xref: /titanic_41/usr/src/uts/common/fs/namefs/namevno.c (revision 0cfdb6036e046270988a17ac442e4d717d426a44)
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 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
22 /*	  All Rights Reserved  	*/
23 
24 
25 /*
26  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 /*
33  * This file defines the vnode operations for mounted file descriptors.
34  * The routines in this file act as a layer between the NAMEFS file
35  * system and SPECFS/FIFOFS.  With the exception of nm_open(), nm_setattr(),
36  * nm_getattr() and nm_access(), the routines simply apply the VOP operation
37  * to the vnode representing the file descriptor.  This switches control
38  * to the underlying file system to which the file descriptor belongs.
39  */
40 #include <sys/types.h>
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/cred.h>
44 #include <sys/errno.h>
45 #include <sys/time.h>
46 #include <sys/file.h>
47 #include <sys/fcntl.h>
48 #include <sys/flock.h>
49 #include <sys/kmem.h>
50 #include <sys/uio.h>
51 #include <sys/vfs.h>
52 #include <sys/vfs_opreg.h>
53 #include <sys/vnode.h>
54 #include <sys/pcb.h>
55 #include <sys/signal.h>
56 #include <sys/user.h>
57 #include <sys/proc.h>
58 #include <sys/conf.h>
59 #include <sys/debug.h>
60 #include <vm/seg.h>
61 #include <sys/fs/namenode.h>
62 #include <sys/stream.h>
63 #include <fs/fs_subr.h>
64 #include <sys/policy.h>
65 
66 /*
67  * Create a reference to the vnode representing the file descriptor.
68  * Then, apply the VOP_OPEN operation to that vnode.
69  *
70  * The vnode for the file descriptor may be switched under you.
71  * If it is, search the hash list for an nodep - nodep->nm_filevp
72  * pair. If it exists, return that nodep to the user.
73  * If it does not exist, create a new namenode to attach
74  * to the nodep->nm_filevp then place the pair on the hash list.
75  *
76  * Newly created objects are like children/nodes in the mounted
77  * file system, with the parent being the initial mount.
78  */
79 int
80 nm_open(vnode_t **vpp, int flag, cred_t *crp, caller_context_t *ct)
81 {
82 	struct namenode *nodep = VTONM(*vpp);
83 	int error = 0;
84 	struct namenode *newnamep;
85 	struct vnode *newvp;
86 	struct vnode *infilevp;
87 	struct vnode *outfilevp;
88 
89 	/*
90 	 * If the vnode is switched under us, the corresponding
91 	 * VN_RELE for this VN_HOLD will be done by the file system
92 	 * performing the switch. Otherwise, the corresponding
93 	 * VN_RELE will be done by nm_close().
94 	 */
95 	infilevp = outfilevp = nodep->nm_filevp;
96 	VN_HOLD(outfilevp);
97 
98 	if ((error = VOP_OPEN(&outfilevp, flag, crp, ct)) != 0) {
99 		VN_RELE(outfilevp);
100 		return (error);
101 	}
102 	if (infilevp != outfilevp) {
103 		/*
104 		 * See if the new filevp (outfilevp) is already associated
105 		 * with the mount point. If it is, then it already has a
106 		 * namenode associated with it.
107 		 */
108 		mutex_enter(&ntable_lock);
109 		if ((newnamep =
110 		    namefind(outfilevp, nodep->nm_mountpt)) != NULL) {
111 			struct vnode *vp = NMTOV(newnamep);
112 
113 			VN_HOLD(vp);
114 			goto gotit;
115 		}
116 
117 		newnamep = kmem_zalloc(sizeof (struct namenode), KM_SLEEP);
118 		newvp = vn_alloc(KM_SLEEP);
119 		newnamep->nm_vnode = newvp;
120 
121 		mutex_init(&newnamep->nm_lock, NULL, MUTEX_DEFAULT, NULL);
122 
123 		mutex_enter(&nodep->nm_lock);
124 		newvp->v_flag = ((*vpp)->v_flag | VNOMAP | VNOSWAP) & ~VROOT;
125 		vn_setops(newvp, vn_getops(*vpp));
126 		newvp->v_vfsp = &namevfs;
127 		newvp->v_stream = outfilevp->v_stream;
128 		newvp->v_type = outfilevp->v_type;
129 		newvp->v_rdev = outfilevp->v_rdev;
130 		newvp->v_data = (caddr_t)newnamep;
131 		vn_exists(newvp);
132 		bcopy(&nodep->nm_vattr, &newnamep->nm_vattr, sizeof (vattr_t));
133 		newnamep->nm_vattr.va_type = outfilevp->v_type;
134 		newnamep->nm_vattr.va_nodeid = namenodeno_alloc();
135 		newnamep->nm_vattr.va_size = (u_offset_t)0;
136 		newnamep->nm_vattr.va_rdev = outfilevp->v_rdev;
137 		newnamep->nm_flag = NMNMNT;
138 		newnamep->nm_filevp = outfilevp;
139 		newnamep->nm_filep = nodep->nm_filep;
140 		newnamep->nm_mountpt = nodep->nm_mountpt;
141 		mutex_exit(&nodep->nm_lock);
142 
143 		/*
144 		 * Insert the new namenode into the hash list.
145 		 */
146 		nameinsert(newnamep);
147 gotit:
148 		mutex_exit(&ntable_lock);
149 		/*
150 		 * Release the above reference to the infilevp, the reference
151 		 * to the NAMEFS vnode, create a reference to the new vnode
152 		 * and return the new vnode to the user.
153 		 */
154 		VN_RELE(*vpp);
155 		*vpp = NMTOV(newnamep);
156 	}
157 	return (0);
158 }
159 
160 /*
161  * Close a mounted file descriptor.
162  * Remove any locks and apply the VOP_CLOSE operation to the vnode for
163  * the file descriptor.
164  */
165 static int
166 nm_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *crp,
167 	caller_context_t *ct)
168 {
169 	struct namenode *nodep = VTONM(vp);
170 	int error = 0;
171 
172 	(void) cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
173 	cleanshares(vp, ttoproc(curthread)->p_pid);
174 	error = VOP_CLOSE(nodep->nm_filevp, flag, count, offset, crp, ct);
175 	if (count == 1) {
176 		(void) VOP_FSYNC(nodep->nm_filevp, FSYNC, crp, ct);
177 		/*
178 		 * Before VN_RELE() we need to remove the vnode from
179 		 * the hash table.  We should only do so in the  NMNMNT case.
180 		 * In other cases, nodep->nm_filep keeps a reference
181 		 * to nm_filevp and the entry in the hash table doesn't
182 		 * hurt.
183 		 */
184 		if ((nodep->nm_flag & NMNMNT) != 0) {
185 			mutex_enter(&ntable_lock);
186 			nameremove(nodep);
187 			mutex_exit(&ntable_lock);
188 		}
189 		VN_RELE(nodep->nm_filevp);
190 	}
191 	return (error);
192 }
193 
194 static int
195 nm_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *crp,
196 	caller_context_t *ct)
197 {
198 	return (VOP_READ(VTONM(vp)->nm_filevp, uiop, ioflag, crp, ct));
199 }
200 
201 static int
202 nm_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *crp,
203 	caller_context_t *ct)
204 {
205 	return (VOP_WRITE(VTONM(vp)->nm_filevp, uiop, ioflag, crp, ct));
206 }
207 
208 static int
209 nm_ioctl(vnode_t *vp, int cmd, intptr_t arg, int mode, cred_t *cr, int *rvalp,
210 	caller_context_t *ct)
211 {
212 	return (VOP_IOCTL(VTONM(vp)->nm_filevp, cmd, arg, mode, cr, rvalp, ct));
213 }
214 
215 /*
216  * Return in vap the attributes that are stored in the namenode
217  * structure.  Only the size is taken from the mounted object.
218  */
219 /* ARGSUSED */
220 static int
221 nm_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *crp,
222 	caller_context_t *ct)
223 {
224 	struct namenode *nodep = VTONM(vp);
225 	struct vattr va;
226 	int error;
227 
228 	mutex_enter(&nodep->nm_lock);
229 	bcopy(&nodep->nm_vattr, vap, sizeof (vattr_t));
230 	mutex_exit(&nodep->nm_lock);
231 
232 	if ((va.va_mask = vap->va_mask & AT_SIZE) != 0) {
233 		if (error = VOP_GETATTR(nodep->nm_filevp, &va, flags, crp, ct))
234 			return (error);
235 		vap->va_size = va.va_size;
236 	}
237 
238 	return (0);
239 }
240 
241 /*
242  * Standard access() like check.  Figure out which mode bits apply
243  * to the caller then pass the missing mode bits to the secpolicy function.
244  */
245 static int
246 nm_access_unlocked(void *vnp, int mode, cred_t *crp)
247 {
248 	struct namenode *nodep = vnp;
249 	int shift = 0;
250 
251 	if (crgetuid(crp) != nodep->nm_vattr.va_uid) {
252 		shift += 3;
253 		if (!groupmember(nodep->nm_vattr.va_gid, crp))
254 			shift += 3;
255 	}
256 	mode &= ~(nodep->nm_vattr.va_mode << shift);
257 
258 	if (mode == 0)
259 		return (0);
260 
261 	return (secpolicy_vnode_access(crp, NMTOV(nodep),
262 						nodep->nm_vattr.va_uid, mode));
263 }
264 /*
265  * Set the attributes of the namenode from the attributes in vap.
266  */
267 /* ARGSUSED */
268 static int
269 nm_setattr(
270 	vnode_t *vp,
271 	vattr_t *vap,
272 	int flags,
273 	cred_t *crp,
274 	caller_context_t *ctp)
275 {
276 	struct namenode *nodep = VTONM(vp);
277 	struct vattr *nmvap = &nodep->nm_vattr;
278 	long mask = vap->va_mask;
279 	int error = 0;
280 
281 	/*
282 	 * Cannot set these attributes.
283 	 */
284 	if (mask & (AT_NOSET|AT_SIZE))
285 		return (EINVAL);
286 
287 	(void) VOP_RWLOCK(nodep->nm_filevp, V_WRITELOCK_TRUE, ctp);
288 	mutex_enter(&nodep->nm_lock);
289 
290 	/*
291 	 * Change ownership/group/time/access mode of mounted file
292 	 * descriptor.
293 	 */
294 
295 	error = secpolicy_vnode_setattr(crp, vp, vap, nmvap, flags,
296 					    nm_access_unlocked, nodep);
297 	if (error)
298 		goto out;
299 
300 	mask = vap->va_mask;
301 	/*
302 	 * If request to change mode, copy new
303 	 * mode into existing attribute structure.
304 	 */
305 	if (mask & AT_MODE)
306 		nmvap->va_mode = vap->va_mode & ~VSVTX;
307 
308 	/*
309 	 * If request was to change user or group, turn off suid and sgid
310 	 * bits.
311 	 * If the system was configured with the "rstchown" option, the
312 	 * owner is not permitted to give away the file, and can change
313 	 * the group id only to a group of which he or she is a member.
314 	 */
315 	if (mask & AT_UID)
316 		nmvap->va_uid = vap->va_uid;
317 	if (mask & AT_GID)
318 		nmvap->va_gid = vap->va_gid;
319 	/*
320 	 * If request is to modify times, make sure user has write
321 	 * permissions on the file.
322 	 */
323 	if (mask & AT_ATIME)
324 		nmvap->va_atime = vap->va_atime;
325 	if (mask & AT_MTIME) {
326 		nmvap->va_mtime = vap->va_mtime;
327 		gethrestime(&nmvap->va_ctime);
328 	}
329 out:
330 	mutex_exit(&nodep->nm_lock);
331 	VOP_RWUNLOCK(nodep->nm_filevp, V_WRITELOCK_TRUE, ctp);
332 	return (error);
333 }
334 
335 /*
336  * Check mode permission on the namenode.  First nm_access_unlocked()
337  * checks the bits on the name node, then an access check is performed
338  * on the underlying file.
339  */
340 /* ARGSUSED */
341 static int
342 nm_access(vnode_t *vp, int mode, int flags, cred_t *crp,
343 	caller_context_t *ct)
344 {
345 	struct namenode *nodep = VTONM(vp);
346 	int error;
347 
348 	mutex_enter(&nodep->nm_lock);
349 	error = nm_access_unlocked(nodep, mode, crp);
350 	mutex_exit(&nodep->nm_lock);
351 	if (error == 0)
352 		return (VOP_ACCESS(nodep->nm_filevp, mode, flags, crp, ct));
353 	else
354 		return (error);
355 }
356 
357 /*
358  * We can get here if a creat or open with O_CREAT is done on a namefs
359  * mount point, for example, as the object of a shell output redirection to
360  * the mount point.
361  */
362 /*ARGSUSED*/
363 static int
364 nm_create(vnode_t *dvp, char *name, vattr_t *vap, enum vcexcl excl,
365 	int mode, vnode_t **vpp, cred_t *cr, int flag,
366 	caller_context_t *ct, vsecattr_t *vsecp)
367 {
368 	int error;
369 
370 	ASSERT(dvp && *name == '\0');
371 	if (excl == NONEXCL) {
372 		if (mode && (error = nm_access(dvp, mode, 0, cr, ct)) != 0)
373 			return (error);
374 		VN_HOLD(dvp);
375 		return (0);
376 	}
377 	return (EEXIST);
378 }
379 
380 /*
381  * Links are not allowed on mounted file descriptors.
382  */
383 /*ARGSUSED*/
384 static int
385 nm_link(vnode_t *tdvp, vnode_t *vp, char *tnm, cred_t *crp,
386 	caller_context_t *ct, int flags)
387 {
388 	return (EXDEV);
389 }
390 
391 static int
392 nm_fsync(vnode_t *vp, int syncflag, cred_t *crp, caller_context_t *ct)
393 {
394 	return (VOP_FSYNC(VTONM(vp)->nm_filevp, syncflag, crp, ct));
395 }
396 
397 /* Free the namenode */
398 /* ARGSUSED */
399 static void
400 nm_inactive(vnode_t *vp, cred_t *crp, caller_context_t *ct)
401 {
402 	struct namenode *nodep = VTONM(vp);
403 
404 	mutex_enter(&vp->v_lock);
405 	ASSERT(vp->v_count >= 1);
406 	if (--vp->v_count != 0) {
407 		mutex_exit(&vp->v_lock);
408 		return;
409 	}
410 	mutex_exit(&vp->v_lock);
411 	if (!(nodep->nm_flag & NMNMNT)) {
412 		ASSERT(nodep->nm_filep->f_vnode == nodep->nm_filevp);
413 		(void) closef(nodep->nm_filep);
414 	}
415 	vn_invalid(vp);
416 	vn_free(vp);
417 	namenodeno_free(nodep->nm_vattr.va_nodeid);
418 	kmem_free(nodep, sizeof (struct namenode));
419 }
420 
421 static int
422 nm_fid(vnode_t *vp, struct fid *fidnodep, caller_context_t *ct)
423 {
424 	return (VOP_FID(VTONM(vp)->nm_filevp, fidnodep, ct));
425 }
426 
427 static int
428 nm_rwlock(vnode_t *vp, int write, caller_context_t *ctp)
429 {
430 	return (VOP_RWLOCK(VTONM(vp)->nm_filevp, write, ctp));
431 }
432 
433 static void
434 nm_rwunlock(vnode_t *vp, int write, caller_context_t *ctp)
435 {
436 	VOP_RWUNLOCK(VTONM(vp)->nm_filevp, write, ctp);
437 }
438 
439 static int
440 nm_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
441 {
442 	return (VOP_SEEK(VTONM(vp)->nm_filevp, ooff, noffp, ct));
443 }
444 
445 /*
446  * Return the vnode representing the file descriptor in vpp.
447  */
448 static int
449 nm_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
450 {
451 	struct vnode *rvp;
452 
453 	vp = VTONM(vp)->nm_filevp;
454 	if (VOP_REALVP(vp, &rvp, ct) == 0)
455 		vp = rvp;
456 	*vpp = vp;
457 	return (0);
458 }
459 
460 static int
461 nm_poll(vnode_t *vp, short events, int anyyet, short *reventsp,
462 	pollhead_t **phpp, caller_context_t *ct)
463 {
464 	return (VOP_POLL(VTONM(vp)->nm_filevp, events, anyyet, reventsp,
465 	    phpp, ct));
466 }
467 
468 struct vnodeops *nm_vnodeops;
469 
470 const fs_operation_def_t nm_vnodeops_template[] = {
471 	VOPNAME_OPEN,		{ .vop_open = nm_open },
472 	VOPNAME_CLOSE,		{ .vop_close = nm_close },
473 	VOPNAME_READ,		{ .vop_read = nm_read },
474 	VOPNAME_WRITE,		{ .vop_write = nm_write },
475 	VOPNAME_IOCTL,		{ .vop_ioctl = nm_ioctl },
476 	VOPNAME_GETATTR,	{ .vop_getattr = nm_getattr },
477 	VOPNAME_SETATTR,	{ .vop_setattr = nm_setattr },
478 	VOPNAME_ACCESS,		{ .vop_access = nm_access },
479 	VOPNAME_CREATE,		{ .vop_create = nm_create },
480 	VOPNAME_LINK,		{ .vop_link = nm_link },
481 	VOPNAME_FSYNC,		{ .vop_fsync = nm_fsync },
482 	VOPNAME_INACTIVE,	{ .vop_inactive = nm_inactive },
483 	VOPNAME_FID,		{ .vop_fid = nm_fid },
484 	VOPNAME_RWLOCK,		{ .vop_rwlock = nm_rwlock },
485 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = nm_rwunlock },
486 	VOPNAME_SEEK,		{ .vop_seek = nm_seek },
487 	VOPNAME_REALVP,		{ .vop_realvp = nm_realvp },
488 	VOPNAME_POLL,		{ .vop_poll = nm_poll },
489 	VOPNAME_DISPOSE,	{ .error = fs_error },
490 	NULL,			NULL
491 };
492