xref: /illumos-gate/usr/src/uts/common/fs/namefs/namevfs.c (revision e75b2cb0)
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) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2017 by Delphix. All rights reserved.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved	*/
28 
29 
30 /*
31  * This file supports the vfs operations for the NAMEFS file system.
32  */
33 
34 #include <sys/types.h>
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/debug.h>
38 #include <sys/errno.h>
39 #include <sys/kmem.h>
40 #include <sys/inline.h>
41 #include <sys/file.h>
42 #include <sys/proc.h>
43 #include <sys/stat.h>
44 #include <sys/statvfs.h>
45 #include <sys/mount.h>
46 #include <sys/sysmacros.h>
47 #include <sys/var.h>
48 #include <sys/vfs.h>
49 #include <sys/vfs_opreg.h>
50 #include <sys/vnode.h>
51 #include <sys/mode.h>
52 #include <sys/pcb.h>
53 #include <sys/signal.h>
54 #include <sys/user.h>
55 #include <sys/uio.h>
56 #include <sys/cred.h>
57 #include <sys/fs/namenode.h>
58 #include <sys/stream.h>
59 #include <sys/strsubr.h>
60 #include <sys/cmn_err.h>
61 #include <sys/modctl.h>
62 #include <fs/fs_subr.h>
63 #include <sys/policy.h>
64 #include <sys/vmem.h>
65 #include <sys/fs/sdev_impl.h>
66 
67 #define	NM_INOQUANT		(64 * 1024)
68 
69 /*
70  * Define global data structures.
71  */
72 dev_t	namedev;
73 int	namefstype;
74 struct	namenode *nm_filevp_hash[NM_FILEVP_HASH_SIZE];
75 struct	vfs namevfs;
76 kmutex_t ntable_lock;
77 
78 static vmem_t	*nm_inoarena;	/* vmem arena to allocate inode no's from */
79 static kmutex_t	nm_inolock;
80 
81 vfsops_t *namefs_vfsops;
82 /*
83  * Functions to allocate node id's starting from 1. Based on vmem routines.
84  * The vmem arena is extended in NM_INOQUANT chunks.
85  */
86 uint64_t
87 namenodeno_alloc(void)
88 {
89 	uint64_t nno;
90 
91 	mutex_enter(&nm_inolock);
92 	nno = (uint64_t)(uintptr_t)
93 	    vmem_alloc(nm_inoarena, 1, VM_NOSLEEP + VM_FIRSTFIT);
94 	if (nno == 0) {
95 		(void) vmem_add(nm_inoarena, (void *)(vmem_size(nm_inoarena,
96 		    VMEM_ALLOC | VMEM_FREE) + 1), NM_INOQUANT, VM_SLEEP);
97 		nno = (uint64_t)(uintptr_t)
98 		    vmem_alloc(nm_inoarena, 1, VM_SLEEP + VM_FIRSTFIT);
99 		ASSERT(nno != 0);
100 	}
101 	mutex_exit(&nm_inolock);
102 	ASSERT32(nno <= ULONG_MAX);
103 	return (nno);
104 }
105 
106 static void
107 namenodeno_init(void)
108 {
109 	nm_inoarena = vmem_create("namefs_inodes", (void *)1, NM_INOQUANT, 1,
110 	    NULL, NULL, NULL, 1, VM_SLEEP);
111 	mutex_init(&nm_inolock, NULL, MUTEX_DEFAULT, NULL);
112 }
113 
114 void
115 namenodeno_free(uint64_t nn)
116 {
117 	void *vaddr = (void *)(uintptr_t)nn;
118 
119 	ASSERT32((uint64_t)(uintptr_t)vaddr == nn);
120 
121 	mutex_enter(&nm_inolock);
122 	vmem_free(nm_inoarena, vaddr, 1);
123 	mutex_exit(&nm_inolock);
124 }
125 
126 /*
127  * Insert a namenode into the nm_filevp_hash table.
128  *
129  * Each link has a unique namenode with a unique nm_mountvp field.
130  * The nm_filevp field of the namenode need not be unique, since a
131  * file descriptor may be mounted to multiple nodes at the same time.
132  * We hash on nm_filevp since that's what discriminates the searches
133  * in namefind() and nm_unmountall().
134  */
135 void
136 nameinsert(struct namenode *nodep)
137 {
138 	struct namenode **bucket;
139 
140 	ASSERT(MUTEX_HELD(&ntable_lock));
141 
142 	bucket = NM_FILEVP_HASH(nodep->nm_filevp);
143 	nodep->nm_nextp = *bucket;
144 	*bucket = nodep;
145 }
146 
147 /*
148  * Remove a namenode from the hash table, if present.
149  */
150 void
151 nameremove(struct namenode *nodep)
152 {
153 	struct namenode *np, **npp;
154 
155 	ASSERT(MUTEX_HELD(&ntable_lock));
156 
157 	for (npp = NM_FILEVP_HASH(nodep->nm_filevp); (np = *npp) != NULL;
158 	    npp = &np->nm_nextp) {
159 		if (np == nodep) {
160 			*npp = np->nm_nextp;
161 			return;
162 		}
163 	}
164 }
165 
166 /*
167  * Search for a namenode that has a nm_filevp == vp and nm_mountpt == mnt.
168  * If mnt is NULL, return the first link with nm_filevp of vp.
169  * Returns namenode pointer on success, NULL on failure.
170  */
171 struct namenode *
172 namefind(vnode_t *vp, vnode_t *mnt)
173 {
174 	struct namenode *np;
175 
176 	ASSERT(MUTEX_HELD(&ntable_lock));
177 	for (np = *NM_FILEVP_HASH(vp); np != NULL; np = np->nm_nextp)
178 		if (np->nm_filevp == vp &&
179 		    (mnt == NULL || np->nm_mountpt == mnt))
180 			break;
181 	return (np);
182 }
183 
184 /*
185  * Force the unmouting of a file descriptor from ALL of the nodes
186  * that it was mounted to.
187  * At the present time, the only usage for this routine is in the
188  * event one end of a pipe was mounted. At the time the unmounted
189  * end gets closed down, the mounted end is forced to be unmounted.
190  *
191  * This routine searches the namenode hash list for all namenodes
192  * that have a nm_filevp field equal to vp. Each time one is found,
193  * the dounmount() routine is called. This causes the nm_unmount()
194  * routine to be called and thus, the file descriptor is unmounted
195  * from the node.
196  *
197  * At the start of this routine, the reference count for vp is
198  * incremented to protect the vnode from being released in the
199  * event the mount was the only thing keeping the vnode active.
200  * If that is the case, the VOP_CLOSE operation is applied to
201  * the vnode, prior to it being released.
202  */
203 static int
204 nm_umountall(vnode_t *vp, cred_t *crp)
205 {
206 	vfs_t *vfsp;
207 	struct namenode *nodep;
208 	int error = 0;
209 	int realerr = 0;
210 
211 	/*
212 	 * For each namenode that is associated with the file:
213 	 * If the v_vfsp field is not namevfs, dounmount it.  Otherwise,
214 	 * it was created in nm_open() and will be released in time.
215 	 * The following loop replicates some code from nm_find.  That
216 	 * routine can't be used as is since the list isn't strictly
217 	 * consumed as it is traversed.
218 	 */
219 	mutex_enter(&ntable_lock);
220 	nodep = *NM_FILEVP_HASH(vp);
221 	while (nodep) {
222 		if (nodep->nm_filevp == vp &&
223 		    (vfsp = NMTOV(nodep)->v_vfsp) != NULL &&
224 		    vfsp != &namevfs && (NMTOV(nodep)->v_flag & VROOT)) {
225 
226 			/*
227 			 * If the vn_vfswlock fails, skip the vfs since
228 			 * somebody else may be unmounting it.
229 			 */
230 			if (vn_vfswlock(vfsp->vfs_vnodecovered)) {
231 				realerr = EBUSY;
232 				nodep = nodep->nm_nextp;
233 				continue;
234 			}
235 
236 			/*
237 			 * Can't hold ntable_lock across call to do_unmount
238 			 * because nm_unmount tries to acquire it.  This means
239 			 * there is a window where another mount of vp can
240 			 * happen so it is possible that after nm_unmountall
241 			 * there are still some mounts.  This situation existed
242 			 * without MT locking because dounmount can sleep
243 			 * so another mount could happen during that time.
244 			 * This situation is unlikely and doesn't really cause
245 			 * any problems.
246 			 */
247 			mutex_exit(&ntable_lock);
248 			if ((error = dounmount(vfsp, 0, crp)) != 0)
249 				realerr = error;
250 			mutex_enter(&ntable_lock);
251 			/*
252 			 * Since we dropped the ntable_lock, we
253 			 * have to start over from the beginning.
254 			 * If for some reasons dounmount() fails,
255 			 * start from beginning means that we will keep on
256 			 * trying unless another thread unmounts it for us.
257 			 */
258 			nodep = *NM_FILEVP_HASH(vp);
259 		} else
260 			nodep = nodep->nm_nextp;
261 	}
262 	mutex_exit(&ntable_lock);
263 	return (realerr);
264 }
265 
266 /*
267  * Force the unmouting of a file descriptor from ALL of the nodes
268  * that it was mounted to.  XXX: fifo_close() calls this routine.
269  *
270  * nm_umountall() may return EBUSY.
271  * nm_unmountall() will keep on trying until it succeeds.
272  */
273 int
274 nm_unmountall(vnode_t *vp, cred_t *crp)
275 {
276 	int error;
277 
278 	/*
279 	 * Nm_umuontall() returns only if it succeeds or
280 	 * return with error EBUSY.  If EBUSY, that means
281 	 * it cannot acquire the lock on the covered vnode,
282 	 * and we will keep on trying.
283 	 */
284 	for (;;) {
285 		error = nm_umountall(vp, crp);
286 		if (error != EBUSY)
287 			break;
288 		delay(1);	/* yield cpu briefly, then try again */
289 	}
290 	return (error);
291 }
292 
293 /*
294  * Mount a file descriptor onto the node in the file system.
295  * Create a new vnode, update the attributes with info from the
296  * file descriptor and the mount point.  The mask, mode, uid, gid,
297  * atime, mtime and ctime are taken from the mountpt.  Link count is
298  * set to one, the file system id is namedev and nodeid is unique
299  * for each mounted object.  Other attributes are taken from mount point.
300  * Make sure user is owner (or root) with write permissions on mount point.
301  * Hash the new vnode and return 0.
302  * Upon entry to this routine, the file descriptor is in the
303  * fd field of a struct namefd.  Copy that structure from user
304  * space and retrieve the file descriptor.
305  */
306 static int
307 nm_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *crp)
308 {
309 	struct namefd namefdp;
310 	struct vnode *filevp;		/* file descriptor vnode */
311 	struct file *fp;
312 	struct vnode *newvp;		/* vnode representing this mount */
313 	struct vnode *rvp;		/* realvp (if any) for the mountpt */
314 	struct namenode *nodep;		/* namenode for this mount */
315 	struct vattr filevattr;		/* attributes of file dec.  */
316 	struct vattr *vattrp;		/* attributes of this mount */
317 	char *resource_name;
318 	char *resource_nodetype;
319 	statvfs64_t *svfsp;
320 	int error = 0;
321 
322 	/*
323 	 * Get the file descriptor from user space.
324 	 * Make sure the file descriptor is valid and has an
325 	 * associated file pointer.
326 	 * If so, extract the vnode from the file pointer.
327 	 */
328 	if (uap->datalen != sizeof (struct namefd))
329 		return (EINVAL);
330 
331 	if (copyin(uap->dataptr, &namefdp, uap->datalen))
332 		return (EFAULT);
333 
334 	if ((fp = getf(namefdp.fd)) == NULL)
335 		return (EBADF);
336 
337 	/*
338 	 * If the mount point already has something mounted
339 	 * on it, disallow this mount.  (This restriction may
340 	 * be removed in a later release).
341 	 * Or unmount has completed but the namefs ROOT vnode
342 	 * count has not decremented to zero, disallow this mount.
343 	 */
344 
345 	mutex_enter(&mvp->v_lock);
346 	if ((mvp->v_flag & VROOT) ||
347 	    vfs_matchops(mvp->v_vfsp, namefs_vfsops)) {
348 		mutex_exit(&mvp->v_lock);
349 		releasef(namefdp.fd);
350 		return (EBUSY);
351 	}
352 	mutex_exit(&mvp->v_lock);
353 
354 	/*
355 	 * Cannot allow users to fattach() in /dev/pts.
356 	 * First, there is no need for doing so and secondly
357 	 * we cannot allow arbitrary users to park on a node in
358 	 * /dev/pts or /dev/vt.
359 	 */
360 	rvp = NULLVP;
361 	if (vn_matchops(mvp, spec_getvnodeops()) &&
362 	    VOP_REALVP(mvp, &rvp, NULL) == 0 && rvp &&
363 	    (vn_matchops(rvp, devpts_getvnodeops()) ||
364 	    vn_matchops(rvp, devvt_getvnodeops()))) {
365 		releasef(namefdp.fd);
366 		return (ENOTSUP);
367 	}
368 
369 	filevp = fp->f_vnode;
370 	if (filevp->v_type == VDIR || filevp->v_type == VPORT) {
371 		releasef(namefdp.fd);
372 		return (EINVAL);
373 	}
374 
375 	/*
376 	 * If the fd being mounted refers to neither a door nor a stream,
377 	 * make sure the caller is privileged.
378 	 */
379 	if (filevp->v_type != VDOOR && filevp->v_stream == NULL) {
380 		if (secpolicy_fs_mount(crp, filevp, vfsp) != 0) {
381 			/* fd is neither a stream nor a door */
382 			releasef(namefdp.fd);
383 			return (EINVAL);
384 		}
385 	}
386 
387 	/*
388 	 * Make sure the file descriptor is not the root of some
389 	 * file system.
390 	 * If it's not, create a reference and allocate a namenode
391 	 * to represent this mount request.
392 	 */
393 	if (filevp->v_flag & VROOT) {
394 		releasef(namefdp.fd);
395 		return (EBUSY);
396 	}
397 
398 	nodep = kmem_zalloc(sizeof (struct namenode), KM_SLEEP);
399 
400 	mutex_init(&nodep->nm_lock, NULL, MUTEX_DEFAULT, NULL);
401 	vattrp = &nodep->nm_vattr;
402 	vattrp->va_mask = AT_ALL;
403 	if (error = VOP_GETATTR(mvp, vattrp, 0, crp, NULL))
404 		goto out;
405 
406 	filevattr.va_mask = AT_ALL;
407 	if (error = VOP_GETATTR(filevp, &filevattr, 0, crp, NULL))
408 		goto out;
409 	/*
410 	 * Make sure the user is the owner of the mount point
411 	 * or has sufficient privileges.
412 	 */
413 	if (error = secpolicy_vnode_owner(crp, vattrp->va_uid))
414 		goto out;
415 
416 	/*
417 	 * Make sure the user has write permissions on the
418 	 * mount point (or has sufficient privileges).
419 	 */
420 	if (secpolicy_vnode_access2(crp, mvp, vattrp->va_uid, vattrp->va_mode,
421 	    VWRITE) != 0) {
422 		error = EACCES;
423 		goto out;
424 	}
425 
426 	/*
427 	 * If the file descriptor has file/record locking, don't
428 	 * allow the mount to succeed.
429 	 */
430 	if (vn_has_flocks(filevp)) {
431 		error = EACCES;
432 		goto out;
433 	}
434 
435 	/*
436 	 * Initialize the namenode.
437 	 */
438 	if (filevp->v_stream) {
439 		struct stdata *stp = filevp->v_stream;
440 		mutex_enter(&stp->sd_lock);
441 		stp->sd_flag |= STRMOUNT;
442 		mutex_exit(&stp->sd_lock);
443 	}
444 	nodep->nm_filevp = filevp;
445 	mutex_enter(&fp->f_tlock);
446 	fp->f_count++;
447 	mutex_exit(&fp->f_tlock);
448 
449 	releasef(namefdp.fd);
450 	nodep->nm_filep = fp;
451 	nodep->nm_mountpt = mvp;
452 
453 	/*
454 	 * The attributes for the mounted file descriptor were initialized
455 	 * above by applying VOP_GETATTR to the mount point.  Some of
456 	 * the fields of the attributes structure will be overwritten
457 	 * by the attributes from the file descriptor.
458 	 */
459 	vattrp->va_type    = filevattr.va_type;
460 	vattrp->va_fsid    = namedev;
461 	vattrp->va_nodeid  = namenodeno_alloc();
462 	vattrp->va_nlink   = 1;
463 	vattrp->va_size    = filevattr.va_size;
464 	vattrp->va_rdev    = filevattr.va_rdev;
465 	vattrp->va_blksize = filevattr.va_blksize;
466 	vattrp->va_nblocks = filevattr.va_nblocks;
467 	vattrp->va_seq	   = 0;
468 
469 	/*
470 	 * Initialize new vnode structure for the mounted file descriptor.
471 	 */
472 	nodep->nm_vnode = vn_alloc(KM_SLEEP);
473 	newvp = NMTOV(nodep);
474 
475 	newvp->v_flag = filevp->v_flag | VROOT | VNOMAP | VNOSWAP;
476 	vn_setops(newvp, nm_vnodeops);
477 	newvp->v_vfsp = vfsp;
478 	newvp->v_stream = filevp->v_stream;
479 	newvp->v_type = filevp->v_type;
480 	newvp->v_rdev = filevp->v_rdev;
481 	newvp->v_data = (caddr_t)nodep;
482 	VFS_HOLD(vfsp);
483 	vn_exists(newvp);
484 
485 	/*
486 	 * Initialize the vfs structure.
487 	 */
488 	vfsp->vfs_vnodecovered = NULL;
489 	vfsp->vfs_flag |= VFS_UNLINKABLE;
490 	vfsp->vfs_bsize = 1024;
491 	vfsp->vfs_fstype = namefstype;
492 	vfs_make_fsid(&vfsp->vfs_fsid, namedev, namefstype);
493 	vfsp->vfs_data = (caddr_t)nodep;
494 	vfsp->vfs_dev = namedev;
495 	vfsp->vfs_bcount = 0;
496 
497 	/*
498 	 * Set the name we mounted from.
499 	 */
500 	switch (filevp->v_type) {
501 	case VPROC:	/* VOP_GETATTR() translates this to VREG */
502 	case VREG:	resource_nodetype = "file"; break;
503 	case VDIR:	resource_nodetype = "directory"; break;
504 	case VBLK:	resource_nodetype = "device"; break;
505 	case VCHR:	resource_nodetype = "device"; break;
506 	case VLNK:	resource_nodetype = "link"; break;
507 	case VFIFO:	resource_nodetype = "fifo"; break;
508 	case VDOOR:	resource_nodetype = "door"; break;
509 	case VSOCK:	resource_nodetype = "socket"; break;
510 	default:	resource_nodetype = "resource"; break;
511 	}
512 
513 #define	RESOURCE_NAME_SZ 128 /* Maximum length of the resource name */
514 	resource_name = kmem_alloc(RESOURCE_NAME_SZ, KM_SLEEP);
515 	svfsp = kmem_alloc(sizeof (statvfs64_t), KM_SLEEP);
516 
517 	error = VFS_STATVFS(filevp->v_vfsp, svfsp);
518 	if (error == 0) {
519 		(void) snprintf(resource_name, RESOURCE_NAME_SZ,
520 		    "unspecified_%s_%s", svfsp->f_basetype, resource_nodetype);
521 	} else {
522 		(void) snprintf(resource_name, RESOURCE_NAME_SZ,
523 		    "unspecified_%s", resource_nodetype);
524 	}
525 
526 	vfs_setresource(vfsp, resource_name, 0);
527 
528 	kmem_free(svfsp, sizeof (statvfs64_t));
529 	kmem_free(resource_name, RESOURCE_NAME_SZ);
530 #undef RESOURCE_NAME_SZ
531 
532 	/*
533 	 * Insert the namenode.
534 	 */
535 	mutex_enter(&ntable_lock);
536 	nameinsert(nodep);
537 	mutex_exit(&ntable_lock);
538 	return (0);
539 out:
540 	releasef(namefdp.fd);
541 	kmem_free(nodep, sizeof (struct namenode));
542 	return (error);
543 }
544 
545 /*
546  * Unmount a file descriptor from a node in the file system.
547  * If the user is not the owner of the file and is not privileged,
548  * the request is denied.
549  * Otherwise, remove the namenode from the hash list.
550  * If the mounted file descriptor was that of a stream and this
551  * was the last mount of the stream, turn off the STRMOUNT flag.
552  * If the rootvp is referenced other than through the mount,
553  * nm_inactive will clean up.
554  */
555 static int
556 nm_unmount(vfs_t *vfsp, int flag, cred_t *crp)
557 {
558 	struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
559 	vnode_t *vp, *thisvp;
560 	struct file *fp = NULL;
561 
562 	ASSERT((nodep->nm_flag & NMNMNT) == 0);
563 
564 	/*
565 	 * forced unmount is not supported by this file system
566 	 * and thus, ENOTSUP, is being returned.
567 	 */
568 	if (flag & MS_FORCE) {
569 		return (ENOTSUP);
570 	}
571 
572 	vp = nodep->nm_filevp;
573 	mutex_enter(&nodep->nm_lock);
574 	if (secpolicy_vnode_owner(crp, nodep->nm_vattr.va_uid) != 0) {
575 		mutex_exit(&nodep->nm_lock);
576 		return (EPERM);
577 	}
578 
579 	mutex_exit(&nodep->nm_lock);
580 
581 	mutex_enter(&ntable_lock);
582 	nameremove(nodep);
583 	thisvp = NMTOV(nodep);
584 	mutex_enter(&thisvp->v_lock);
585 	VN_RELE_LOCKED(thisvp);
586 	if (thisvp->v_count == 0) {
587 		fp = nodep->nm_filep;
588 		mutex_exit(&thisvp->v_lock);
589 		vn_invalid(thisvp);
590 		vn_free(thisvp);
591 		VFS_RELE(vfsp);
592 		namenodeno_free(nodep->nm_vattr.va_nodeid);
593 		kmem_free(nodep, sizeof (struct namenode));
594 	} else {
595 		thisvp->v_flag &= ~VROOT;
596 		mutex_exit(&thisvp->v_lock);
597 	}
598 	if (namefind(vp, NULLVP) == NULL && vp->v_stream) {
599 		struct stdata *stp = vp->v_stream;
600 		mutex_enter(&stp->sd_lock);
601 		stp->sd_flag &= ~STRMOUNT;
602 		mutex_exit(&stp->sd_lock);
603 	}
604 	mutex_exit(&ntable_lock);
605 	if (fp != NULL)
606 		(void) closef(fp);
607 	return (0);
608 }
609 
610 /*
611  * Create a reference to the root of a mounted file descriptor.
612  * This routine is called from lookupname() in the event a path
613  * is being searched that has a mounted file descriptor in it.
614  */
615 static int
616 nm_root(vfs_t *vfsp, vnode_t **vpp)
617 {
618 	struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
619 	struct vnode *vp = NMTOV(nodep);
620 
621 	VN_HOLD(vp);
622 	*vpp = vp;
623 	return (0);
624 }
625 
626 /*
627  * Return in sp the status of this file system.
628  */
629 static int
630 nm_statvfs(vfs_t *vfsp, struct statvfs64 *sp)
631 {
632 	dev32_t d32;
633 
634 	bzero(sp, sizeof (*sp));
635 	sp->f_bsize	= 1024;
636 	sp->f_frsize	= 1024;
637 	(void) cmpldev(&d32, vfsp->vfs_dev);
638 	sp->f_fsid = d32;
639 	(void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
640 	sp->f_flag	= vf_to_stf(vfsp->vfs_flag);
641 	return (0);
642 }
643 
644 /*
645  * Since this file system has no disk blocks of its own, apply
646  * the VOP_FSYNC operation on the mounted file descriptor.
647  */
648 static int
649 nm_sync(vfs_t *vfsp, short flag, cred_t *crp)
650 {
651 	struct namenode *nodep;
652 
653 	if (vfsp == NULL)
654 		return (0);
655 
656 	nodep = (struct namenode *)vfsp->vfs_data;
657 	if (flag & SYNC_CLOSE)
658 		return (nm_umountall(nodep->nm_filevp, crp));
659 
660 	return (VOP_FSYNC(nodep->nm_filevp, FSYNC, crp, NULL));
661 }
662 
663 /*
664  * File system initialization routine. Save the file system type,
665  * establish a file system device number and initialize nm_filevp_hash[].
666  */
667 int
668 nameinit(int fstype, char *name)
669 {
670 	static const fs_operation_def_t nm_vfsops_template[] = {
671 		VFSNAME_MOUNT,		{ .vfs_mount = nm_mount },
672 		VFSNAME_UNMOUNT,	{ .vfs_unmount = nm_unmount },
673 		VFSNAME_ROOT,		{ .vfs_root = nm_root },
674 		VFSNAME_STATVFS,	{ .vfs_statvfs = nm_statvfs },
675 		VFSNAME_SYNC,		{ .vfs_sync = nm_sync },
676 		NULL,			NULL
677 	};
678 	static const fs_operation_def_t nm_dummy_vfsops_template[] = {
679 		VFSNAME_STATVFS,	{ .vfs_statvfs = nm_statvfs },
680 		VFSNAME_SYNC,		{ .vfs_sync = nm_sync },
681 		NULL,			NULL
682 	};
683 	int error;
684 	int dev;
685 	vfsops_t *dummy_vfsops;
686 
687 	error = vfs_setfsops(fstype, nm_vfsops_template, &namefs_vfsops);
688 	if (error != 0) {
689 		cmn_err(CE_WARN, "nameinit: bad vfs ops template");
690 		return (error);
691 	}
692 
693 	error = vfs_makefsops(nm_dummy_vfsops_template, &dummy_vfsops);
694 	if (error != 0) {
695 		(void) vfs_freevfsops_by_type(fstype);
696 		cmn_err(CE_WARN, "nameinit: bad dummy vfs ops template");
697 		return (error);
698 	}
699 
700 	error = vn_make_ops(name, nm_vnodeops_template, &nm_vnodeops);
701 	if (error != 0) {
702 		(void) vfs_freevfsops_by_type(fstype);
703 		vfs_freevfsops(dummy_vfsops);
704 		cmn_err(CE_WARN, "nameinit: bad vnode ops template");
705 		return (error);
706 	}
707 
708 	namefstype = fstype;
709 
710 	if ((dev = getudev()) == (major_t)-1) {
711 		cmn_err(CE_WARN, "nameinit: can't get unique device");
712 		dev = 0;
713 	}
714 	mutex_init(&ntable_lock, NULL, MUTEX_DEFAULT, NULL);
715 	namedev = makedevice(dev, 0);
716 	bzero(nm_filevp_hash, sizeof (nm_filevp_hash));
717 	vfs_setops(&namevfs, dummy_vfsops);
718 	namevfs.vfs_vnodecovered = NULL;
719 	namevfs.vfs_bsize = 1024;
720 	namevfs.vfs_fstype = namefstype;
721 	vfs_make_fsid(&namevfs.vfs_fsid, namedev, namefstype);
722 	namevfs.vfs_dev = namedev;
723 	return (0);
724 }
725 
726 static mntopts_t nm_mntopts = {
727 	.mo_count = 0,
728 	.mo_list = NULL
729 };
730 
731 static vfsdef_t vfw = {
732 	VFSDEF_VERSION,
733 	"namefs",
734 	nameinit,
735 	VSW_HASPROTO | VSW_ZMOUNT,
736 	&nm_mntopts
737 };
738 
739 /*
740  * Module linkage information for the kernel.
741  */
742 static struct modlfs modlfs = {
743 	&mod_fsops, "filesystem for namefs", &vfw
744 };
745 
746 static struct modlinkage modlinkage = {
747 	MODREV_1, (void *)&modlfs, NULL
748 };
749 
750 int
751 _init(void)
752 {
753 	namenodeno_init();
754 	return (mod_install(&modlinkage));
755 }
756 
757 int
758 _fini(void)
759 {
760 	return (EBUSY);
761 }
762 
763 int
764 _info(struct modinfo *modinfop)
765 {
766 	return (mod_info(&modlinkage, modinfop));
767 }
768