_add_name(name, parent, NULL, dnp, nm_pp)) != 0) {
1399 devfs_dn_free(dnp); /* 1->0 for dir, 0->(-1) for other */
1400 printf("devfs: %s: name slot allocation failed (Errno=%d)\n",
1401 name, error);
1402 }
1403 return error;
1404 }
1405
1406 static void
devfs_bulk_notify(devfs_event_log_t delp)1407 devfs_bulk_notify(devfs_event_log_t delp)
1408 {
1409 uint32_t i;
1410 for (i = 0; i < delp->del_used; i++) {
1411 devfs_vnode_event_t dvep = &delp->del_entries[i];
1412 if (vnode_getwithvid(dvep->dve_vp, dvep->dve_vid) == 0) {
1413 vnode_notify(dvep->dve_vp, dvep->dve_events, NULL);
1414 vnode_put(dvep->dve_vp);
1415 }
1416 vnode_drop(dvep->dve_vp);
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 vnode_hold(dvep->dve_vp);
1433 dvep->dve_events = events;
1434 delp->del_used++;
1435 }
1436 }
1437
1438 static int
devfs_init_event_log(devfs_event_log_t delp,uint32_t count,devfs_vnode_event_t buf)1439 devfs_init_event_log(devfs_event_log_t delp, uint32_t count, devfs_vnode_event_t buf)
1440 {
1441 devfs_vnode_event_t dvearr;
1442
1443 if (buf == NULL) {
1444 dvearr = kalloc_type(struct devfs_vnode_event, count,
1445 Z_WAITOK | Z_ZERO);
1446 if (dvearr == NULL) {
1447 return ENOMEM;
1448 }
1449 } else {
1450 dvearr = buf;
1451 }
1452
1453 delp->del_max = count;
1454 delp->del_used = 0;
1455 delp->del_entries = dvearr;
1456 return 0;
1457 }
1458
1459 static void
devfs_release_event_log(devfs_event_log_t delp,int need_free)1460 devfs_release_event_log(devfs_event_log_t delp, int need_free)
1461 {
1462 if (delp->del_entries == NULL) {
1463 panic("Free of devfs notify info that has not been intialized.");
1464 }
1465
1466 if (need_free) {
1467 kfree_type(struct devfs_vnode_event, delp->del_max,
1468 delp->del_entries);
1469 }
1470
1471 delp->del_entries = NULL;
1472 }
1473
1474 /*
1475 * Function: devfs_make_node
1476 *
1477 * Purpose
1478 * Create a device node with the given pathname in the devfs namespace.
1479 *
1480 * Parameters:
1481 * dev - the dev_t value to associate
1482 * chrblk - block or character device (DEVFS_CHAR or DEVFS_BLOCK)
1483 * uid, gid - ownership
1484 * perms - permissions
1485 * clone - minor number cloning function
1486 * fmt, ... - path format string with printf args to format the path name
1487 * Returns:
1488 * A handle to a device node if successful, NULL otherwise.
1489 */
1490 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,...)1491 devfs_make_node_clone(dev_t dev, int chrblk, uid_t uid,
1492 gid_t gid, int perms, int (*clone)(dev_t dev, int action),
1493 const char *fmt, ...)
1494 {
1495 devdirent_t * new_dev = NULL;
1496 devfstype_t type;
1497 va_list ap;
1498
1499 switch (chrblk) {
1500 case DEVFS_CHAR:
1501 type = DEV_CDEV;
1502 break;
1503 case DEVFS_BLOCK:
1504 type = DEV_BDEV;
1505 break;
1506 default:
1507 goto out;
1508 }
1509
1510 va_start(ap, fmt);
1511 new_dev = devfs_make_node_internal(dev, type, uid, gid, perms, clone, fmt, ap);
1512 va_end(ap);
1513 out:
1514 return new_dev;
1515 }
1516
1517
1518 /*
1519 * Function: devfs_make_node
1520 *
1521 * Purpose
1522 * Create a device node with the given pathname in the devfs namespace.
1523 *
1524 * Parameters:
1525 * dev - the dev_t value to associate
1526 * chrblk - block or character device (DEVFS_CHAR or DEVFS_BLOCK)
1527 * uid, gid - ownership
1528 * perms - permissions
1529 * fmt, ... - path format string with printf args to format the path name
1530 * Returns:
1531 * A handle to a device node if successful, NULL otherwise.
1532 */
1533 void *
devfs_make_node(dev_t dev,int chrblk,uid_t uid,gid_t gid,int perms,const char * fmt,...)1534 devfs_make_node(dev_t dev, int chrblk, uid_t uid,
1535 gid_t gid, int perms, const char *fmt, ...)
1536 {
1537 devdirent_t * new_dev = NULL;
1538 devfstype_t type;
1539 va_list ap;
1540
1541 if (chrblk != DEVFS_CHAR && chrblk != DEVFS_BLOCK) {
1542 goto out;
1543 }
1544
1545 type = (chrblk == DEVFS_BLOCK ? DEV_BDEV : DEV_CDEV);
1546
1547 va_start(ap, fmt);
1548 new_dev = devfs_make_node_internal(dev, type, uid, gid, perms, NULL, fmt, ap);
1549 va_end(ap);
1550
1551 out:
1552 return new_dev;
1553 }
1554
1555 __printflike(7, 0)
1556 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)1557 devfs_make_node_internal(dev_t dev, devfstype_t type, uid_t uid,
1558 gid_t gid, int perms, int (*clone)(dev_t dev, int action), const char *fmt, va_list ap)
1559 {
1560 devdirent_t * new_dev = NULL;
1561 devnode_t * dnp;
1562 devnode_type_t typeinfo;
1563
1564 char *name, buf[256]; /* XXX */
1565 const char *path;
1566 #if CONFIG_MACF
1567 char buff[sizeof(buf)];
1568 #endif
1569 size_t i;
1570 uint32_t log_count;
1571 struct devfs_event_log event_log;
1572 struct devfs_vnode_event stackbuf[NUM_STACK_ENTRIES];
1573 int need_free = 0;
1574
1575 vsnprintf(buf, sizeof(buf), fmt, ap);
1576
1577 #if CONFIG_MACF
1578 bcopy(buf, buff, sizeof(buff));
1579 buff[sizeof(buff) - 1] = 0;
1580 #endif
1581 name = NULL;
1582
1583 for (i = strlen(buf); i > 0; i--) {
1584 if (buf[i] == '/') {
1585 name = &buf[i];
1586 buf[i] = 0;
1587 break;
1588 }
1589 }
1590
1591 if (name) {
1592 *name++ = '\0';
1593 path = buf;
1594 } else {
1595 name = buf;
1596 path = "/";
1597 }
1598
1599 log_count = devfs_nmountplanes;
1600 if (log_count > NUM_STACK_ENTRIES) {
1601 wrongsize:
1602 need_free = 1;
1603 if (devfs_init_event_log(&event_log, log_count, NULL) != 0) {
1604 return NULL;
1605 }
1606 } else {
1607 need_free = 0;
1608 log_count = NUM_STACK_ENTRIES;
1609 if (devfs_init_event_log(&event_log, log_count, &stackbuf[0]) != 0) {
1610 return NULL;
1611 }
1612 }
1613
1614 DEVFS_LOCK();
1615 if (log_count < devfs_nmountplanes) {
1616 DEVFS_UNLOCK();
1617 devfs_release_event_log(&event_log, need_free);
1618 log_count = log_count * 2;
1619 goto wrongsize;
1620 }
1621
1622 if (!devfs_ready) {
1623 printf("devfs_make_node: not ready for devices!\n");
1624 goto out;
1625 }
1626
1627 /* find/create directory path ie. mkdir -p */
1628 if (dev_finddir(path, NULL, DEVFS_CREATE, &dnp, &event_log) == 0) {
1629 typeinfo.dev = dev;
1630 if (dev_add_entry(name, dnp, type, &typeinfo, NULL, NULL, &new_dev) == 0) {
1631 new_dev->de_dnp->dn_gid = gid;
1632 new_dev->de_dnp->dn_uid = uid;
1633 new_dev->de_dnp->dn_mode |= perms;
1634 new_dev->de_dnp->dn_clone = clone;
1635 #if CONFIG_MACF
1636 mac_devfs_label_associate_device(dev, new_dev->de_dnp, buff);
1637 #endif
1638 devfs_propogate(dnp->dn_typeinfo.Dir.myname, new_dev, &event_log);
1639 }
1640 }
1641
1642 out:
1643 DEVFS_UNLOCK();
1644
1645 devfs_bulk_notify(&event_log);
1646 devfs_release_event_log(&event_log, need_free);
1647 return new_dev;
1648 }
1649
1650 /*
1651 * Function: devfs_make_link
1652 *
1653 * Purpose:
1654 * Create a link to a previously created device node.
1655 *
1656 * Returns:
1657 * 0 if successful, -1 if failed
1658 */
1659 int
devfs_make_link(void * original,char * fmt,...)1660 devfs_make_link(void *original, char *fmt, ...)
1661 {
1662 devdirent_t * new_dev = NULL;
1663 devdirent_t * orig = (devdirent_t *) original;
1664 devnode_t * dirnode; /* devnode for parent directory */
1665 struct devfs_event_log event_log;
1666 uint32_t log_count;
1667
1668 va_list ap;
1669 char *p, buf[256]; /* XXX */
1670 size_t i;
1671
1672 DEVFS_LOCK();
1673
1674 if (!devfs_ready) {
1675 DEVFS_UNLOCK();
1676 printf("devfs_make_link: not ready for devices!\n");
1677 return -1;
1678 }
1679 DEVFS_UNLOCK();
1680
1681 va_start(ap, fmt);
1682 vsnprintf(buf, sizeof(buf), fmt, ap);
1683 va_end(ap);
1684
1685 p = NULL;
1686
1687 for (i = strlen(buf); i > 0; i--) {
1688 if (buf[i] == '/') {
1689 p = &buf[i];
1690 buf[i] = 0;
1691 break;
1692 }
1693 }
1694
1695 /*
1696 * One slot for each directory, one for each devnode
1697 * whose link count changes
1698 */
1699 log_count = devfs_nmountplanes * 2;
1700 wrongsize:
1701 if (devfs_init_event_log(&event_log, log_count, NULL) != 0) {
1702 /* No lock held, no allocations done, can just return */
1703 return -1;
1704 }
1705
1706 DEVFS_LOCK();
1707
1708 if (log_count < devfs_nmountplanes) {
1709 DEVFS_UNLOCK();
1710 devfs_release_event_log(&event_log, 1);
1711 log_count = log_count * 2;
1712 goto wrongsize;
1713 }
1714
1715 if (p) {
1716 *p++ = '\0';
1717
1718 if (dev_finddir(buf, NULL, DEVFS_CREATE, &dirnode, &event_log)
1719 || dev_add_name(p, dirnode, NULL, orig->de_dnp, &new_dev)) {
1720 goto fail;
1721 }
1722 } else {
1723 if (dev_finddir("", NULL, DEVFS_CREATE, &dirnode, &event_log)
1724 || dev_add_name(buf, dirnode, NULL, orig->de_dnp, &new_dev)) {
1725 goto fail;
1726 }
1727 }
1728 devfs_propogate(dirnode->dn_typeinfo.Dir.myname, new_dev, &event_log);
1729 fail:
1730 DEVFS_UNLOCK();
1731 devfs_bulk_notify(&event_log);
1732 devfs_release_event_log(&event_log, 1);
1733
1734 return (new_dev != NULL) ? 0 : -1;
1735 }
1736