xref: /illumos-gate/usr/src/uts/common/fs/dev/sdev_subr.c (revision 8119dad84d6416f13557b0ba8e2aaf9064cbcfd3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013, 2016 Joyent, Inc. All rights reserved.
24  * Copyright (c) 2017 by Delphix. All rights reserved.
25  */
26 
27 /*
28  * utility routines for the /dev fs
29  */
30 
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/t_lock.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/user.h>
37 #include <sys/time.h>
38 #include <sys/vfs.h>
39 #include <sys/vnode.h>
40 #include <sys/file.h>
41 #include <sys/fcntl.h>
42 #include <sys/flock.h>
43 #include <sys/kmem.h>
44 #include <sys/uio.h>
45 #include <sys/errno.h>
46 #include <sys/stat.h>
47 #include <sys/cred.h>
48 #include <sys/dirent.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/debug.h>
52 #include <sys/mode.h>
53 #include <sys/policy.h>
54 #include <fs/fs_subr.h>
55 #include <sys/mount.h>
56 #include <sys/fs/snode.h>
57 #include <sys/fs/dv_node.h>
58 #include <sys/fs/sdev_impl.h>
59 #include <sys/sunndi.h>
60 #include <sys/sunmdi.h>
61 #include <sys/conf.h>
62 #include <sys/proc.h>
63 #include <sys/user.h>
64 #include <sys/modctl.h>
65 
66 #ifdef DEBUG
67 int sdev_debug = 0x00000001;
68 int sdev_debug_cache_flags = 0;
69 #endif
70 
71 /*
72  * globals
73  */
74 /* prototype memory vattrs */
75 vattr_t sdev_vattr_dir = {
76 	AT_TYPE|AT_MODE|AT_UID|AT_GID,		/* va_mask */
77 	VDIR,					/* va_type */
78 	SDEV_DIRMODE_DEFAULT,			/* va_mode */
79 	SDEV_UID_DEFAULT,			/* va_uid */
80 	SDEV_GID_DEFAULT,			/* va_gid */
81 	0,					/* va_fsid */
82 	0,					/* va_nodeid */
83 	0,					/* va_nlink */
84 	0,					/* va_size */
85 	0,					/* va_atime */
86 	0,					/* va_mtime */
87 	0,					/* va_ctime */
88 	0,					/* va_rdev */
89 	0,					/* va_blksize */
90 	0,					/* va_nblocks */
91 	0					/* va_vcode */
92 };
93 
94 vattr_t sdev_vattr_lnk = {
95 	AT_TYPE|AT_MODE,			/* va_mask */
96 	VLNK,					/* va_type */
97 	SDEV_LNKMODE_DEFAULT,			/* va_mode */
98 	SDEV_UID_DEFAULT,			/* va_uid */
99 	SDEV_GID_DEFAULT,			/* va_gid */
100 	0,					/* va_fsid */
101 	0,					/* va_nodeid */
102 	0,					/* va_nlink */
103 	0,					/* va_size */
104 	0,					/* va_atime */
105 	0,					/* va_mtime */
106 	0,					/* va_ctime */
107 	0,					/* va_rdev */
108 	0,					/* va_blksize */
109 	0,					/* va_nblocks */
110 	0					/* va_vcode */
111 };
112 
113 vattr_t sdev_vattr_blk = {
114 	AT_TYPE|AT_MODE|AT_UID|AT_GID,		/* va_mask */
115 	VBLK,					/* va_type */
116 	S_IFBLK | SDEV_DEVMODE_DEFAULT,		/* va_mode */
117 	SDEV_UID_DEFAULT,			/* va_uid */
118 	SDEV_GID_DEFAULT,			/* va_gid */
119 	0,					/* va_fsid */
120 	0,					/* va_nodeid */
121 	0,					/* va_nlink */
122 	0,					/* va_size */
123 	0,					/* va_atime */
124 	0,					/* va_mtime */
125 	0,					/* va_ctime */
126 	0,					/* va_rdev */
127 	0,					/* va_blksize */
128 	0,					/* va_nblocks */
129 	0					/* va_vcode */
130 };
131 
132 vattr_t sdev_vattr_chr = {
133 	AT_TYPE|AT_MODE|AT_UID|AT_GID,		/* va_mask */
134 	VCHR,					/* va_type */
135 	S_IFCHR | SDEV_DEVMODE_DEFAULT,		/* va_mode */
136 	SDEV_UID_DEFAULT,			/* va_uid */
137 	SDEV_GID_DEFAULT,			/* va_gid */
138 	0,					/* va_fsid */
139 	0,					/* va_nodeid */
140 	0,					/* va_nlink */
141 	0,					/* va_size */
142 	0,					/* va_atime */
143 	0,					/* va_mtime */
144 	0,					/* va_ctime */
145 	0,					/* va_rdev */
146 	0,					/* va_blksize */
147 	0,					/* va_nblocks */
148 	0					/* va_vcode */
149 };
150 
151 kmem_cache_t	*sdev_node_cache;	/* sdev_node cache */
152 int		devtype;		/* fstype */
153 
154 static void
155 sdev_prof_free(struct sdev_node *dv)
156 {
157 	ASSERT(!SDEV_IS_GLOBAL(dv));
158 	nvlist_free(dv->sdev_prof.dev_name);
159 	nvlist_free(dv->sdev_prof.dev_map);
160 	nvlist_free(dv->sdev_prof.dev_symlink);
161 	nvlist_free(dv->sdev_prof.dev_glob_incdir);
162 	nvlist_free(dv->sdev_prof.dev_glob_excdir);
163 	bzero(&dv->sdev_prof, sizeof (dv->sdev_prof));
164 }
165 
166 /* sdev_node cache constructor */
167 /*ARGSUSED1*/
168 static int
169 i_sdev_node_ctor(void *buf, void *cfarg, int flag)
170 {
171 	struct sdev_node *dv = (struct sdev_node *)buf;
172 	struct vnode *vp;
173 
174 	bzero(buf, sizeof (struct sdev_node));
175 	vp = dv->sdev_vnode = vn_alloc(flag);
176 	if (vp == NULL) {
177 		return (-1);
178 	}
179 	vp->v_data = dv;
180 	rw_init(&dv->sdev_contents, NULL, RW_DEFAULT, NULL);
181 	return (0);
182 }
183 
184 /* sdev_node cache destructor */
185 /*ARGSUSED1*/
186 static void
187 i_sdev_node_dtor(void *buf, void *arg)
188 {
189 	struct sdev_node *dv = (struct sdev_node *)buf;
190 	struct vnode *vp = SDEVTOV(dv);
191 
192 	rw_destroy(&dv->sdev_contents);
193 	vn_free(vp);
194 }
195 
196 /* initialize sdev_node cache */
197 void
198 sdev_node_cache_init()
199 {
200 	int flags = 0;
201 
202 #ifdef	DEBUG
203 	flags = sdev_debug_cache_flags;
204 	if (flags)
205 		sdcmn_err(("cache debug flags 0x%x\n", flags));
206 #endif	/* DEBUG */
207 
208 	ASSERT(sdev_node_cache == NULL);
209 	sdev_node_cache = kmem_cache_create("sdev_node_cache",
210 	    sizeof (struct sdev_node), 0, i_sdev_node_ctor, i_sdev_node_dtor,
211 	    NULL, NULL, NULL, flags);
212 }
213 
214 /* destroy sdev_node cache */
215 void
216 sdev_node_cache_fini()
217 {
218 	ASSERT(sdev_node_cache != NULL);
219 	kmem_cache_destroy(sdev_node_cache);
220 	sdev_node_cache = NULL;
221 }
222 
223 /*
224  * Compare two nodes lexographically to balance avl tree
225  */
226 static int
227 sdev_compare_nodes(const struct sdev_node *dv1, const struct sdev_node *dv2)
228 {
229 	int rv;
230 	if ((rv = strcmp(dv1->sdev_name, dv2->sdev_name)) == 0)
231 		return (0);
232 	return ((rv < 0) ? -1 : 1);
233 }
234 
235 void
236 sdev_set_nodestate(struct sdev_node *dv, sdev_node_state_t state)
237 {
238 	ASSERT(dv);
239 	ASSERT(RW_WRITE_HELD(&dv->sdev_contents));
240 	dv->sdev_state = state;
241 }
242 
243 static void
244 sdev_attr_update(struct sdev_node *dv, vattr_t *vap)
245 {
246 	timestruc_t	now;
247 	struct vattr	*attrp;
248 	uint_t		mask;
249 
250 	ASSERT(dv->sdev_attr);
251 	ASSERT(vap);
252 
253 	attrp = dv->sdev_attr;
254 	mask = vap->va_mask;
255 	if (mask & AT_TYPE)
256 		attrp->va_type = vap->va_type;
257 	if (mask & AT_MODE)
258 		attrp->va_mode = vap->va_mode;
259 	if (mask & AT_UID)
260 		attrp->va_uid = vap->va_uid;
261 	if (mask & AT_GID)
262 		attrp->va_gid = vap->va_gid;
263 	if (mask & AT_RDEV)
264 		attrp->va_rdev = vap->va_rdev;
265 
266 	gethrestime(&now);
267 	attrp->va_atime = (mask & AT_ATIME) ? vap->va_atime : now;
268 	attrp->va_mtime = (mask & AT_MTIME) ? vap->va_mtime : now;
269 	attrp->va_ctime = (mask & AT_CTIME) ? vap->va_ctime : now;
270 }
271 
272 static void
273 sdev_attr_alloc(struct sdev_node *dv, vattr_t *vap)
274 {
275 	ASSERT(dv->sdev_attr == NULL);
276 	ASSERT(vap->va_mask & AT_TYPE);
277 	ASSERT(vap->va_mask & AT_MODE);
278 
279 	dv->sdev_attr = kmem_zalloc(sizeof (struct vattr), KM_SLEEP);
280 	sdev_attr_update(dv, vap);
281 }
282 
283 /* alloc and initialize a sdev_node */
284 int
285 sdev_nodeinit(struct sdev_node *ddv, char *nm, struct sdev_node **newdv,
286     vattr_t *vap)
287 {
288 	struct sdev_node *dv = NULL;
289 	struct vnode *vp;
290 	size_t nmlen, len;
291 	devname_handle_t  *dhl;
292 
293 	nmlen = strlen(nm) + 1;
294 	if (nmlen > MAXNAMELEN) {
295 		sdcmn_err9(("sdev_nodeinit: node name %s"
296 		    " too long\n", nm));
297 		*newdv = NULL;
298 		return (ENAMETOOLONG);
299 	}
300 
301 	dv = kmem_cache_alloc(sdev_node_cache, KM_SLEEP);
302 
303 	dv->sdev_name = kmem_alloc(nmlen, KM_SLEEP);
304 	bcopy(nm, dv->sdev_name, nmlen);
305 	dv->sdev_namelen = nmlen - 1;	/* '\0' not included */
306 	len = strlen(ddv->sdev_path) + strlen(nm) + 2;
307 	dv->sdev_path = kmem_alloc(len, KM_SLEEP);
308 	(void) snprintf(dv->sdev_path, len, "%s/%s", ddv->sdev_path, nm);
309 	/* overwritten for VLNK nodes */
310 	dv->sdev_symlink = NULL;
311 	list_link_init(&dv->sdev_plist);
312 
313 	vp = SDEVTOV(dv);
314 	vn_reinit(vp);
315 	vp->v_vfsp = SDEVTOV(ddv)->v_vfsp;
316 	if (vap)
317 		vp->v_type = vap->va_type;
318 
319 	/*
320 	 * initialized to the parent's vnodeops.
321 	 * maybe overwriten for a VDIR
322 	 */
323 	vn_setops(vp, vn_getops(SDEVTOV(ddv)));
324 	vn_exists(vp);
325 
326 	dv->sdev_dotdot = NULL;
327 	dv->sdev_attrvp = NULL;
328 	if (vap) {
329 		sdev_attr_alloc(dv, vap);
330 	} else {
331 		dv->sdev_attr = NULL;
332 	}
333 
334 	dv->sdev_ino = sdev_mkino(dv);
335 	dv->sdev_nlink = 0;		/* updated on insert */
336 	dv->sdev_flags = ddv->sdev_flags; /* inherit from the parent first */
337 	dv->sdev_flags |= SDEV_BUILD;
338 	mutex_init(&dv->sdev_lookup_lock, NULL, MUTEX_DEFAULT, NULL);
339 	cv_init(&dv->sdev_lookup_cv, NULL, CV_DEFAULT, NULL);
340 	if (SDEV_IS_GLOBAL(ddv)) {
341 		dv->sdev_flags |= SDEV_GLOBAL;
342 		dhl = &(dv->sdev_handle);
343 		dhl->dh_data = dv;
344 		dhl->dh_args = NULL;
345 		sdev_set_no_negcache(dv);
346 		dv->sdev_gdir_gen = 0;
347 	} else {
348 		dv->sdev_flags &= ~SDEV_GLOBAL;
349 		dv->sdev_origin = NULL; /* set later */
350 		bzero(&dv->sdev_prof, sizeof (dv->sdev_prof));
351 		dv->sdev_ldir_gen = 0;
352 		dv->sdev_devtree_gen = 0;
353 	}
354 
355 	rw_enter(&dv->sdev_contents, RW_WRITER);
356 	sdev_set_nodestate(dv, SDEV_INIT);
357 	rw_exit(&dv->sdev_contents);
358 	*newdv = dv;
359 
360 	return (0);
361 }
362 
363 /*
364  * Transition a sdev_node into SDEV_READY state. If this fails, it is up to the
365  * caller to transition the node to the SDEV_ZOMBIE state.
366  */
367 int
368 sdev_nodeready(struct sdev_node *dv, struct vattr *vap, struct vnode *avp,
369     void *args, struct cred *cred)
370 {
371 	int error = 0;
372 	struct vnode *vp = SDEVTOV(dv);
373 	vtype_t type;
374 
375 	ASSERT(dv && (dv->sdev_state != SDEV_READY) && vap);
376 
377 	type = vap->va_type;
378 	vp->v_type = type;
379 	vp->v_rdev = vap->va_rdev;
380 	rw_enter(&dv->sdev_contents, RW_WRITER);
381 	if (type == VDIR) {
382 		dv->sdev_nlink = 2;
383 		dv->sdev_flags &= ~SDEV_PERSIST;
384 		dv->sdev_flags &= ~SDEV_DYNAMIC;
385 		vn_setops(vp, sdev_get_vop(dv)); /* from internal vtab */
386 		ASSERT(dv->sdev_dotdot);
387 		ASSERT(SDEVTOV(dv->sdev_dotdot)->v_type == VDIR);
388 		vp->v_rdev = SDEVTOV(dv->sdev_dotdot)->v_rdev;
389 		avl_create(&dv->sdev_entries,
390 		    (int (*)(const void *, const void *))sdev_compare_nodes,
391 		    sizeof (struct sdev_node),
392 		    offsetof(struct sdev_node, sdev_avllink));
393 	} else if (type == VLNK) {
394 		ASSERT(args);
395 		dv->sdev_nlink = 1;
396 		dv->sdev_symlink = i_ddi_strdup((char *)args, KM_SLEEP);
397 	} else {
398 		dv->sdev_nlink = 1;
399 	}
400 	sdev_plugin_nodeready(dv);
401 
402 	if (!(SDEV_IS_GLOBAL(dv))) {
403 		dv->sdev_origin = (struct sdev_node *)args;
404 		dv->sdev_flags &= ~SDEV_PERSIST;
405 	}
406 
407 	/*
408 	 * shadow node is created here OR
409 	 * if failed (indicated by dv->sdev_attrvp == NULL),
410 	 * created later in sdev_setattr
411 	 */
412 	if (avp) {
413 		dv->sdev_attrvp = avp;
414 	} else {
415 		if (dv->sdev_attr == NULL) {
416 			sdev_attr_alloc(dv, vap);
417 		} else {
418 			sdev_attr_update(dv, vap);
419 		}
420 
421 		if ((dv->sdev_attrvp == NULL) && SDEV_IS_PERSIST(dv))
422 			error = sdev_shadow_node(dv, cred);
423 	}
424 
425 	if (error == 0) {
426 		/* transition to READY state */
427 		sdev_set_nodestate(dv, SDEV_READY);
428 		sdev_nc_node_exists(dv);
429 	}
430 	rw_exit(&dv->sdev_contents);
431 	return (error);
432 }
433 
434 /*
435  * Build the VROOT sdev_node.
436  */
437 /*ARGSUSED*/
438 struct sdev_node *
439 sdev_mkroot(struct vfs *vfsp, dev_t devdev, struct vnode *mvp,
440     struct vnode *avp, struct cred *cred)
441 {
442 	struct sdev_node *dv;
443 	struct vnode *vp;
444 	char devdir[] = "/dev";
445 
446 	ASSERT(sdev_node_cache != NULL);
447 	ASSERT(avp);
448 	dv = kmem_cache_alloc(sdev_node_cache, KM_SLEEP);
449 	vp = SDEVTOV(dv);
450 	vn_reinit(vp);
451 	vp->v_flag |= VROOT;
452 	vp->v_vfsp = vfsp;
453 	vp->v_type = VDIR;
454 	vp->v_rdev = devdev;
455 	vn_setops(vp, sdev_vnodeops); /* apply the default vnodeops at /dev */
456 	vn_exists(vp);
457 
458 	if (vfsp->vfs_mntpt)
459 		dv->sdev_name = i_ddi_strdup(
460 		    (char *)refstr_value(vfsp->vfs_mntpt), KM_SLEEP);
461 	else
462 		/* vfs_mountdev1 set mount point later */
463 		dv->sdev_name = i_ddi_strdup("/dev", KM_SLEEP);
464 	dv->sdev_namelen = strlen(dv->sdev_name); /* '\0' not included */
465 	dv->sdev_path = i_ddi_strdup(devdir, KM_SLEEP);
466 	dv->sdev_ino = SDEV_ROOTINO;
467 	dv->sdev_nlink = 2;		/* name + . (no sdev_insert) */
468 	dv->sdev_dotdot = dv;		/* .. == self */
469 	dv->sdev_attrvp = avp;
470 	dv->sdev_attr = NULL;
471 	mutex_init(&dv->sdev_lookup_lock, NULL, MUTEX_DEFAULT, NULL);
472 	cv_init(&dv->sdev_lookup_cv, NULL, CV_DEFAULT, NULL);
473 	if (strcmp(dv->sdev_name, "/dev") == 0) {
474 		dv->sdev_flags = SDEV_BUILD|SDEV_GLOBAL|SDEV_PERSIST;
475 		bzero(&dv->sdev_handle, sizeof (dv->sdev_handle));
476 		dv->sdev_gdir_gen = 0;
477 	} else {
478 		dv->sdev_flags = SDEV_BUILD;
479 		dv->sdev_flags &= ~SDEV_PERSIST;
480 		bzero(&dv->sdev_prof, sizeof (dv->sdev_prof));
481 		dv->sdev_ldir_gen = 0;
482 		dv->sdev_devtree_gen = 0;
483 	}
484 
485 	avl_create(&dv->sdev_entries,
486 	    (int (*)(const void *, const void *))sdev_compare_nodes,
487 	    sizeof (struct sdev_node),
488 	    offsetof(struct sdev_node, sdev_avllink));
489 
490 	rw_enter(&dv->sdev_contents, RW_WRITER);
491 	sdev_set_nodestate(dv, SDEV_READY);
492 	rw_exit(&dv->sdev_contents);
493 	sdev_nc_node_exists(dv);
494 	return (dv);
495 }
496 
497 struct sdev_vop_table vtab[] = {
498 	{ "pts", devpts_vnodeops_tbl, &devpts_vnodeops, devpts_validate,
499 	SDEV_DYNAMIC | SDEV_VTOR },
500 
501 	{ "vt", devvt_vnodeops_tbl, &devvt_vnodeops, devvt_validate,
502 	SDEV_DYNAMIC | SDEV_VTOR },
503 
504 	{ "zvol", devzvol_vnodeops_tbl, &devzvol_vnodeops,
505 	devzvol_validate, SDEV_ZONED | SDEV_DYNAMIC | SDEV_VTOR | SDEV_SUBDIR },
506 
507 	{ "zcons", NULL, NULL, NULL, SDEV_NO_NCACHE },
508 
509 	{ "net", devnet_vnodeops_tbl, &devnet_vnodeops, devnet_validate,
510 	SDEV_DYNAMIC | SDEV_VTOR | SDEV_SUBDIR },
511 
512 	{ "ipnet", devipnet_vnodeops_tbl, &devipnet_vnodeops,
513 	devipnet_validate, SDEV_DYNAMIC | SDEV_VTOR | SDEV_NO_NCACHE },
514 
515 	/*
516 	 * SDEV_DYNAMIC: prevent calling out to devfsadm, since only the
517 	 * lofi driver controls child nodes.
518 	 *
519 	 * SDEV_PERSIST: ensure devfsadm knows to clean up any persisted
520 	 * stale nodes (e.g. from devfsadm -R).
521 	 *
522 	 * In addition, devfsadm knows not to attempt a rmdir: a zone
523 	 * may hold a reference, which would zombify the node,
524 	 * preventing a mkdir.
525 	 */
526 
527 	{ "lofi", NULL, NULL, NULL,
528 	    SDEV_ZONED | SDEV_DYNAMIC | SDEV_PERSIST },
529 	{ "rlofi", NULL, NULL, NULL,
530 	    SDEV_ZONED | SDEV_DYNAMIC | SDEV_PERSIST },
531 
532 	{ NULL, NULL, NULL, NULL, 0}
533 };
534 
535 
536 /*
537  * Build the base root inode
538  */
539 ino_t
540 sdev_mkino(struct sdev_node *dv)
541 {
542 	ino_t	ino;
543 
544 	/*
545 	 * for now, follow the lead of tmpfs here
546 	 * need to someday understand the requirements here
547 	 */
548 	ino = (ino_t)(uint32_t)((uintptr_t)dv >> 3);
549 	ino += SDEV_ROOTINO + 1;
550 
551 	return (ino);
552 }
553 
554 int
555 sdev_getlink(struct vnode *linkvp, char **link)
556 {
557 	int err;
558 	char *buf;
559 	struct uio uio = {0};
560 	struct iovec iov = {0};
561 
562 	if (linkvp == NULL)
563 		return (ENOENT);
564 	ASSERT(linkvp->v_type == VLNK);
565 
566 	buf = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
567 	iov.iov_base = buf;
568 	iov.iov_len = MAXPATHLEN;
569 	uio.uio_iov = &iov;
570 	uio.uio_iovcnt = 1;
571 	uio.uio_resid = MAXPATHLEN;
572 	uio.uio_segflg = UIO_SYSSPACE;
573 	uio.uio_llimit = MAXOFFSET_T;
574 
575 	err = VOP_READLINK(linkvp, &uio, kcred, NULL);
576 	if (err) {
577 		cmn_err(CE_WARN, "readlink %s failed in dev\n", buf);
578 		kmem_free(buf, MAXPATHLEN);
579 		return (ENOENT);
580 	}
581 
582 	/* mission complete */
583 	*link = i_ddi_strdup(buf, KM_SLEEP);
584 	kmem_free(buf, MAXPATHLEN);
585 	return (0);
586 }
587 
588 /*
589  * A convenient wrapper to get the devfs node vnode for a device
590  * minor functionality: readlink() of a /dev symlink
591  * Place the link into dv->sdev_symlink
592  */
593 static int
594 sdev_follow_link(struct sdev_node *dv)
595 {
596 	int err;
597 	struct vnode *linkvp;
598 	char *link = NULL;
599 
600 	linkvp = SDEVTOV(dv);
601 	if (linkvp == NULL)
602 		return (ENOENT);
603 	ASSERT(linkvp->v_type == VLNK);
604 	err = sdev_getlink(linkvp, &link);
605 	if (err) {
606 		dv->sdev_symlink = NULL;
607 		return (ENOENT);
608 	}
609 
610 	ASSERT(link != NULL);
611 	dv->sdev_symlink = link;
612 	return (0);
613 }
614 
615 static int
616 sdev_node_check(struct sdev_node *dv, struct vattr *nvap, void *nargs)
617 {
618 	vtype_t otype = SDEVTOV(dv)->v_type;
619 
620 	/*
621 	 * existing sdev_node has a different type.
622 	 */
623 	if (otype != nvap->va_type) {
624 		sdcmn_err9(("sdev_node_check: existing node "
625 		    "  %s type %d does not match new node type %d\n",
626 		    dv->sdev_name, otype, nvap->va_type));
627 		return (EEXIST);
628 	}
629 
630 	/*
631 	 * For a symlink, the target should be the same.
632 	 */
633 	if (otype == VLNK) {
634 		ASSERT(nargs != NULL);
635 		ASSERT(dv->sdev_symlink != NULL);
636 		if (strcmp(dv->sdev_symlink, (char *)nargs) != 0) {
637 			sdcmn_err9(("sdev_node_check: existing node "
638 			    " %s has different symlink %s as new node "
639 			    " %s\n", dv->sdev_name, dv->sdev_symlink,
640 			    (char *)nargs));
641 			return (EEXIST);
642 		}
643 	}
644 
645 	return (0);
646 }
647 
648 /*
649  * sdev_mknode - a wrapper for sdev_nodeinit(), sdev_nodeready()
650  *
651  * arguments:
652  *	- ddv (parent)
653  *	- nm (child name)
654  *	- newdv (sdev_node for nm is returned here)
655  *	- vap (vattr for the node to be created, va_type should be set.
656  *	- avp (attribute vnode)
657  *	  the defaults should be used if unknown)
658  *	- cred
659  *	- args
660  *	    . tnm (for VLNK)
661  *	    . global sdev_node (for !SDEV_GLOBAL)
662  * 	- state: SDEV_INIT, SDEV_READY
663  *
664  * only ddv, nm, newddv, vap, cred are required for sdev_mknode(SDEV_INIT)
665  *
666  * NOTE:  directory contents writers lock needs to be held before
667  *	  calling this routine.
668  */
669 int
670 sdev_mknode(struct sdev_node *ddv, char *nm, struct sdev_node **newdv,
671     struct vattr *vap, struct vnode *avp, void *args, struct cred *cred,
672     sdev_node_state_t state)
673 {
674 	int error = 0;
675 	sdev_node_state_t node_state;
676 	struct sdev_node *dv = NULL;
677 
678 	ASSERT(state != SDEV_ZOMBIE);
679 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
680 
681 	if (*newdv) {
682 		dv = *newdv;
683 	} else {
684 		/* allocate and initialize a sdev_node */
685 		if (ddv->sdev_state == SDEV_ZOMBIE) {
686 			sdcmn_err9(("sdev_mknode: parent %s ZOMBIEd\n",
687 			    ddv->sdev_path));
688 			return (ENOENT);
689 		}
690 
691 		error = sdev_nodeinit(ddv, nm, &dv, vap);
692 		if (error != 0) {
693 			sdcmn_err9(("sdev_mknode: error %d,"
694 			    " name %s can not be initialized\n",
695 			    error, nm));
696 			return (error);
697 		}
698 		ASSERT(dv);
699 
700 		/* insert into the directory cache */
701 		sdev_cache_update(ddv, &dv, nm, SDEV_CACHE_ADD);
702 	}
703 
704 	ASSERT(dv);
705 	node_state = dv->sdev_state;
706 	ASSERT(node_state != SDEV_ZOMBIE);
707 
708 	if (state == SDEV_READY) {
709 		switch (node_state) {
710 		case SDEV_INIT:
711 			error = sdev_nodeready(dv, vap, avp, args, cred);
712 			if (error) {
713 				sdcmn_err9(("sdev_mknode: node %s can NOT"
714 				    " be transitioned into READY state, "
715 				    "error %d\n", nm, error));
716 			}
717 			break;
718 		case SDEV_READY:
719 			/*
720 			 * Do some sanity checking to make sure
721 			 * the existing sdev_node is what has been
722 			 * asked for.
723 			 */
724 			error = sdev_node_check(dv, vap, args);
725 			break;
726 		default:
727 			break;
728 		}
729 	}
730 
731 	if (!error) {
732 		*newdv = dv;
733 		ASSERT((*newdv)->sdev_state != SDEV_ZOMBIE);
734 	} else {
735 		sdev_cache_update(ddv, &dv, nm, SDEV_CACHE_DELETE);
736 		/*
737 		 * We created this node, it wasn't passed into us. Therefore it
738 		 * is up to us to delete it.
739 		 */
740 		if (*newdv == NULL)
741 			SDEV_SIMPLE_RELE(dv);
742 		*newdv = NULL;
743 	}
744 
745 	return (error);
746 }
747 
748 /*
749  * convenient wrapper to change vp's ATIME, CTIME and MTIME
750  */
751 void
752 sdev_update_timestamps(struct vnode *vp, cred_t *cred, uint_t mask)
753 {
754 	struct vattr attr;
755 	timestruc_t now;
756 	int err;
757 
758 	ASSERT(vp);
759 	gethrestime(&now);
760 	if (mask & AT_CTIME)
761 		attr.va_ctime = now;
762 	if (mask & AT_MTIME)
763 		attr.va_mtime = now;
764 	if (mask & AT_ATIME)
765 		attr.va_atime = now;
766 
767 	attr.va_mask = (mask & AT_TIMES);
768 	err = VOP_SETATTR(vp, &attr, 0, cred, NULL);
769 	if (err && (err != EROFS)) {
770 		sdcmn_err(("update timestamps error %d\n", err));
771 	}
772 }
773 
774 /*
775  * the backing store vnode is released here
776  */
777 /*ARGSUSED1*/
778 void
779 sdev_nodedestroy(struct sdev_node *dv, uint_t flags)
780 {
781 	/* no references */
782 	ASSERT(dv->sdev_nlink == 0);
783 
784 	if (dv->sdev_attrvp != NULLVP) {
785 		VN_RELE(dv->sdev_attrvp);
786 		/*
787 		 * reset the attrvp so that no more
788 		 * references can be made on this already
789 		 * vn_rele() vnode
790 		 */
791 		dv->sdev_attrvp = NULLVP;
792 	}
793 
794 	if (dv->sdev_attr != NULL) {
795 		kmem_free(dv->sdev_attr, sizeof (struct vattr));
796 		dv->sdev_attr = NULL;
797 	}
798 
799 	if (dv->sdev_name != NULL) {
800 		kmem_free(dv->sdev_name, dv->sdev_namelen + 1);
801 		dv->sdev_name = NULL;
802 	}
803 
804 	if (dv->sdev_symlink != NULL) {
805 		kmem_free(dv->sdev_symlink, strlen(dv->sdev_symlink) + 1);
806 		dv->sdev_symlink = NULL;
807 	}
808 
809 	if (dv->sdev_path) {
810 		kmem_free(dv->sdev_path, strlen(dv->sdev_path) + 1);
811 		dv->sdev_path = NULL;
812 	}
813 
814 	if (!SDEV_IS_GLOBAL(dv)) {
815 		sdev_prof_free(dv);
816 		if (dv->sdev_vnode->v_type != VLNK && dv->sdev_origin != NULL)
817 			SDEV_RELE(dv->sdev_origin);
818 	}
819 
820 	if (SDEVTOV(dv)->v_type == VDIR) {
821 		ASSERT(SDEV_FIRST_ENTRY(dv) == NULL);
822 		avl_destroy(&dv->sdev_entries);
823 	}
824 
825 	mutex_destroy(&dv->sdev_lookup_lock);
826 	cv_destroy(&dv->sdev_lookup_cv);
827 
828 	/* return node to initial state as per constructor */
829 	(void) memset((void *)&dv->sdev_instance_data, 0,
830 	    sizeof (dv->sdev_instance_data));
831 	vn_invalid(SDEVTOV(dv));
832 	dv->sdev_private = NULL;
833 	kmem_cache_free(sdev_node_cache, dv);
834 }
835 
836 /*
837  * DIRECTORY CACHE lookup
838  */
839 struct sdev_node *
840 sdev_findbyname(struct sdev_node *ddv, char *nm)
841 {
842 	struct sdev_node *dv;
843 	struct sdev_node dvtmp;
844 	avl_index_t	where;
845 
846 	ASSERT(RW_LOCK_HELD(&ddv->sdev_contents));
847 
848 	dvtmp.sdev_name = nm;
849 	dv = avl_find(&ddv->sdev_entries, &dvtmp, &where);
850 	if (dv) {
851 		ASSERT(dv->sdev_dotdot == ddv);
852 		ASSERT(strcmp(dv->sdev_name, nm) == 0);
853 		ASSERT(dv->sdev_state != SDEV_ZOMBIE);
854 		SDEV_HOLD(dv);
855 		return (dv);
856 	}
857 	return (NULL);
858 }
859 
860 /*
861  * Inserts a new sdev_node in a parent directory
862  */
863 void
864 sdev_direnter(struct sdev_node *ddv, struct sdev_node *dv)
865 {
866 	avl_index_t where;
867 
868 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
869 	ASSERT(SDEVTOV(ddv)->v_type == VDIR);
870 	ASSERT(ddv->sdev_nlink >= 2);
871 	ASSERT(dv->sdev_nlink == 0);
872 	ASSERT(dv->sdev_state != SDEV_ZOMBIE);
873 
874 	dv->sdev_dotdot = ddv;
875 	VERIFY(avl_find(&ddv->sdev_entries, dv, &where) == NULL);
876 	avl_insert(&ddv->sdev_entries, dv, where);
877 	ddv->sdev_nlink++;
878 }
879 
880 /*
881  * The following check is needed because while sdev_nodes are linked
882  * in SDEV_INIT state, they have their link counts incremented only
883  * in SDEV_READY state.
884  */
885 static void
886 decr_link(struct sdev_node *dv)
887 {
888 	VERIFY(RW_WRITE_HELD(&dv->sdev_contents));
889 	if (dv->sdev_state != SDEV_INIT) {
890 		VERIFY(dv->sdev_nlink >= 1);
891 		dv->sdev_nlink--;
892 	} else {
893 		VERIFY(dv->sdev_nlink == 0);
894 	}
895 }
896 
897 /*
898  * Delete an existing dv from directory cache
899  *
900  * In the case of a node is still held by non-zero reference count, the node is
901  * put into ZOMBIE state. The node is always unlinked from its parent, but it is
902  * not destroyed via sdev_inactive until its reference count reaches "0".
903  */
904 static void
905 sdev_dirdelete(struct sdev_node *ddv, struct sdev_node *dv)
906 {
907 	struct vnode *vp;
908 	sdev_node_state_t os;
909 
910 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
911 
912 	vp = SDEVTOV(dv);
913 	mutex_enter(&vp->v_lock);
914 	rw_enter(&dv->sdev_contents, RW_WRITER);
915 	os = dv->sdev_state;
916 	ASSERT(os != SDEV_ZOMBIE);
917 	dv->sdev_state = SDEV_ZOMBIE;
918 
919 	/*
920 	 * unlink ourselves from the parent directory now to take care of the ..
921 	 * link. However, if we're a directory, we don't remove our reference to
922 	 * ourself eg. '.' until we are torn down in the inactive callback.
923 	 */
924 	decr_link(ddv);
925 	avl_remove(&ddv->sdev_entries, dv);
926 	/*
927 	 * sdev_inactive expects nodes to have a link to themselves when we're
928 	 * tearing them down. If we're transitioning from the initial state to
929 	 * zombie and not via ready, then we're not going to have this link that
930 	 * comes from the node being ready. As a result, we need to increment
931 	 * our link count by one to account for this.
932 	 */
933 	if (os == SDEV_INIT && dv->sdev_nlink == 0)
934 		dv->sdev_nlink++;
935 	rw_exit(&dv->sdev_contents);
936 	mutex_exit(&vp->v_lock);
937 }
938 
939 /*
940  * check if the source is in the path of the target
941  *
942  * source and target are different
943  */
944 /*ARGSUSED2*/
945 static int
946 sdev_checkpath(struct sdev_node *sdv, struct sdev_node *tdv, struct cred *cred)
947 {
948 	int error = 0;
949 	struct sdev_node *dotdot, *dir;
950 
951 	dotdot = tdv->sdev_dotdot;
952 	ASSERT(dotdot);
953 
954 	/* fs root */
955 	if (dotdot == tdv) {
956 		return (0);
957 	}
958 
959 	for (;;) {
960 		/*
961 		 * avoid error cases like
962 		 *	mv a a/b
963 		 *	mv a a/b/c
964 		 *	etc.
965 		 */
966 		if (dotdot == sdv) {
967 			error = EINVAL;
968 			break;
969 		}
970 
971 		dir = dotdot;
972 		dotdot = dir->sdev_dotdot;
973 
974 		/* done checking because root is reached */
975 		if (dir == dotdot) {
976 			break;
977 		}
978 	}
979 	return (error);
980 }
981 
982 int
983 sdev_rnmnode(struct sdev_node *oddv, struct sdev_node *odv,
984     struct sdev_node *nddv, struct sdev_node **ndvp, char *nnm,
985     struct cred *cred)
986 {
987 	int error = 0;
988 	struct vnode *ovp = SDEVTOV(odv);
989 	struct vnode *nvp;
990 	struct vattr vattr;
991 	int doingdir = (ovp->v_type == VDIR);
992 	char *link = NULL;
993 	int samedir = (oddv == nddv) ? 1 : 0;
994 	int bkstore = 0;
995 	struct sdev_node *idv = NULL;
996 	struct sdev_node *ndv = NULL;
997 	timestruc_t now;
998 
999 	vattr.va_mask = AT_TYPE|AT_MODE|AT_UID|AT_GID;
1000 	error = VOP_GETATTR(ovp, &vattr, 0, cred, NULL);
1001 	if (error)
1002 		return (error);
1003 
1004 	if (!samedir)
1005 		rw_enter(&oddv->sdev_contents, RW_WRITER);
1006 	rw_enter(&nddv->sdev_contents, RW_WRITER);
1007 
1008 	/*
1009 	 * the source may have been deleted by another thread before
1010 	 * we gets here.
1011 	 */
1012 	if (odv->sdev_state != SDEV_READY) {
1013 		error = ENOENT;
1014 		goto err_out;
1015 	}
1016 
1017 	if (doingdir && (odv == nddv)) {
1018 		error = EINVAL;
1019 		goto err_out;
1020 	}
1021 
1022 	/*
1023 	 * If renaming a directory, and the parents are different (".." must be
1024 	 * changed) then the source dir must not be in the dir hierarchy above
1025 	 * the target since it would orphan everything below the source dir.
1026 	 */
1027 	if (doingdir && (oddv != nddv)) {
1028 		error = sdev_checkpath(odv, nddv, cred);
1029 		if (error)
1030 			goto err_out;
1031 	}
1032 
1033 	/* fix the source for a symlink */
1034 	if (vattr.va_type == VLNK) {
1035 		if (odv->sdev_symlink == NULL) {
1036 			error = sdev_follow_link(odv);
1037 			if (error) {
1038 				/*
1039 				 * The underlying symlink doesn't exist. This
1040 				 * node probably shouldn't even exist. While
1041 				 * it's a bit jarring to consumers, we're going
1042 				 * to remove the node from /dev.
1043 				 */
1044 				if (SDEV_IS_PERSIST((*ndvp)))
1045 					bkstore = 1;
1046 				sdev_dirdelete(oddv, odv);
1047 				if (bkstore) {
1048 					ASSERT(nddv->sdev_attrvp);
1049 					error = VOP_REMOVE(nddv->sdev_attrvp,
1050 					    nnm, cred, NULL, 0);
1051 					if (error)
1052 						goto err_out;
1053 				}
1054 				error = ENOENT;
1055 				goto err_out;
1056 			}
1057 		}
1058 		ASSERT(odv->sdev_symlink);
1059 		link = i_ddi_strdup(odv->sdev_symlink, KM_SLEEP);
1060 	}
1061 
1062 	/* destination existing */
1063 	if (*ndvp) {
1064 		nvp = SDEVTOV(*ndvp);
1065 		ASSERT(nvp);
1066 
1067 		/* handling renaming to itself */
1068 		if (odv == *ndvp) {
1069 			error = 0;
1070 			goto err_out;
1071 		}
1072 
1073 		if (nvp->v_type == VDIR) {
1074 			if (!doingdir) {
1075 				error = EISDIR;
1076 				goto err_out;
1077 			}
1078 
1079 			if (vn_vfswlock(nvp)) {
1080 				error = EBUSY;
1081 				goto err_out;
1082 			}
1083 
1084 			if (vn_mountedvfs(nvp) != NULL) {
1085 				vn_vfsunlock(nvp);
1086 				error = EBUSY;
1087 				goto err_out;
1088 			}
1089 
1090 			/* in case dir1 exists in dir2 and "mv dir1 dir2" */
1091 			if ((*ndvp)->sdev_nlink > 2) {
1092 				vn_vfsunlock(nvp);
1093 				error = EEXIST;
1094 				goto err_out;
1095 			}
1096 			vn_vfsunlock(nvp);
1097 
1098 			/*
1099 			 * We did not place the hold on *ndvp, so even though
1100 			 * we're deleting the node, we should not get rid of our
1101 			 * reference.
1102 			 */
1103 			sdev_dirdelete(nddv, *ndvp);
1104 			*ndvp = NULL;
1105 			ASSERT(nddv->sdev_attrvp);
1106 			error = VOP_RMDIR(nddv->sdev_attrvp, nnm,
1107 			    nddv->sdev_attrvp, cred, NULL, 0);
1108 			if (error)
1109 				goto err_out;
1110 		} else {
1111 			if (doingdir) {
1112 				error = ENOTDIR;
1113 				goto err_out;
1114 			}
1115 
1116 			if (SDEV_IS_PERSIST((*ndvp))) {
1117 				bkstore = 1;
1118 			}
1119 
1120 			/*
1121 			 * Get rid of the node from the directory cache note.
1122 			 * Don't forget that it's not up to us to remove the vn
1123 			 * ref on the sdev node, as we did not place it.
1124 			 */
1125 			sdev_dirdelete(nddv, *ndvp);
1126 			*ndvp = NULL;
1127 			if (bkstore) {
1128 				ASSERT(nddv->sdev_attrvp);
1129 				error = VOP_REMOVE(nddv->sdev_attrvp,
1130 				    nnm, cred, NULL, 0);
1131 				if (error)
1132 					goto err_out;
1133 			}
1134 		}
1135 	}
1136 
1137 	/*
1138 	 * make a fresh node from the source attrs
1139 	 */
1140 	ASSERT(RW_WRITE_HELD(&nddv->sdev_contents));
1141 	error = sdev_mknode(nddv, nnm, ndvp, &vattr,
1142 	    NULL, (void *)link, cred, SDEV_READY);
1143 
1144 	if (link != NULL) {
1145 		kmem_free(link, strlen(link) + 1);
1146 		link = NULL;
1147 	}
1148 
1149 	if (error)
1150 		goto err_out;
1151 	ASSERT(*ndvp);
1152 	ASSERT((*ndvp)->sdev_state == SDEV_READY);
1153 
1154 	/* move dir contents */
1155 	if (doingdir) {
1156 		for (idv = SDEV_FIRST_ENTRY(odv); idv;
1157 		    idv = SDEV_NEXT_ENTRY(odv, idv)) {
1158 			SDEV_HOLD(idv);
1159 			error = sdev_rnmnode(odv, idv,
1160 			    (struct sdev_node *)(*ndvp), &ndv,
1161 			    idv->sdev_name, cred);
1162 			SDEV_RELE(idv);
1163 			if (error)
1164 				goto err_out;
1165 			ndv = NULL;
1166 		}
1167 	}
1168 
1169 	if ((*ndvp)->sdev_attrvp) {
1170 		sdev_update_timestamps((*ndvp)->sdev_attrvp, kcred,
1171 		    AT_CTIME|AT_ATIME);
1172 	} else {
1173 		ASSERT((*ndvp)->sdev_attr);
1174 		gethrestime(&now);
1175 		(*ndvp)->sdev_attr->va_ctime = now;
1176 		(*ndvp)->sdev_attr->va_atime = now;
1177 	}
1178 
1179 	if (nddv->sdev_attrvp) {
1180 		sdev_update_timestamps(nddv->sdev_attrvp, kcred,
1181 		    AT_MTIME|AT_ATIME);
1182 	} else {
1183 		ASSERT(nddv->sdev_attr);
1184 		gethrestime(&now);
1185 		nddv->sdev_attr->va_mtime = now;
1186 		nddv->sdev_attr->va_atime = now;
1187 	}
1188 	rw_exit(&nddv->sdev_contents);
1189 	if (!samedir)
1190 		rw_exit(&oddv->sdev_contents);
1191 
1192 	SDEV_RELE(*ndvp);
1193 	return (error);
1194 
1195 err_out:
1196 	if (link != NULL) {
1197 		kmem_free(link, strlen(link) + 1);
1198 		link = NULL;
1199 	}
1200 
1201 	rw_exit(&nddv->sdev_contents);
1202 	if (!samedir)
1203 		rw_exit(&oddv->sdev_contents);
1204 	return (error);
1205 }
1206 
1207 /*
1208  * Merge sdev_node specific information into an attribute structure.
1209  *
1210  * note: sdev_node is not locked here
1211  */
1212 void
1213 sdev_vattr_merge(struct sdev_node *dv, struct vattr *vap)
1214 {
1215 	struct vnode *vp = SDEVTOV(dv);
1216 
1217 	vap->va_nlink = dv->sdev_nlink;
1218 	vap->va_nodeid = dv->sdev_ino;
1219 	vap->va_fsid = SDEVTOV(dv->sdev_dotdot)->v_rdev;
1220 	vap->va_type = vp->v_type;
1221 
1222 	if (vp->v_type == VDIR) {
1223 		vap->va_rdev = 0;
1224 		vap->va_fsid = vp->v_rdev;
1225 	} else if (vp->v_type == VLNK) {
1226 		vap->va_rdev = 0;
1227 		vap->va_mode  &= ~S_IFMT;
1228 		vap->va_mode |= S_IFLNK;
1229 	} else if ((vp->v_type == VCHR) || (vp->v_type == VBLK)) {
1230 		vap->va_rdev = vp->v_rdev;
1231 		vap->va_mode &= ~S_IFMT;
1232 		if (vap->va_type == VCHR)
1233 			vap->va_mode |= S_IFCHR;
1234 		else
1235 			vap->va_mode |= S_IFBLK;
1236 	} else {
1237 		vap->va_rdev = 0;
1238 	}
1239 }
1240 
1241 struct vattr *
1242 sdev_getdefault_attr(enum vtype type)
1243 {
1244 	if (type == VDIR)
1245 		return (&sdev_vattr_dir);
1246 	else if (type == VCHR)
1247 		return (&sdev_vattr_chr);
1248 	else if (type == VBLK)
1249 		return (&sdev_vattr_blk);
1250 	else if (type == VLNK)
1251 		return (&sdev_vattr_lnk);
1252 	else
1253 		return (NULL);
1254 }
1255 int
1256 sdev_to_vp(struct sdev_node *dv, struct vnode **vpp)
1257 {
1258 	int rv = 0;
1259 	struct vnode *vp = SDEVTOV(dv);
1260 
1261 	switch (vp->v_type) {
1262 	case VCHR:
1263 	case VBLK:
1264 		/*
1265 		 * If vnode is a device, return special vnode instead
1266 		 * (though it knows all about -us- via sp->s_realvp)
1267 		 */
1268 		*vpp = specvp(vp, vp->v_rdev, vp->v_type, kcred);
1269 		VN_RELE(vp);
1270 		if (*vpp == NULLVP)
1271 			rv = ENOSYS;
1272 		break;
1273 	default:	/* most types are returned as is */
1274 		*vpp = vp;
1275 		break;
1276 	}
1277 	return (rv);
1278 }
1279 
1280 /*
1281  * junction between devname and root file system, e.g. ufs
1282  */
1283 int
1284 devname_backstore_lookup(struct sdev_node *ddv, char *nm, struct vnode **rvp)
1285 {
1286 	struct vnode *rdvp = ddv->sdev_attrvp;
1287 	int rval = 0;
1288 
1289 	ASSERT(rdvp);
1290 
1291 	rval = VOP_LOOKUP(rdvp, nm, rvp, NULL, 0, NULL, kcred, NULL, NULL,
1292 	    NULL);
1293 	return (rval);
1294 }
1295 
1296 static int
1297 sdev_filldir_from_store(struct sdev_node *ddv, int dlen, struct cred *cred)
1298 {
1299 	struct sdev_node *dv = NULL;
1300 	char	*nm;
1301 	struct vnode *dirvp;
1302 	int	error;
1303 	vnode_t	*vp;
1304 	int eof;
1305 	struct iovec iov;
1306 	struct uio uio;
1307 	struct dirent64 *dp;
1308 	dirent64_t *dbuf;
1309 	size_t dbuflen;
1310 	struct vattr vattr;
1311 	char *link = NULL;
1312 
1313 	if (ddv->sdev_attrvp == NULL)
1314 		return (0);
1315 	if (!(ddv->sdev_flags & SDEV_BUILD))
1316 		return (0);
1317 
1318 	dirvp = ddv->sdev_attrvp;
1319 	VN_HOLD(dirvp);
1320 	dbuf = kmem_zalloc(dlen, KM_SLEEP);
1321 
1322 	uio.uio_iov = &iov;
1323 	uio.uio_iovcnt = 1;
1324 	uio.uio_segflg = UIO_SYSSPACE;
1325 	uio.uio_fmode = 0;
1326 	uio.uio_extflg = UIO_COPY_CACHED;
1327 	uio.uio_loffset = 0;
1328 	uio.uio_llimit = MAXOFFSET_T;
1329 
1330 	eof = 0;
1331 	error = 0;
1332 	while (!error && !eof) {
1333 		uio.uio_resid = dlen;
1334 		iov.iov_base = (char *)dbuf;
1335 		iov.iov_len = dlen;
1336 		(void) VOP_RWLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1337 		error = VOP_READDIR(dirvp, &uio, kcred, &eof, NULL, 0);
1338 		VOP_RWUNLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1339 
1340 		dbuflen = dlen - uio.uio_resid;
1341 		if (error || dbuflen == 0)
1342 			break;
1343 
1344 		if (!(ddv->sdev_flags & SDEV_BUILD))
1345 			break;
1346 
1347 		for (dp = dbuf; ((intptr_t)dp <
1348 		    (intptr_t)dbuf + dbuflen);
1349 		    dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) {
1350 			nm = dp->d_name;
1351 
1352 			if (strcmp(nm, ".") == 0 ||
1353 			    strcmp(nm, "..") == 0)
1354 				continue;
1355 
1356 			vp = NULLVP;
1357 			dv = sdev_cache_lookup(ddv, nm);
1358 			if (dv) {
1359 				VERIFY(dv->sdev_state != SDEV_ZOMBIE);
1360 				SDEV_SIMPLE_RELE(dv);
1361 				continue;
1362 			}
1363 
1364 			/* refill the cache if not already */
1365 			error = devname_backstore_lookup(ddv, nm, &vp);
1366 			if (error)
1367 				continue;
1368 
1369 			vattr.va_mask = AT_TYPE|AT_MODE|AT_UID|AT_GID;
1370 			error = VOP_GETATTR(vp, &vattr, 0, cred, NULL);
1371 			if (error)
1372 				continue;
1373 
1374 			if (vattr.va_type == VLNK) {
1375 				error = sdev_getlink(vp, &link);
1376 				if (error) {
1377 					continue;
1378 				}
1379 				ASSERT(link != NULL);
1380 			}
1381 
1382 			if (!rw_tryupgrade(&ddv->sdev_contents)) {
1383 				rw_exit(&ddv->sdev_contents);
1384 				rw_enter(&ddv->sdev_contents, RW_WRITER);
1385 			}
1386 			error = sdev_mknode(ddv, nm, &dv, &vattr, vp, link,
1387 			    cred, SDEV_READY);
1388 			rw_downgrade(&ddv->sdev_contents);
1389 
1390 			if (link != NULL) {
1391 				kmem_free(link, strlen(link) + 1);
1392 				link = NULL;
1393 			}
1394 
1395 			if (!error) {
1396 				ASSERT(dv);
1397 				ASSERT(dv->sdev_state != SDEV_ZOMBIE);
1398 				SDEV_SIMPLE_RELE(dv);
1399 			}
1400 			vp = NULL;
1401 			dv = NULL;
1402 		}
1403 	}
1404 
1405 done:
1406 	VN_RELE(dirvp);
1407 	kmem_free(dbuf, dlen);
1408 
1409 	return (error);
1410 }
1411 
1412 void
1413 sdev_filldir_dynamic(struct sdev_node *ddv)
1414 {
1415 	int error;
1416 	int i;
1417 	struct vattr vattr;
1418 	struct vattr *vap = &vattr;
1419 	char *nm = NULL;
1420 	struct sdev_node *dv = NULL;
1421 
1422 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1423 	ASSERT((ddv->sdev_flags & SDEV_BUILD));
1424 
1425 	*vap = *sdev_getdefault_attr(VDIR);	/* note structure copy here */
1426 	gethrestime(&vap->va_atime);
1427 	vap->va_mtime = vap->va_atime;
1428 	vap->va_ctime = vap->va_atime;
1429 	for (i = 0; vtab[i].vt_name != NULL; i++) {
1430 		/*
1431 		 * This early, we may be in a read-only /dev environment: leave
1432 		 * the creation of any nodes we'd attempt to persist to
1433 		 * devfsadm. Because /dev itself is normally persistent, any
1434 		 * node which is not marked dynamic will end up being marked
1435 		 * persistent. However, some nodes are both dynamic and
1436 		 * persistent, mostly lofi and rlofi, so we need to be careful
1437 		 * in our check.
1438 		 */
1439 		if ((vtab[i].vt_flags & SDEV_PERSIST) ||
1440 		    !(vtab[i].vt_flags & SDEV_DYNAMIC))
1441 			continue;
1442 		nm = vtab[i].vt_name;
1443 		ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1444 		dv = NULL;
1445 		error = sdev_mknode(ddv, nm, &dv, vap, NULL,
1446 		    NULL, kcred, SDEV_READY);
1447 		if (error) {
1448 			cmn_err(CE_WARN, "%s/%s: error %d\n",
1449 			    ddv->sdev_name, nm, error);
1450 		} else {
1451 			ASSERT(dv);
1452 			ASSERT(dv->sdev_state != SDEV_ZOMBIE);
1453 			SDEV_SIMPLE_RELE(dv);
1454 		}
1455 	}
1456 }
1457 
1458 /*
1459  * Creating a backing store entry based on sdev_attr.
1460  * This is called either as part of node creation in a persistent directory
1461  * or from setattr/setsecattr to persist access attributes across reboot.
1462  */
1463 int
1464 sdev_shadow_node(struct sdev_node *dv, struct cred *cred)
1465 {
1466 	int error = 0;
1467 	struct vnode *dvp = SDEVTOV(dv->sdev_dotdot);
1468 	struct vnode *rdvp = VTOSDEV(dvp)->sdev_attrvp;
1469 	struct vattr *vap = dv->sdev_attr;
1470 	char *nm = dv->sdev_name;
1471 	struct vnode *tmpvp, **rvp = &tmpvp, *rrvp = NULL;
1472 
1473 	ASSERT(dv && dv->sdev_name && rdvp);
1474 	ASSERT(RW_WRITE_HELD(&dv->sdev_contents) && dv->sdev_attrvp == NULL);
1475 
1476 lookup:
1477 	/* try to find it in the backing store */
1478 	error = VOP_LOOKUP(rdvp, nm, rvp, NULL, 0, NULL, cred, NULL, NULL,
1479 	    NULL);
1480 	if (error == 0) {
1481 		if (VOP_REALVP(*rvp, &rrvp, NULL) == 0) {
1482 			VN_HOLD(rrvp);
1483 			VN_RELE(*rvp);
1484 			*rvp = rrvp;
1485 		}
1486 
1487 		kmem_free(dv->sdev_attr, sizeof (vattr_t));
1488 		dv->sdev_attr = NULL;
1489 		dv->sdev_attrvp = *rvp;
1490 		return (0);
1491 	}
1492 
1493 	/* let's try to persist the node */
1494 	gethrestime(&vap->va_atime);
1495 	vap->va_mtime = vap->va_atime;
1496 	vap->va_ctime = vap->va_atime;
1497 	vap->va_mask |= AT_TYPE|AT_MODE;
1498 	switch (vap->va_type) {
1499 	case VDIR:
1500 		error = VOP_MKDIR(rdvp, nm, vap, rvp, cred, NULL, 0, NULL);
1501 		sdcmn_err9(("sdev_shadow_node: mkdir vp %p error %d\n",
1502 		    (void *)(*rvp), error));
1503 		if (!error)
1504 			VN_RELE(*rvp);
1505 		break;
1506 	case VCHR:
1507 	case VBLK:
1508 	case VREG:
1509 	case VDOOR:
1510 		error = VOP_CREATE(rdvp, nm, vap, NONEXCL, VREAD|VWRITE,
1511 		    rvp, cred, 0, NULL, NULL);
1512 		sdcmn_err9(("sdev_shadow_node: create vp %p, error %d\n",
1513 		    (void *)(*rvp), error));
1514 		if (!error)
1515 			VN_RELE(*rvp);
1516 		break;
1517 	case VLNK:
1518 		ASSERT(dv->sdev_symlink);
1519 		error = VOP_SYMLINK(rdvp, nm, vap, dv->sdev_symlink, cred,
1520 		    NULL, 0);
1521 		sdcmn_err9(("sdev_shadow_node: create symlink error %d\n",
1522 		    error));
1523 		break;
1524 	default:
1525 		cmn_err(CE_PANIC, "dev: %s: sdev_shadow_node "
1526 		    "create\n", nm);
1527 		/*NOTREACHED*/
1528 	}
1529 
1530 	/* go back to lookup to factor out spec node and set attrvp */
1531 	if (error == 0)
1532 		goto lookup;
1533 
1534 	sdcmn_err(("cannot persist %s - error %d\n", dv->sdev_path, error));
1535 	return (error);
1536 }
1537 
1538 static void
1539 sdev_cache_add(struct sdev_node *ddv, struct sdev_node **dv, char *nm)
1540 {
1541 	struct sdev_node *dup = NULL;
1542 
1543 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1544 	if ((dup = sdev_findbyname(ddv, nm)) == NULL) {
1545 		sdev_direnter(ddv, *dv);
1546 	} else {
1547 		VERIFY(dup->sdev_state != SDEV_ZOMBIE);
1548 		SDEV_SIMPLE_RELE(*dv);
1549 		sdev_nodedestroy(*dv, 0);
1550 		*dv = dup;
1551 	}
1552 }
1553 
1554 static void
1555 sdev_cache_delete(struct sdev_node *ddv, struct sdev_node **dv)
1556 {
1557 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1558 	sdev_dirdelete(ddv, *dv);
1559 }
1560 
1561 /*
1562  * update the in-core directory cache
1563  */
1564 void
1565 sdev_cache_update(struct sdev_node *ddv, struct sdev_node **dv, char *nm,
1566     sdev_cache_ops_t ops)
1567 {
1568 	ASSERT((SDEV_HELD(*dv)));
1569 
1570 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1571 	switch (ops) {
1572 	case SDEV_CACHE_ADD:
1573 		sdev_cache_add(ddv, dv, nm);
1574 		break;
1575 	case SDEV_CACHE_DELETE:
1576 		sdev_cache_delete(ddv, dv);
1577 		break;
1578 	default:
1579 		break;
1580 	}
1581 }
1582 
1583 /*
1584  * retrieve the named entry from the directory cache
1585  */
1586 struct sdev_node *
1587 sdev_cache_lookup(struct sdev_node *ddv, char *nm)
1588 {
1589 	struct sdev_node *dv = NULL;
1590 
1591 	ASSERT(RW_LOCK_HELD(&ddv->sdev_contents));
1592 	dv = sdev_findbyname(ddv, nm);
1593 
1594 	return (dv);
1595 }
1596 
1597 /*
1598  * Implicit reconfig for nodes constructed by a link generator
1599  * Start devfsadm if needed, or if devfsadm is in progress,
1600  * prepare to block on devfsadm either completing or
1601  * constructing the desired node.  As devfsadmd is global
1602  * in scope, constructing all necessary nodes, we only
1603  * need to initiate it once.
1604  */
1605 static int
1606 sdev_call_devfsadmd(struct sdev_node *ddv, struct sdev_node *dv, char *nm)
1607 {
1608 	int error = 0;
1609 
1610 	if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state)) {
1611 		sdcmn_err6(("lookup: waiting for %s/%s, 0x%x\n",
1612 		    ddv->sdev_name, nm, devfsadm_state));
1613 		mutex_enter(&dv->sdev_lookup_lock);
1614 		SDEV_BLOCK_OTHERS(dv, (SDEV_LOOKUP | SDEV_LGWAITING));
1615 		mutex_exit(&dv->sdev_lookup_lock);
1616 		error = 0;
1617 	} else if (!DEVNAME_DEVFSADM_HAS_RUN(devfsadm_state)) {
1618 		sdcmn_err6(("lookup %s/%s starting devfsadm, 0x%x\n",
1619 		    ddv->sdev_name, nm, devfsadm_state));
1620 
1621 		sdev_devfsadmd_thread(ddv, dv, kcred);
1622 		mutex_enter(&dv->sdev_lookup_lock);
1623 		SDEV_BLOCK_OTHERS(dv,
1624 		    (SDEV_LOOKUP | SDEV_LGWAITING));
1625 		mutex_exit(&dv->sdev_lookup_lock);
1626 		error = 0;
1627 	} else {
1628 		error = -1;
1629 	}
1630 
1631 	return (error);
1632 }
1633 
1634 /*
1635  *  Support for specialized device naming construction mechanisms
1636  */
1637 static int
1638 sdev_call_dircallback(struct sdev_node *ddv, struct sdev_node **dvp, char *nm,
1639     int (*callback)(struct sdev_node *, char *, void **, struct cred *,
1640     void *, char *), int flags, struct cred *cred)
1641 {
1642 	int rv = 0;
1643 	char *physpath = NULL;
1644 	struct vattr vattr;
1645 	struct vattr *vap = &vattr;
1646 	struct sdev_node *dv = NULL;
1647 
1648 	ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
1649 	if (flags & SDEV_VLINK) {
1650 		physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1651 		rv = callback(ddv, nm, (void *)&physpath, kcred, NULL,
1652 		    NULL);
1653 		if (rv) {
1654 			kmem_free(physpath, MAXPATHLEN);
1655 			return (-1);
1656 		}
1657 
1658 		*vap = *sdev_getdefault_attr(VLNK);	/* structure copy */
1659 		vap->va_size = strlen(physpath);
1660 		gethrestime(&vap->va_atime);
1661 		vap->va_mtime = vap->va_atime;
1662 		vap->va_ctime = vap->va_atime;
1663 
1664 		rv = sdev_mknode(ddv, nm, &dv, vap, NULL,
1665 		    (void *)physpath, cred, SDEV_READY);
1666 		kmem_free(physpath, MAXPATHLEN);
1667 		if (rv)
1668 			return (rv);
1669 	} else if (flags & SDEV_VATTR) {
1670 		/*
1671 		 * /dev/pts
1672 		 *
1673 		 * callback is responsible to set the basic attributes,
1674 		 * e.g. va_type/va_uid/va_gid/
1675 		 *    dev_t if VCHR or VBLK/
1676 		 */
1677 		ASSERT(callback);
1678 		rv = callback(ddv, nm, (void *)&vattr, kcred, NULL, NULL);
1679 		if (rv) {
1680 			sdcmn_err3(("devname_lookup_func: SDEV_NONE "
1681 			    "callback failed \n"));
1682 			return (-1);
1683 		}
1684 
1685 		rv = sdev_mknode(ddv, nm, &dv, &vattr, NULL, NULL,
1686 		    cred, SDEV_READY);
1687 
1688 		if (rv)
1689 			return (rv);
1690 
1691 	} else {
1692 		impossible(("lookup: %s/%s by %s not supported (%d)\n",
1693 		    SDEVTOV(ddv)->v_path, nm, curproc->p_user.u_comm,
1694 		    __LINE__));
1695 		rv = -1;
1696 	}
1697 
1698 	*dvp = dv;
1699 	return (rv);
1700 }
1701 
1702 static int
1703 is_devfsadm_thread(char *exec_name)
1704 {
1705 	/*
1706 	 * note: because devfsadmd -> /usr/sbin/devfsadm
1707 	 * it is safe to use "devfsadm" to capture the lookups
1708 	 * from devfsadm and its daemon version.
1709 	 */
1710 	if (strcmp(exec_name, "devfsadm") == 0)
1711 		return (1);
1712 	return (0);
1713 }
1714 
1715 /*
1716  * Lookup Order:
1717  *	sdev_node cache;
1718  *	backing store (SDEV_PERSIST);
1719  *	DBNR: a. dir_ops implemented in the loadable modules;
1720  *	      b. vnode ops in vtab.
1721  */
1722 int
1723 devname_lookup_func(struct sdev_node *ddv, char *nm, struct vnode **vpp,
1724     struct cred *cred, int (*callback)(struct sdev_node *, char *, void **,
1725     struct cred *, void *, char *), int flags)
1726 {
1727 	int rv = 0, nmlen;
1728 	struct vnode *rvp = NULL;
1729 	struct sdev_node *dv = NULL;
1730 	int	retried = 0;
1731 	int	error = 0;
1732 	struct vattr vattr;
1733 	char *lookup_thread = curproc->p_user.u_comm;
1734 	int failed_flags = 0;
1735 	int (*vtor)(struct sdev_node *) = NULL;
1736 	int state;
1737 	int parent_state;
1738 	char *link = NULL;
1739 
1740 	if (SDEVTOV(ddv)->v_type != VDIR)
1741 		return (ENOTDIR);
1742 
1743 	/*
1744 	 * Empty name or ., return node itself.
1745 	 */
1746 	nmlen = strlen(nm);
1747 	if ((nmlen == 0) || ((nmlen == 1) && (nm[0] == '.'))) {
1748 		*vpp = SDEVTOV(ddv);
1749 		VN_HOLD(*vpp);
1750 		return (0);
1751 	}
1752 
1753 	/*
1754 	 * .., return the parent directory
1755 	 */
1756 	if ((nmlen == 2) && (strcmp(nm, "..") == 0)) {
1757 		*vpp = SDEVTOV(ddv->sdev_dotdot);
1758 		VN_HOLD(*vpp);
1759 		return (0);
1760 	}
1761 
1762 	rw_enter(&ddv->sdev_contents, RW_READER);
1763 	if (ddv->sdev_flags & SDEV_VTOR) {
1764 		vtor = (int (*)(struct sdev_node *))sdev_get_vtor(ddv);
1765 		ASSERT(vtor);
1766 	}
1767 
1768 tryagain:
1769 	/*
1770 	 * (a) directory cache lookup:
1771 	 */
1772 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
1773 	parent_state = ddv->sdev_state;
1774 	dv = sdev_cache_lookup(ddv, nm);
1775 	if (dv) {
1776 		state = dv->sdev_state;
1777 		switch (state) {
1778 		case SDEV_INIT:
1779 			if (is_devfsadm_thread(lookup_thread))
1780 				break;
1781 
1782 			/* ZOMBIED parent won't allow node creation */
1783 			if (parent_state == SDEV_ZOMBIE) {
1784 				SD_TRACE_FAILED_LOOKUP(ddv, nm,
1785 				    retried);
1786 				goto nolock_notfound;
1787 			}
1788 
1789 			mutex_enter(&dv->sdev_lookup_lock);
1790 			/* compensate the threads started after devfsadm */
1791 			if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state) &&
1792 			    !(SDEV_IS_LOOKUP(dv)))
1793 				SDEV_BLOCK_OTHERS(dv,
1794 				    (SDEV_LOOKUP | SDEV_LGWAITING));
1795 
1796 			if (SDEV_IS_LOOKUP(dv)) {
1797 				failed_flags |= SLF_REBUILT;
1798 				rw_exit(&ddv->sdev_contents);
1799 				error = sdev_wait4lookup(dv, SDEV_LOOKUP);
1800 				mutex_exit(&dv->sdev_lookup_lock);
1801 				rw_enter(&ddv->sdev_contents, RW_READER);
1802 
1803 				if (error != 0) {
1804 					SD_TRACE_FAILED_LOOKUP(ddv, nm,
1805 					    retried);
1806 					goto nolock_notfound;
1807 				}
1808 
1809 				state = dv->sdev_state;
1810 				if (state == SDEV_INIT) {
1811 					SD_TRACE_FAILED_LOOKUP(ddv, nm,
1812 					    retried);
1813 					goto nolock_notfound;
1814 				} else if (state == SDEV_READY) {
1815 					goto found;
1816 				} else if (state == SDEV_ZOMBIE) {
1817 					rw_exit(&ddv->sdev_contents);
1818 					SD_TRACE_FAILED_LOOKUP(ddv, nm,
1819 					    retried);
1820 					SDEV_RELE(dv);
1821 					goto lookup_failed;
1822 				}
1823 			} else {
1824 				mutex_exit(&dv->sdev_lookup_lock);
1825 			}
1826 			break;
1827 		case SDEV_READY:
1828 			goto found;
1829 		case SDEV_ZOMBIE:
1830 			rw_exit(&ddv->sdev_contents);
1831 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1832 			SDEV_RELE(dv);
1833 			goto lookup_failed;
1834 		default:
1835 			rw_exit(&ddv->sdev_contents);
1836 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1837 			sdev_lookup_failed(ddv, nm, failed_flags);
1838 			*vpp = NULLVP;
1839 			return (ENOENT);
1840 		}
1841 	}
1842 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
1843 
1844 	/*
1845 	 * ZOMBIED parent does not allow new node creation.
1846 	 * bail out early
1847 	 */
1848 	if (parent_state == SDEV_ZOMBIE) {
1849 		rw_exit(&ddv->sdev_contents);
1850 		*vpp = NULLVP;
1851 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1852 		return (ENOENT);
1853 	}
1854 
1855 	/*
1856 	 * (b0): backing store lookup
1857 	 *	SDEV_PERSIST is default except:
1858 	 *		1) pts nodes
1859 	 *		2) non-chmod'ed local nodes
1860 	 *		3) zvol nodes
1861 	 */
1862 	if (SDEV_IS_PERSIST(ddv)) {
1863 		error = devname_backstore_lookup(ddv, nm, &rvp);
1864 
1865 		if (!error) {
1866 
1867 			vattr.va_mask = AT_TYPE|AT_MODE|AT_UID|AT_GID;
1868 			error = VOP_GETATTR(rvp, &vattr, 0, cred, NULL);
1869 			if (error) {
1870 				rw_exit(&ddv->sdev_contents);
1871 				if (dv)
1872 					SDEV_RELE(dv);
1873 				SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1874 				sdev_lookup_failed(ddv, nm, failed_flags);
1875 				*vpp = NULLVP;
1876 				return (ENOENT);
1877 			}
1878 
1879 			if (vattr.va_type == VLNK) {
1880 				error = sdev_getlink(rvp, &link);
1881 				if (error) {
1882 					rw_exit(&ddv->sdev_contents);
1883 					if (dv)
1884 						SDEV_RELE(dv);
1885 					SD_TRACE_FAILED_LOOKUP(ddv, nm,
1886 					    retried);
1887 					sdev_lookup_failed(ddv, nm,
1888 					    failed_flags);
1889 					*vpp = NULLVP;
1890 					return (ENOENT);
1891 				}
1892 				ASSERT(link != NULL);
1893 			}
1894 
1895 			if (!rw_tryupgrade(&ddv->sdev_contents)) {
1896 				rw_exit(&ddv->sdev_contents);
1897 				rw_enter(&ddv->sdev_contents, RW_WRITER);
1898 			}
1899 			error = sdev_mknode(ddv, nm, &dv, &vattr,
1900 			    rvp, link, cred, SDEV_READY);
1901 			rw_downgrade(&ddv->sdev_contents);
1902 
1903 			if (link != NULL) {
1904 				kmem_free(link, strlen(link) + 1);
1905 				link = NULL;
1906 			}
1907 
1908 			if (error) {
1909 				SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1910 				rw_exit(&ddv->sdev_contents);
1911 				if (dv)
1912 					SDEV_RELE(dv);
1913 				goto lookup_failed;
1914 			} else {
1915 				goto found;
1916 			}
1917 		} else if (retried) {
1918 			rw_exit(&ddv->sdev_contents);
1919 			sdcmn_err3(("retry of lookup of %s/%s: failed\n",
1920 			    ddv->sdev_name, nm));
1921 			if (dv)
1922 				SDEV_RELE(dv);
1923 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1924 			sdev_lookup_failed(ddv, nm, failed_flags);
1925 			*vpp = NULLVP;
1926 			return (ENOENT);
1927 		}
1928 	}
1929 
1930 lookup_create_node:
1931 	/* first thread that is doing the lookup on this node */
1932 	if (callback) {
1933 		ASSERT(dv == NULL);
1934 		if (!rw_tryupgrade(&ddv->sdev_contents)) {
1935 			rw_exit(&ddv->sdev_contents);
1936 			rw_enter(&ddv->sdev_contents, RW_WRITER);
1937 		}
1938 		error = sdev_call_dircallback(ddv, &dv, nm, callback,
1939 		    flags, cred);
1940 		rw_downgrade(&ddv->sdev_contents);
1941 		if (error == 0) {
1942 			goto found;
1943 		} else {
1944 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1945 			rw_exit(&ddv->sdev_contents);
1946 			goto lookup_failed;
1947 		}
1948 	}
1949 	if (!dv) {
1950 		if (!rw_tryupgrade(&ddv->sdev_contents)) {
1951 			rw_exit(&ddv->sdev_contents);
1952 			rw_enter(&ddv->sdev_contents, RW_WRITER);
1953 		}
1954 		error = sdev_mknode(ddv, nm, &dv, NULL, NULL, NULL,
1955 		    cred, SDEV_INIT);
1956 		if (!dv) {
1957 			rw_exit(&ddv->sdev_contents);
1958 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1959 			sdev_lookup_failed(ddv, nm, failed_flags);
1960 			*vpp = NULLVP;
1961 			return (ENOENT);
1962 		}
1963 		rw_downgrade(&ddv->sdev_contents);
1964 	}
1965 
1966 	/*
1967 	 * (b1) invoking devfsadm once per life time for devfsadm nodes
1968 	 */
1969 	ASSERT(SDEV_HELD(dv));
1970 
1971 	if (SDEV_IS_NO_NCACHE(dv))
1972 		failed_flags |= SLF_NO_NCACHE;
1973 	if (sdev_reconfig_boot || !i_ddi_io_initialized() ||
1974 	    SDEV_IS_DYNAMIC(ddv) || SDEV_IS_NO_NCACHE(dv) ||
1975 	    ((moddebug & MODDEBUG_FINI_EBUSY) != 0)) {
1976 		ASSERT(SDEV_HELD(dv));
1977 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1978 		goto nolock_notfound;
1979 	}
1980 
1981 	/*
1982 	 * filter out known non-existent devices recorded
1983 	 * during initial reconfiguration boot for which
1984 	 * reconfig should not be done and lookup may
1985 	 * be short-circuited now.
1986 	 */
1987 	if (sdev_lookup_filter(ddv, nm)) {
1988 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1989 		goto nolock_notfound;
1990 	}
1991 
1992 	/* bypassing devfsadm internal nodes */
1993 	if (is_devfsadm_thread(lookup_thread)) {
1994 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
1995 		goto nolock_notfound;
1996 	}
1997 
1998 	if (sdev_reconfig_disable) {
1999 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
2000 		goto nolock_notfound;
2001 	}
2002 
2003 	error = sdev_call_devfsadmd(ddv, dv, nm);
2004 	if (error == 0) {
2005 		sdcmn_err8(("lookup of %s/%s by %s: reconfig\n",
2006 		    ddv->sdev_name, nm, curproc->p_user.u_comm));
2007 		if (sdev_reconfig_verbose) {
2008 			cmn_err(CE_CONT,
2009 			    "?lookup of %s/%s by %s: reconfig\n",
2010 			    ddv->sdev_name, nm, curproc->p_user.u_comm);
2011 		}
2012 		retried = 1;
2013 		failed_flags |= SLF_REBUILT;
2014 		ASSERT(dv->sdev_state != SDEV_ZOMBIE);
2015 		SDEV_SIMPLE_RELE(dv);
2016 		goto tryagain;
2017 	} else {
2018 		SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
2019 		goto nolock_notfound;
2020 	}
2021 
2022 found:
2023 	ASSERT(dv->sdev_state == SDEV_READY);
2024 	if (vtor) {
2025 		/*
2026 		 * Check validity of returned node
2027 		 */
2028 		switch (vtor(dv)) {
2029 		case SDEV_VTOR_VALID:
2030 			break;
2031 		case SDEV_VTOR_STALE:
2032 			/*
2033 			 * The name exists, but the cache entry is
2034 			 * stale and needs to be re-created.
2035 			 */
2036 			ASSERT(RW_READ_HELD(&ddv->sdev_contents));
2037 			if (rw_tryupgrade(&ddv->sdev_contents) == 0) {
2038 				rw_exit(&ddv->sdev_contents);
2039 				rw_enter(&ddv->sdev_contents, RW_WRITER);
2040 			}
2041 			sdev_cache_update(ddv, &dv, nm, SDEV_CACHE_DELETE);
2042 			rw_downgrade(&ddv->sdev_contents);
2043 			SDEV_RELE(dv);
2044 			dv = NULL;
2045 			goto lookup_create_node;
2046 			/* FALLTHRU */
2047 		case SDEV_VTOR_INVALID:
2048 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
2049 			sdcmn_err7(("lookup: destroy invalid "
2050 			    "node: %s(%p)\n", dv->sdev_name, (void *)dv));
2051 			goto nolock_notfound;
2052 		case SDEV_VTOR_SKIP:
2053 			sdcmn_err7(("lookup: node not applicable - "
2054 			    "skipping: %s(%p)\n", dv->sdev_name, (void *)dv));
2055 			rw_exit(&ddv->sdev_contents);
2056 			SD_TRACE_FAILED_LOOKUP(ddv, nm, retried);
2057 			SDEV_RELE(dv);
2058 			goto lookup_failed;
2059 		default:
2060 			cmn_err(CE_PANIC,
2061 			    "dev fs: validator failed: %s(%p)\n",
2062 			    dv->sdev_name, (void *)dv);
2063 			break;
2064 		}
2065 	}
2066 
2067 	rw_exit(&ddv->sdev_contents);
2068 	rv = sdev_to_vp(dv, vpp);
2069 	sdcmn_err3(("devname_lookup_func: returning vp %p v_count %d state %d "
2070 	    "for nm %s, error %d\n", (void *)*vpp, (*vpp)->v_count,
2071 	    dv->sdev_state, nm, rv));
2072 	return (rv);
2073 
2074 nolock_notfound:
2075 	/*
2076 	 * Destroy the node that is created for synchronization purposes.
2077 	 */
2078 	sdcmn_err3(("devname_lookup_func: %s with state %d\n",
2079 	    nm, dv->sdev_state));
2080 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
2081 	if (dv->sdev_state == SDEV_INIT) {
2082 		if (!rw_tryupgrade(&ddv->sdev_contents)) {
2083 			rw_exit(&ddv->sdev_contents);
2084 			rw_enter(&ddv->sdev_contents, RW_WRITER);
2085 		}
2086 
2087 		/*
2088 		 * Node state may have changed during the lock
2089 		 * changes. Re-check.
2090 		 */
2091 		if (dv->sdev_state == SDEV_INIT) {
2092 			sdev_dirdelete(ddv, dv);
2093 			rw_exit(&ddv->sdev_contents);
2094 			sdev_lookup_failed(ddv, nm, failed_flags);
2095 			SDEV_RELE(dv);
2096 			*vpp = NULL;
2097 			return (ENOENT);
2098 		}
2099 	}
2100 
2101 	rw_exit(&ddv->sdev_contents);
2102 	SDEV_RELE(dv);
2103 
2104 lookup_failed:
2105 	sdev_lookup_failed(ddv, nm, failed_flags);
2106 	*vpp = NULL;
2107 	return (ENOENT);
2108 }
2109 
2110 /*
2111  * Given a directory node, mark all nodes beneath as
2112  * STALE, i.e. nodes that don't exist as far as new
2113  * consumers are concerned.  Remove them from the
2114  * list of directory entries so that no lookup or
2115  * directory traversal will find them.  The node
2116  * not deallocated so existing holds are not affected.
2117  */
2118 void
2119 sdev_stale(struct sdev_node *ddv)
2120 {
2121 	struct sdev_node *dv;
2122 	struct vnode *vp;
2123 
2124 	ASSERT(SDEVTOV(ddv)->v_type == VDIR);
2125 
2126 	rw_enter(&ddv->sdev_contents, RW_WRITER);
2127 	while ((dv = SDEV_FIRST_ENTRY(ddv)) != NULL) {
2128 		vp = SDEVTOV(dv);
2129 		SDEV_HOLD(dv);
2130 		if (vp->v_type == VDIR)
2131 			sdev_stale(dv);
2132 
2133 		sdev_dirdelete(ddv, dv);
2134 		SDEV_RELE(dv);
2135 	}
2136 	ddv->sdev_flags |= SDEV_BUILD;
2137 	rw_exit(&ddv->sdev_contents);
2138 }
2139 
2140 /*
2141  * Given a directory node, clean out all the nodes beneath.
2142  * If expr is specified, clean node with names matching expr.
2143  * If SDEV_ENFORCE is specified in flags, busy nodes are made stale,
2144  *	so they are excluded from future lookups.
2145  */
2146 int
2147 sdev_cleandir(struct sdev_node *ddv, char *expr, uint_t flags)
2148 {
2149 	int error = 0;
2150 	int busy = 0;
2151 	struct vnode *vp;
2152 	struct sdev_node *dv, *next;
2153 	int bkstore = 0;
2154 	int len = 0;
2155 	char *bks_name = NULL;
2156 
2157 	ASSERT(SDEVTOV(ddv)->v_type == VDIR);
2158 
2159 	/*
2160 	 * We try our best to destroy all unused sdev_node's
2161 	 */
2162 	rw_enter(&ddv->sdev_contents, RW_WRITER);
2163 	for (dv = SDEV_FIRST_ENTRY(ddv); dv != NULL; dv = next) {
2164 		next = SDEV_NEXT_ENTRY(ddv, dv);
2165 		vp = SDEVTOV(dv);
2166 
2167 		if (expr && gmatch(dv->sdev_name, expr) == 0)
2168 			continue;
2169 
2170 		if (vp->v_type == VDIR &&
2171 		    sdev_cleandir(dv, NULL, flags) != 0) {
2172 			sdcmn_err9(("sdev_cleandir: dir %s busy\n",
2173 			    dv->sdev_name));
2174 			busy++;
2175 			continue;
2176 		}
2177 
2178 		if (vp->v_count > 0 && (flags & SDEV_ENFORCE) == 0) {
2179 			sdcmn_err9(("sdev_cleandir: dir %s busy\n",
2180 			    dv->sdev_name));
2181 			busy++;
2182 			continue;
2183 		}
2184 
2185 		/*
2186 		 * at this point, either dv is not held or SDEV_ENFORCE
2187 		 * is specified. In either case, dv needs to be deleted
2188 		 */
2189 		SDEV_HOLD(dv);
2190 
2191 		bkstore = SDEV_IS_PERSIST(dv) ? 1 : 0;
2192 		if (bkstore && (vp->v_type == VDIR))
2193 			bkstore += 1;
2194 
2195 		if (bkstore) {
2196 			len = strlen(dv->sdev_name) + 1;
2197 			bks_name = kmem_alloc(len, KM_SLEEP);
2198 			bcopy(dv->sdev_name, bks_name, len);
2199 		}
2200 
2201 		sdev_dirdelete(ddv, dv);
2202 
2203 		/* take care the backing store clean up */
2204 		if (bkstore) {
2205 			ASSERT(bks_name);
2206 			ASSERT(ddv->sdev_attrvp);
2207 
2208 			if (bkstore == 1) {
2209 				error = VOP_REMOVE(ddv->sdev_attrvp,
2210 				    bks_name, kcred, NULL, 0);
2211 			} else if (bkstore == 2) {
2212 				error = VOP_RMDIR(ddv->sdev_attrvp,
2213 				    bks_name, ddv->sdev_attrvp, kcred, NULL, 0);
2214 			}
2215 
2216 			/* do not propagate the backing store errors */
2217 			if (error) {
2218 				sdcmn_err9(("sdev_cleandir: backing store"
2219 				    "not cleaned\n"));
2220 				error = 0;
2221 			}
2222 
2223 			bkstore = 0;
2224 			kmem_free(bks_name, len);
2225 			bks_name = NULL;
2226 			len = 0;
2227 		}
2228 
2229 		ddv->sdev_flags |= SDEV_BUILD;
2230 		SDEV_RELE(dv);
2231 	}
2232 
2233 	ddv->sdev_flags |= SDEV_BUILD;
2234 	rw_exit(&ddv->sdev_contents);
2235 
2236 	if (busy) {
2237 		error = EBUSY;
2238 	}
2239 
2240 	return (error);
2241 }
2242 
2243 /*
2244  * a convenient wrapper for readdir() funcs
2245  */
2246 size_t
2247 add_dir_entry(dirent64_t *de, char *nm, size_t size, ino_t ino, offset_t off)
2248 {
2249 	size_t reclen = DIRENT64_RECLEN(strlen(nm));
2250 	if (reclen > size)
2251 		return (0);
2252 
2253 	de->d_ino = (ino64_t)ino;
2254 	de->d_off = (off64_t)off + 1;
2255 	de->d_reclen = (ushort_t)reclen;
2256 	(void) strncpy(de->d_name, nm, DIRENT64_NAMELEN(reclen));
2257 	return (reclen);
2258 }
2259 
2260 /*
2261  * sdev_mount service routines
2262  */
2263 int
2264 sdev_copyin_mountargs(struct mounta *uap, struct sdev_mountargs *args)
2265 {
2266 	int	error;
2267 
2268 	if (uap->datalen != sizeof (*args))
2269 		return (EINVAL);
2270 
2271 	if (error = copyin(uap->dataptr, args, sizeof (*args))) {
2272 		cmn_err(CE_WARN, "sdev_copyin_mountargs: can not"
2273 		    "get user data. error %d\n", error);
2274 		return (EFAULT);
2275 	}
2276 
2277 	return (0);
2278 }
2279 
2280 #ifdef nextdp
2281 #undef nextdp
2282 #endif
2283 #define	nextdp(dp)	((struct dirent64 *) \
2284 			    (intptr_t)((char *)(dp) + (dp)->d_reclen))
2285 
2286 /*
2287  * readdir helper func
2288  */
2289 int
2290 devname_readdir_func(vnode_t *vp, uio_t *uiop, cred_t *cred, int *eofp,
2291     int flags)
2292 {
2293 	struct sdev_node *ddv = VTOSDEV(vp);
2294 	struct sdev_node *dv;
2295 	dirent64_t	*dp;
2296 	ulong_t		outcount = 0;
2297 	size_t		namelen;
2298 	ulong_t		alloc_count;
2299 	void		*outbuf;
2300 	struct iovec	*iovp;
2301 	int		error = 0;
2302 	size_t		reclen;
2303 	offset_t	diroff;
2304 	offset_t	soff;
2305 	int		this_reclen;
2306 	int (*vtor)(struct sdev_node *) = NULL;
2307 	struct vattr attr;
2308 	timestruc_t now;
2309 
2310 	ASSERT(ddv->sdev_attr || ddv->sdev_attrvp);
2311 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
2312 
2313 	if (uiop->uio_loffset >= MAXOFF_T) {
2314 		if (eofp)
2315 			*eofp = 1;
2316 		return (0);
2317 	}
2318 
2319 	if (uiop->uio_iovcnt != 1)
2320 		return (EINVAL);
2321 
2322 	if (vp->v_type != VDIR)
2323 		return (ENOTDIR);
2324 
2325 	if (ddv->sdev_flags & SDEV_VTOR) {
2326 		vtor = (int (*)(struct sdev_node *))sdev_get_vtor(ddv);
2327 		ASSERT(vtor);
2328 	}
2329 
2330 	if (eofp != NULL)
2331 		*eofp = 0;
2332 
2333 	soff = uiop->uio_loffset;
2334 	iovp = uiop->uio_iov;
2335 	alloc_count = iovp->iov_len;
2336 	dp = outbuf = kmem_alloc(alloc_count, KM_SLEEP);
2337 	outcount = 0;
2338 
2339 	if (ddv->sdev_state == SDEV_ZOMBIE)
2340 		goto get_cache;
2341 
2342 	if (SDEV_IS_GLOBAL(ddv)) {
2343 
2344 		if ((sdev_boot_state == SDEV_BOOT_STATE_COMPLETE) &&
2345 		    !sdev_reconfig_boot && (flags & SDEV_BROWSE) &&
2346 		    !SDEV_IS_DYNAMIC(ddv) && !SDEV_IS_NO_NCACHE(ddv) &&
2347 		    ((moddebug & MODDEBUG_FINI_EBUSY) == 0) &&
2348 		    !DEVNAME_DEVFSADM_HAS_RUN(devfsadm_state) &&
2349 		    !DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state) &&
2350 		    !sdev_reconfig_disable) {
2351 			/*
2352 			 * invoking "devfsadm" to do system device reconfig
2353 			 */
2354 			mutex_enter(&ddv->sdev_lookup_lock);
2355 			SDEV_BLOCK_OTHERS(ddv,
2356 			    (SDEV_READDIR|SDEV_LGWAITING));
2357 			mutex_exit(&ddv->sdev_lookup_lock);
2358 
2359 			sdcmn_err8(("readdir of %s by %s: reconfig\n",
2360 			    ddv->sdev_path, curproc->p_user.u_comm));
2361 			if (sdev_reconfig_verbose) {
2362 				cmn_err(CE_CONT,
2363 				    "?readdir of %s by %s: reconfig\n",
2364 				    ddv->sdev_path, curproc->p_user.u_comm);
2365 			}
2366 
2367 			sdev_devfsadmd_thread(ddv, NULL, kcred);
2368 		} else if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state)) {
2369 			/*
2370 			 * compensate the "ls" started later than "devfsadm"
2371 			 */
2372 			mutex_enter(&ddv->sdev_lookup_lock);
2373 			SDEV_BLOCK_OTHERS(ddv, (SDEV_READDIR|SDEV_LGWAITING));
2374 			mutex_exit(&ddv->sdev_lookup_lock);
2375 		}
2376 
2377 		/*
2378 		 * release the contents lock so that
2379 		 * the cache may be updated by devfsadmd
2380 		 */
2381 		rw_exit(&ddv->sdev_contents);
2382 		mutex_enter(&ddv->sdev_lookup_lock);
2383 		if (SDEV_IS_READDIR(ddv))
2384 			(void) sdev_wait4lookup(ddv, SDEV_READDIR);
2385 		mutex_exit(&ddv->sdev_lookup_lock);
2386 		rw_enter(&ddv->sdev_contents, RW_READER);
2387 
2388 		sdcmn_err4(("readdir of directory %s by %s\n",
2389 		    ddv->sdev_name, curproc->p_user.u_comm));
2390 		if (ddv->sdev_flags & SDEV_BUILD) {
2391 			if (SDEV_IS_PERSIST(ddv)) {
2392 				error = sdev_filldir_from_store(ddv,
2393 				    alloc_count, cred);
2394 			}
2395 			ddv->sdev_flags &= ~SDEV_BUILD;
2396 		}
2397 	}
2398 
2399 get_cache:
2400 	/* handle "." and ".." */
2401 	diroff = 0;
2402 	if (soff == 0) {
2403 		/* first time */
2404 		this_reclen = DIRENT64_RECLEN(1);
2405 		if (alloc_count < this_reclen) {
2406 			error = EINVAL;
2407 			goto done;
2408 		}
2409 
2410 		dp->d_ino = (ino64_t)ddv->sdev_ino;
2411 		dp->d_off = (off64_t)1;
2412 		dp->d_reclen = (ushort_t)this_reclen;
2413 
2414 		(void) strncpy(dp->d_name, ".",
2415 		    DIRENT64_NAMELEN(this_reclen));
2416 		outcount += dp->d_reclen;
2417 		dp = nextdp(dp);
2418 	}
2419 
2420 	diroff++;
2421 	if (soff <= 1) {
2422 		this_reclen = DIRENT64_RECLEN(2);
2423 		if (alloc_count < outcount + this_reclen) {
2424 			error = EINVAL;
2425 			goto done;
2426 		}
2427 
2428 		dp->d_reclen = (ushort_t)this_reclen;
2429 		dp->d_ino = (ino64_t)ddv->sdev_dotdot->sdev_ino;
2430 		dp->d_off = (off64_t)2;
2431 
2432 		(void) strncpy(dp->d_name, "..",
2433 		    DIRENT64_NAMELEN(this_reclen));
2434 		outcount += dp->d_reclen;
2435 
2436 		dp = nextdp(dp);
2437 	}
2438 
2439 
2440 	/* gets the cache */
2441 	diroff++;
2442 	for (dv = SDEV_FIRST_ENTRY(ddv); dv;
2443 	    dv = SDEV_NEXT_ENTRY(ddv, dv), diroff++) {
2444 		sdcmn_err3(("sdev_readdir: diroff %lld soff %lld for '%s' \n",
2445 		    diroff, soff, dv->sdev_name));
2446 
2447 		/* bypassing pre-matured nodes */
2448 		if (diroff < soff || (dv->sdev_state != SDEV_READY)) {
2449 			sdcmn_err3(("sdev_readdir: pre-mature node  "
2450 			    "%s %d\n", dv->sdev_name, dv->sdev_state));
2451 			continue;
2452 		}
2453 
2454 		/*
2455 		 * Check validity of node
2456 		 * Drop invalid and nodes to be skipped.
2457 		 * A node the validator indicates as stale needs
2458 		 * to be returned as presumably the node name itself
2459 		 * is valid and the node data itself will be refreshed
2460 		 * on lookup.  An application performing a readdir then
2461 		 * stat on each entry should thus always see consistent
2462 		 * data.  In any case, it is not possible to synchronize
2463 		 * with dynamic kernel state, and any view we return can
2464 		 * never be anything more than a snapshot at a point in time.
2465 		 */
2466 		if (vtor) {
2467 			switch (vtor(dv)) {
2468 			case SDEV_VTOR_VALID:
2469 				break;
2470 			case SDEV_VTOR_INVALID:
2471 			case SDEV_VTOR_SKIP:
2472 				continue;
2473 			case SDEV_VTOR_STALE:
2474 				sdcmn_err3(("sdev_readir: %s stale\n",
2475 				    dv->sdev_name));
2476 				break;
2477 			default:
2478 				cmn_err(CE_PANIC,
2479 				    "dev fs: validator failed: %s(%p)\n",
2480 				    dv->sdev_name, (void *)dv);
2481 				break;
2482 			/*NOTREACHED*/
2483 			}
2484 		}
2485 
2486 		namelen = strlen(dv->sdev_name);
2487 		reclen = DIRENT64_RECLEN(namelen);
2488 		if (outcount + reclen > alloc_count) {
2489 			goto full;
2490 		}
2491 		dp->d_reclen = (ushort_t)reclen;
2492 		dp->d_ino = (ino64_t)dv->sdev_ino;
2493 		dp->d_off = (off64_t)diroff + 1;
2494 		(void) strncpy(dp->d_name, dv->sdev_name,
2495 		    DIRENT64_NAMELEN(reclen));
2496 		outcount += reclen;
2497 		dp = nextdp(dp);
2498 	}
2499 
2500 full:
2501 	sdcmn_err4(("sdev_readdir: moving %lu bytes: "
2502 	    "diroff %lld, soff %lld, dv %p\n", outcount, diroff, soff,
2503 	    (void *)dv));
2504 
2505 	if (outcount)
2506 		error = uiomove(outbuf, outcount, UIO_READ, uiop);
2507 
2508 	if (!error) {
2509 		uiop->uio_loffset = diroff;
2510 		if (eofp)
2511 			*eofp = dv ? 0 : 1;
2512 	}
2513 
2514 
2515 	if (ddv->sdev_attrvp) {
2516 		gethrestime(&now);
2517 		attr.va_ctime = now;
2518 		attr.va_atime = now;
2519 		attr.va_mask = AT_CTIME|AT_ATIME;
2520 
2521 		(void) VOP_SETATTR(ddv->sdev_attrvp, &attr, 0, kcred, NULL);
2522 	}
2523 done:
2524 	kmem_free(outbuf, alloc_count);
2525 	return (error);
2526 }
2527 
2528 static int
2529 sdev_modctl_lookup(const char *path, vnode_t **r_vp)
2530 {
2531 	vnode_t *vp;
2532 	vnode_t *cvp;
2533 	struct sdev_node *svp;
2534 	char *nm;
2535 	struct pathname pn;
2536 	int error;
2537 	int persisted = 0;
2538 
2539 	ASSERT(INGLOBALZONE(curproc));
2540 
2541 	if (error = pn_get((char *)path, UIO_SYSSPACE, &pn))
2542 		return (error);
2543 	nm = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2544 
2545 	vp = rootdir;
2546 	VN_HOLD(vp);
2547 
2548 	while (pn_pathleft(&pn)) {
2549 		ASSERT(vp->v_type == VDIR || vp->v_type == VLNK);
2550 		(void) pn_getcomponent(&pn, nm);
2551 
2552 		/*
2553 		 * Deal with the .. special case where we may be
2554 		 * traversing up across a mount point, to the
2555 		 * root of this filesystem or global root.
2556 		 */
2557 		if (nm[0] == '.' && nm[1] == '.' && nm[2] == 0) {
2558 checkforroot:
2559 			if (VN_CMP(vp, rootdir)) {
2560 				nm[1] = 0;
2561 			} else if (vp->v_flag & VROOT) {
2562 				vfs_t *vfsp;
2563 				cvp = vp;
2564 				vfsp = cvp->v_vfsp;
2565 				vfs_rlock_wait(vfsp);
2566 				vp = cvp->v_vfsp->vfs_vnodecovered;
2567 				if (vp == NULL ||
2568 				    (cvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)) {
2569 					vfs_unlock(vfsp);
2570 					VN_RELE(cvp);
2571 					error = EIO;
2572 					break;
2573 				}
2574 				VN_HOLD(vp);
2575 				vfs_unlock(vfsp);
2576 				VN_RELE(cvp);
2577 				cvp = NULL;
2578 				goto checkforroot;
2579 			}
2580 		}
2581 
2582 		error = VOP_LOOKUP(vp, nm, &cvp, NULL, 0, NULL, kcred, NULL,
2583 		    NULL, NULL);
2584 		if (error) {
2585 			VN_RELE(vp);
2586 			break;
2587 		}
2588 
2589 		/* traverse mount points encountered on our journey */
2590 		if (vn_ismntpt(cvp) && (error = traverse(&cvp)) != 0) {
2591 			VN_RELE(vp);
2592 			VN_RELE(cvp);
2593 			break;
2594 		}
2595 
2596 		/*
2597 		 * symbolic link, can be either relative and absolute
2598 		 */
2599 		if ((cvp->v_type == VLNK) && pn_pathleft(&pn)) {
2600 			struct pathname linkpath;
2601 			pn_alloc(&linkpath);
2602 			if (error = pn_getsymlink(cvp, &linkpath, kcred)) {
2603 				pn_free(&linkpath);
2604 				break;
2605 			}
2606 			if (pn_pathleft(&linkpath) == 0)
2607 				(void) pn_set(&linkpath, ".");
2608 			error = pn_insert(&pn, &linkpath, strlen(nm));
2609 			pn_free(&linkpath);
2610 			if (pn.pn_pathlen == 0) {
2611 				VN_RELE(vp);
2612 				return (ENOENT);
2613 			}
2614 			if (pn.pn_path[0] == '/') {
2615 				pn_skipslash(&pn);
2616 				VN_RELE(vp);
2617 				VN_RELE(cvp);
2618 				vp = rootdir;
2619 				VN_HOLD(vp);
2620 			} else {
2621 				VN_RELE(cvp);
2622 			}
2623 			continue;
2624 		}
2625 
2626 		VN_RELE(vp);
2627 
2628 		/*
2629 		 * Direct the operation to the persisting filesystem
2630 		 * underlying /dev.  Bail if we encounter a
2631 		 * non-persistent dev entity here.
2632 		 */
2633 		if (cvp->v_vfsp->vfs_fstype == devtype) {
2634 
2635 			if ((VTOSDEV(cvp)->sdev_flags & SDEV_PERSIST) == 0) {
2636 				error = ENOENT;
2637 				VN_RELE(cvp);
2638 				break;
2639 			}
2640 
2641 			if (VTOSDEV(cvp) == NULL) {
2642 				error = ENOENT;
2643 				VN_RELE(cvp);
2644 				break;
2645 			}
2646 			svp = VTOSDEV(cvp);
2647 			if ((vp = svp->sdev_attrvp) == NULL) {
2648 				error = ENOENT;
2649 				VN_RELE(cvp);
2650 				break;
2651 			}
2652 			persisted = 1;
2653 			VN_HOLD(vp);
2654 			VN_RELE(cvp);
2655 			cvp = vp;
2656 		}
2657 
2658 		vp = cvp;
2659 		pn_skipslash(&pn);
2660 	}
2661 
2662 	kmem_free(nm, MAXNAMELEN);
2663 	pn_free(&pn);
2664 
2665 	if (error)
2666 		return (error);
2667 
2668 	/*
2669 	 * Only return persisted nodes in the filesystem underlying /dev.
2670 	 */
2671 	if (!persisted) {
2672 		VN_RELE(vp);
2673 		return (ENOENT);
2674 	}
2675 
2676 	*r_vp = vp;
2677 	return (0);
2678 }
2679 
2680 int
2681 sdev_modctl_readdir(const char *dir, char ***dirlistp, int *npathsp,
2682     int *npathsp_alloc, int checking_empty)
2683 {
2684 	char	**pathlist = NULL;
2685 	char	**newlist = NULL;
2686 	int	npaths = 0;
2687 	int	npaths_alloc = 0;
2688 	dirent64_t *dbuf = NULL;
2689 	int	n;
2690 	char	*s;
2691 	int error;
2692 	vnode_t *vp;
2693 	int eof;
2694 	struct iovec iov;
2695 	struct uio uio;
2696 	struct dirent64 *dp;
2697 	size_t dlen;
2698 	size_t dbuflen;
2699 	int ndirents = 64;
2700 	char *nm;
2701 
2702 	error = sdev_modctl_lookup(dir, &vp);
2703 	sdcmn_err11(("modctl readdir: %s by %s: %s\n",
2704 	    dir, curproc->p_user.u_comm,
2705 	    (error == 0) ? "ok" : "failed"));
2706 	if (error)
2707 		return (error);
2708 
2709 	dlen = ndirents * (sizeof (*dbuf));
2710 	dbuf = kmem_alloc(dlen, KM_SLEEP);
2711 
2712 	uio.uio_iov = &iov;
2713 	uio.uio_iovcnt = 1;
2714 	uio.uio_segflg = UIO_SYSSPACE;
2715 	uio.uio_fmode = 0;
2716 	uio.uio_extflg = UIO_COPY_CACHED;
2717 	uio.uio_loffset = 0;
2718 	uio.uio_llimit = MAXOFFSET_T;
2719 
2720 	eof = 0;
2721 	error = 0;
2722 	while (!error && !eof) {
2723 		uio.uio_resid = dlen;
2724 		iov.iov_base = (char *)dbuf;
2725 		iov.iov_len = dlen;
2726 
2727 		(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
2728 		error = VOP_READDIR(vp, &uio, kcred, &eof, NULL, 0);
2729 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
2730 
2731 		dbuflen = dlen - uio.uio_resid;
2732 
2733 		if (error || dbuflen == 0)
2734 			break;
2735 
2736 		for (dp = dbuf; ((intptr_t)dp < (intptr_t)dbuf + dbuflen);
2737 		    dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) {
2738 
2739 			nm = dp->d_name;
2740 
2741 			if (strcmp(nm, ".") == 0 || strcmp(nm, "..") == 0)
2742 				continue;
2743 			if (npaths == npaths_alloc) {
2744 				npaths_alloc += 64;
2745 				newlist = (char **)
2746 				    kmem_zalloc((npaths_alloc + 1) *
2747 				    sizeof (char *), KM_SLEEP);
2748 				if (pathlist) {
2749 					bcopy(pathlist, newlist,
2750 					    npaths * sizeof (char *));
2751 					kmem_free(pathlist,
2752 					    (npaths + 1) * sizeof (char *));
2753 				}
2754 				pathlist = newlist;
2755 			}
2756 			n = strlen(nm) + 1;
2757 			s = kmem_alloc(n, KM_SLEEP);
2758 			bcopy(nm, s, n);
2759 			pathlist[npaths++] = s;
2760 			sdcmn_err11(("  %s/%s\n", dir, s));
2761 
2762 			/* if checking empty, one entry is as good as many */
2763 			if (checking_empty) {
2764 				eof = 1;
2765 				break;
2766 			}
2767 		}
2768 	}
2769 
2770 exit:
2771 	VN_RELE(vp);
2772 
2773 	if (dbuf)
2774 		kmem_free(dbuf, dlen);
2775 
2776 	if (error)
2777 		return (error);
2778 
2779 	*dirlistp = pathlist;
2780 	*npathsp = npaths;
2781 	*npathsp_alloc = npaths_alloc;
2782 
2783 	return (0);
2784 }
2785 
2786 void
2787 sdev_modctl_readdir_free(char **pathlist, int npaths, int npaths_alloc)
2788 {
2789 	int	i, n;
2790 
2791 	for (i = 0; i < npaths; i++) {
2792 		n = strlen(pathlist[i]) + 1;
2793 		kmem_free(pathlist[i], n);
2794 	}
2795 
2796 	kmem_free(pathlist, (npaths_alloc + 1) * sizeof (char *));
2797 }
2798 
2799 int
2800 sdev_modctl_devexists(const char *path)
2801 {
2802 	vnode_t *vp;
2803 	int error;
2804 
2805 	error = sdev_modctl_lookup(path, &vp);
2806 	sdcmn_err11(("modctl dev exists: %s by %s: %s\n",
2807 	    path, curproc->p_user.u_comm,
2808 	    (error == 0) ? "ok" : "failed"));
2809 	if (error == 0)
2810 		VN_RELE(vp);
2811 
2812 	return (error);
2813 }
2814 
2815 /*
2816  * a generic setattr() function
2817  *
2818  * note: flags only supports AT_UID and AT_GID.
2819  *	 Future enhancements can be done for other types, e.g. AT_MODE
2820  */
2821 int
2822 devname_setattr_func(struct vnode *vp, struct vattr *vap, int flags,
2823     struct cred *cred, int (*callback)(struct sdev_node *, struct vattr *,
2824     int), int protocol)
2825 {
2826 	struct sdev_node	*dv = VTOSDEV(vp);
2827 	struct sdev_node	*parent = dv->sdev_dotdot;
2828 	struct vattr		*get;
2829 	uint_t			mask = vap->va_mask;
2830 	int 			error;
2831 
2832 	/* some sanity checks */
2833 	if (vap->va_mask & AT_NOSET)
2834 		return (EINVAL);
2835 
2836 	if (vap->va_mask & AT_SIZE) {
2837 		if (vp->v_type == VDIR) {
2838 			return (EISDIR);
2839 		}
2840 	}
2841 
2842 	/* no need to set attribute, but do not fail either */
2843 	ASSERT(parent);
2844 	rw_enter(&parent->sdev_contents, RW_READER);
2845 	if (dv->sdev_state == SDEV_ZOMBIE) {
2846 		rw_exit(&parent->sdev_contents);
2847 		return (0);
2848 	}
2849 
2850 	/* If backing store exists, just set it. */
2851 	if (dv->sdev_attrvp) {
2852 		rw_exit(&parent->sdev_contents);
2853 		return (VOP_SETATTR(dv->sdev_attrvp, vap, flags, cred, NULL));
2854 	}
2855 
2856 	/*
2857 	 * Otherwise, for nodes with the persistence attribute, create it.
2858 	 */
2859 	ASSERT(dv->sdev_attr);
2860 	if (SDEV_IS_PERSIST(dv) ||
2861 	    ((vap->va_mask & ~AT_TIMES) != 0 && !SDEV_IS_DYNAMIC(dv))) {
2862 		sdev_vattr_merge(dv, vap);
2863 		rw_enter(&dv->sdev_contents, RW_WRITER);
2864 		error = sdev_shadow_node(dv, cred);
2865 		rw_exit(&dv->sdev_contents);
2866 		rw_exit(&parent->sdev_contents);
2867 
2868 		if (error)
2869 			return (error);
2870 		return (VOP_SETATTR(dv->sdev_attrvp, vap, flags, cred, NULL));
2871 	}
2872 
2873 
2874 	/*
2875 	 * sdev_attr was allocated in sdev_mknode
2876 	 */
2877 	rw_enter(&dv->sdev_contents, RW_WRITER);
2878 	error = secpolicy_vnode_setattr(cred, vp, vap,
2879 	    dv->sdev_attr, flags, sdev_unlocked_access, dv);
2880 	if (error) {
2881 		rw_exit(&dv->sdev_contents);
2882 		rw_exit(&parent->sdev_contents);
2883 		return (error);
2884 	}
2885 
2886 	get = dv->sdev_attr;
2887 	if (mask & AT_MODE) {
2888 		get->va_mode &= S_IFMT;
2889 		get->va_mode |= vap->va_mode & ~S_IFMT;
2890 	}
2891 
2892 	if ((mask & AT_UID) || (mask & AT_GID)) {
2893 		if (mask & AT_UID)
2894 			get->va_uid = vap->va_uid;
2895 		if (mask & AT_GID)
2896 			get->va_gid = vap->va_gid;
2897 		/*
2898 		 * a callback must be provided if the protocol is set
2899 		 */
2900 		if ((protocol & AT_UID) || (protocol & AT_GID)) {
2901 			ASSERT(callback);
2902 			error = callback(dv, get, protocol);
2903 			if (error) {
2904 				rw_exit(&dv->sdev_contents);
2905 				rw_exit(&parent->sdev_contents);
2906 				return (error);
2907 			}
2908 		}
2909 	}
2910 
2911 	if (mask & AT_ATIME)
2912 		get->va_atime = vap->va_atime;
2913 	if (mask & AT_MTIME)
2914 		get->va_mtime = vap->va_mtime;
2915 	if (mask & (AT_MODE | AT_UID | AT_GID | AT_CTIME)) {
2916 		gethrestime(&get->va_ctime);
2917 	}
2918 
2919 	sdev_vattr_merge(dv, get);
2920 	rw_exit(&dv->sdev_contents);
2921 	rw_exit(&parent->sdev_contents);
2922 	return (0);
2923 }
2924 
2925 /*
2926  * a generic inactive() function
2927  */
2928 /*ARGSUSED*/
2929 void
2930 devname_inactive_func(struct vnode *vp, struct cred *cred,
2931     void (*callback)(struct vnode *))
2932 {
2933 	int clean;
2934 	struct sdev_node *dv = VTOSDEV(vp);
2935 	int state;
2936 
2937 	mutex_enter(&vp->v_lock);
2938 	ASSERT(vp->v_count >= 1);
2939 
2940 
2941 	if (vp->v_count == 1 && callback != NULL)
2942 		callback(vp);
2943 
2944 	rw_enter(&dv->sdev_contents, RW_WRITER);
2945 	state = dv->sdev_state;
2946 
2947 	clean = (vp->v_count == 1) && (state == SDEV_ZOMBIE);
2948 
2949 	/*
2950 	 * sdev is a rather bad public citizen. It violates the general
2951 	 * agreement that in memory nodes should always have a valid reference
2952 	 * count on their vnode. But that's not the case here. This means that
2953 	 * we do actually have to distinguish between getting inactive callbacks
2954 	 * for zombies and otherwise. This should probably be fixed.
2955 	 */
2956 	if (clean) {
2957 		/* Remove the . entry to ourselves */
2958 		if (vp->v_type == VDIR) {
2959 			decr_link(dv);
2960 		}
2961 		VERIFY(dv->sdev_nlink == 1);
2962 		decr_link(dv);
2963 		VN_RELE_LOCKED(vp);
2964 		rw_exit(&dv->sdev_contents);
2965 		mutex_exit(&vp->v_lock);
2966 		sdev_nodedestroy(dv, 0);
2967 	} else {
2968 		VN_RELE_LOCKED(vp);
2969 		rw_exit(&dv->sdev_contents);
2970 		mutex_exit(&vp->v_lock);
2971 	}
2972 }
2973