xref: /titanic_41/usr/src/uts/common/fs/vnode.c (revision ea394cb00fd96864e34d2841b4a22357b621c78f)
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #include <sys/types.h>
40 #include <sys/param.h>
41 #include <sys/t_lock.h>
42 #include <sys/errno.h>
43 #include <sys/cred.h>
44 #include <sys/user.h>
45 #include <sys/uio.h>
46 #include <sys/file.h>
47 #include <sys/pathname.h>
48 #include <sys/vfs.h>
49 #include <sys/vfs_opreg.h>
50 #include <sys/vnode.h>
51 #include <sys/rwstlock.h>
52 #include <sys/fem.h>
53 #include <sys/stat.h>
54 #include <sys/mode.h>
55 #include <sys/conf.h>
56 #include <sys/sysmacros.h>
57 #include <sys/cmn_err.h>
58 #include <sys/systm.h>
59 #include <sys/kmem.h>
60 #include <sys/debug.h>
61 #include <c2/audit.h>
62 #include <sys/acl.h>
63 #include <sys/nbmlock.h>
64 #include <sys/fcntl.h>
65 #include <fs/fs_subr.h>
66 #include <sys/taskq.h>
67 #include <fs/fs_reparse.h>
68 
69 /* Determine if this vnode is a file that is read-only */
70 #define	ISROFILE(vp)	\
71 	((vp)->v_type != VCHR && (vp)->v_type != VBLK && \
72 	    (vp)->v_type != VFIFO && vn_is_readonly(vp))
73 
74 /* Tunable via /etc/system; used only by admin/install */
75 int nfs_global_client_only;
76 
77 /*
78  * Array of vopstats_t for per-FS-type vopstats.  This array has the same
79  * number of entries as and parallel to the vfssw table.  (Arguably, it could
80  * be part of the vfssw table.)  Once it's initialized, it's accessed using
81  * the same fstype index that is used to index into the vfssw table.
82  */
83 vopstats_t **vopstats_fstype;
84 
85 /* vopstats initialization template used for fast initialization via bcopy() */
86 static vopstats_t *vs_templatep;
87 
88 /* Kmem cache handle for vsk_anchor_t allocations */
89 kmem_cache_t *vsk_anchor_cache;
90 
91 /* file events cleanup routine */
92 extern void free_fopdata(vnode_t *);
93 
94 /*
95  * Root of AVL tree for the kstats associated with vopstats.  Lock protects
96  * updates to vsktat_tree.
97  */
98 avl_tree_t	vskstat_tree;
99 kmutex_t	vskstat_tree_lock;
100 
101 /* Global variable which enables/disables the vopstats collection */
102 int vopstats_enabled = 1;
103 
104 /*
105  * forward declarations for internal vnode specific data (vsd)
106  */
107 static void *vsd_realloc(void *, size_t, size_t);
108 
109 /*
110  * forward declarations for reparse point functions
111  */
112 static int fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr);
113 
114 /*
115  * VSD -- VNODE SPECIFIC DATA
116  * The v_data pointer is typically used by a file system to store a
117  * pointer to the file system's private node (e.g. ufs inode, nfs rnode).
118  * However, there are times when additional project private data needs
119  * to be stored separately from the data (node) pointed to by v_data.
120  * This additional data could be stored by the file system itself or
121  * by a completely different kernel entity.  VSD provides a way for
122  * callers to obtain a key and store a pointer to private data associated
123  * with a vnode.
124  *
125  * Callers are responsible for protecting the vsd by holding v_vsd_lock
126  * for calls to vsd_set() and vsd_get().
127  */
128 
129 /*
130  * vsd_lock protects:
131  *   vsd_nkeys - creation and deletion of vsd keys
132  *   vsd_list - insertion and deletion of vsd_node in the vsd_list
133  *   vsd_destructor - adding and removing destructors to the list
134  */
135 static kmutex_t		vsd_lock;
136 static uint_t		vsd_nkeys;	 /* size of destructor array */
137 /* list of vsd_node's */
138 static list_t *vsd_list = NULL;
139 /* per-key destructor funcs */
140 static void 		(**vsd_destructor)(void *);
141 
142 /*
143  * The following is the common set of actions needed to update the
144  * vopstats structure from a vnode op.  Both VOPSTATS_UPDATE() and
145  * VOPSTATS_UPDATE_IO() do almost the same thing, except for the
146  * recording of the bytes transferred.  Since the code is similar
147  * but small, it is nearly a duplicate.  Consequently any changes
148  * to one may need to be reflected in the other.
149  * Rundown of the variables:
150  * vp - Pointer to the vnode
151  * counter - Partial name structure member to update in vopstats for counts
152  * bytecounter - Partial name structure member to update in vopstats for bytes
153  * bytesval - Value to update in vopstats for bytes
154  * fstype - Index into vsanchor_fstype[], same as index into vfssw[]
155  * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i])
156  */
157 
158 #define	VOPSTATS_UPDATE(vp, counter) {					\
159 	vfs_t *vfsp = (vp)->v_vfsp;					\
160 	if (vfsp && vfsp->vfs_implp &&					\
161 	    (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {	\
162 		vopstats_t *vsp = &vfsp->vfs_vopstats;			\
163 		uint64_t *stataddr = &(vsp->n##counter.value.ui64);	\
164 		extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
165 		    size_t, uint64_t *);				\
166 		__dtrace_probe___fsinfo_##counter(vp, 0, stataddr);	\
167 		(*stataddr)++;						\
168 		if ((vsp = vfsp->vfs_fstypevsp) != NULL) {		\
169 			vsp->n##counter.value.ui64++;			\
170 		}							\
171 	}								\
172 }
173 
174 #define	VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) {	\
175 	vfs_t *vfsp = (vp)->v_vfsp;					\
176 	if (vfsp && vfsp->vfs_implp &&					\
177 	    (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {	\
178 		vopstats_t *vsp = &vfsp->vfs_vopstats;			\
179 		uint64_t *stataddr = &(vsp->n##counter.value.ui64);	\
180 		extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
181 		    size_t, uint64_t *);				\
182 		__dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \
183 		(*stataddr)++;						\
184 		vsp->bytecounter.value.ui64 += bytesval;		\
185 		if ((vsp = vfsp->vfs_fstypevsp) != NULL) {		\
186 			vsp->n##counter.value.ui64++;			\
187 			vsp->bytecounter.value.ui64 += bytesval;	\
188 		}							\
189 	}								\
190 }
191 
192 /*
193  * If the filesystem does not support XIDs map credential
194  * If the vfsp is NULL, perhaps we should also map?
195  */
196 #define	VOPXID_MAP_CR(vp, cr)	{					\
197 	vfs_t *vfsp = (vp)->v_vfsp;					\
198 	if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0)		\
199 		cr = crgetmapped(cr);					\
200 	}
201 
202 /*
203  * Convert stat(2) formats to vnode types and vice versa.  (Knows about
204  * numerical order of S_IFMT and vnode types.)
205  */
206 enum vtype iftovt_tab[] = {
207 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
208 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
209 };
210 
211 ushort_t vttoif_tab[] = {
212 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO,
213 	S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0
214 };
215 
216 /*
217  * The system vnode cache.
218  */
219 
220 kmem_cache_t *vn_cache;
221 
222 
223 /*
224  * Vnode operations vector.
225  */
226 
227 static const fs_operation_trans_def_t vn_ops_table[] = {
228 	VOPNAME_OPEN, offsetof(struct vnodeops, vop_open),
229 	    fs_nosys, fs_nosys,
230 
231 	VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close),
232 	    fs_nosys, fs_nosys,
233 
234 	VOPNAME_READ, offsetof(struct vnodeops, vop_read),
235 	    fs_nosys, fs_nosys,
236 
237 	VOPNAME_WRITE, offsetof(struct vnodeops, vop_write),
238 	    fs_nosys, fs_nosys,
239 
240 	VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl),
241 	    fs_nosys, fs_nosys,
242 
243 	VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl),
244 	    fs_setfl, fs_nosys,
245 
246 	VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr),
247 	    fs_nosys, fs_nosys,
248 
249 	VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr),
250 	    fs_nosys, fs_nosys,
251 
252 	VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access),
253 	    fs_nosys, fs_nosys,
254 
255 	VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup),
256 	    fs_nosys, fs_nosys,
257 
258 	VOPNAME_CREATE, offsetof(struct vnodeops, vop_create),
259 	    fs_nosys, fs_nosys,
260 
261 	VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove),
262 	    fs_nosys, fs_nosys,
263 
264 	VOPNAME_LINK, offsetof(struct vnodeops, vop_link),
265 	    fs_nosys, fs_nosys,
266 
267 	VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename),
268 	    fs_nosys, fs_nosys,
269 
270 	VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir),
271 	    fs_nosys, fs_nosys,
272 
273 	VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir),
274 	    fs_nosys, fs_nosys,
275 
276 	VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir),
277 	    fs_nosys, fs_nosys,
278 
279 	VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink),
280 	    fs_nosys, fs_nosys,
281 
282 	VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink),
283 	    fs_nosys, fs_nosys,
284 
285 	VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync),
286 	    fs_nosys, fs_nosys,
287 
288 	VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive),
289 	    fs_nosys, fs_nosys,
290 
291 	VOPNAME_FID, offsetof(struct vnodeops, vop_fid),
292 	    fs_nosys, fs_nosys,
293 
294 	VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock),
295 	    fs_rwlock, fs_rwlock,
296 
297 	VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock),
298 	    (fs_generic_func_p) fs_rwunlock,
299 	    (fs_generic_func_p) fs_rwunlock,	/* no errors allowed */
300 
301 	VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek),
302 	    fs_nosys, fs_nosys,
303 
304 	VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp),
305 	    fs_cmp, fs_cmp,		/* no errors allowed */
306 
307 	VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock),
308 	    fs_frlock, fs_nosys,
309 
310 	VOPNAME_SPACE, offsetof(struct vnodeops, vop_space),
311 	    fs_nosys, fs_nosys,
312 
313 	VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp),
314 	    fs_nosys, fs_nosys,
315 
316 	VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage),
317 	    fs_nosys, fs_nosys,
318 
319 	VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage),
320 	    fs_nosys, fs_nosys,
321 
322 	VOPNAME_MAP, offsetof(struct vnodeops, vop_map),
323 	    (fs_generic_func_p) fs_nosys_map,
324 	    (fs_generic_func_p) fs_nosys_map,
325 
326 	VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap),
327 	    (fs_generic_func_p) fs_nosys_addmap,
328 	    (fs_generic_func_p) fs_nosys_addmap,
329 
330 	VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap),
331 	    fs_nosys, fs_nosys,
332 
333 	VOPNAME_POLL, offsetof(struct vnodeops, vop_poll),
334 	    (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll,
335 
336 	VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump),
337 	    fs_nosys, fs_nosys,
338 
339 	VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf),
340 	    fs_pathconf, fs_nosys,
341 
342 	VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio),
343 	    fs_nosys, fs_nosys,
344 
345 	VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl),
346 	    fs_nosys, fs_nosys,
347 
348 	VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose),
349 	    (fs_generic_func_p) fs_dispose,
350 	    (fs_generic_func_p) fs_nodispose,
351 
352 	VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr),
353 	    fs_nosys, fs_nosys,
354 
355 	VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr),
356 	    fs_fab_acl, fs_nosys,
357 
358 	VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock),
359 	    fs_shrlock, fs_nosys,
360 
361 	VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent),
362 	    (fs_generic_func_p) fs_vnevent_nosupport,
363 	    (fs_generic_func_p) fs_vnevent_nosupport,
364 
365 	VOPNAME_REQZCBUF, offsetof(struct vnodeops, vop_reqzcbuf),
366 	    fs_nosys, fs_nosys,
367 
368 	VOPNAME_RETZCBUF, offsetof(struct vnodeops, vop_retzcbuf),
369 	    fs_nosys, fs_nosys,
370 
371 	NULL, 0, NULL, NULL
372 };
373 
374 /* Extensible attribute (xva) routines. */
375 
376 /*
377  * Zero out the structure, set the size of the requested/returned bitmaps,
378  * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer
379  * to the returned attributes array.
380  */
381 void
382 xva_init(xvattr_t *xvap)
383 {
384 	bzero(xvap, sizeof (xvattr_t));
385 	xvap->xva_mapsize = XVA_MAPSIZE;
386 	xvap->xva_magic = XVA_MAGIC;
387 	xvap->xva_vattr.va_mask = AT_XVATTR;
388 	xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0];
389 }
390 
391 /*
392  * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t
393  * structure.  Otherwise, returns NULL.
394  */
395 xoptattr_t *
396 xva_getxoptattr(xvattr_t *xvap)
397 {
398 	xoptattr_t *xoap = NULL;
399 	if (xvap->xva_vattr.va_mask & AT_XVATTR)
400 		xoap = &xvap->xva_xoptattrs;
401 	return (xoap);
402 }
403 
404 /*
405  * Used by the AVL routines to compare two vsk_anchor_t structures in the tree.
406  * We use the f_fsid reported by VFS_STATVFS() since we use that for the
407  * kstat name.
408  */
409 static int
410 vska_compar(const void *n1, const void *n2)
411 {
412 	int ret;
413 	ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid;
414 	ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid;
415 
416 	if (p1 < p2) {
417 		ret = -1;
418 	} else if (p1 > p2) {
419 		ret = 1;
420 	} else {
421 		ret = 0;
422 	}
423 
424 	return (ret);
425 }
426 
427 /*
428  * Used to create a single template which will be bcopy()ed to a newly
429  * allocated vsanchor_combo_t structure in new_vsanchor(), below.
430  */
431 static vopstats_t *
432 create_vopstats_template()
433 {
434 	vopstats_t		*vsp;
435 
436 	vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP);
437 	bzero(vsp, sizeof (*vsp));	/* Start fresh */
438 
439 	/* VOP_OPEN */
440 	kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64);
441 	/* VOP_CLOSE */
442 	kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64);
443 	/* VOP_READ I/O */
444 	kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64);
445 	kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64);
446 	/* VOP_WRITE I/O */
447 	kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64);
448 	kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64);
449 	/* VOP_IOCTL */
450 	kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64);
451 	/* VOP_SETFL */
452 	kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64);
453 	/* VOP_GETATTR */
454 	kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64);
455 	/* VOP_SETATTR */
456 	kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64);
457 	/* VOP_ACCESS */
458 	kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64);
459 	/* VOP_LOOKUP */
460 	kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64);
461 	/* VOP_CREATE */
462 	kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64);
463 	/* VOP_REMOVE */
464 	kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64);
465 	/* VOP_LINK */
466 	kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64);
467 	/* VOP_RENAME */
468 	kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64);
469 	/* VOP_MKDIR */
470 	kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64);
471 	/* VOP_RMDIR */
472 	kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64);
473 	/* VOP_READDIR I/O */
474 	kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64);
475 	kstat_named_init(&vsp->readdir_bytes, "readdir_bytes",
476 	    KSTAT_DATA_UINT64);
477 	/* VOP_SYMLINK */
478 	kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64);
479 	/* VOP_READLINK */
480 	kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64);
481 	/* VOP_FSYNC */
482 	kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64);
483 	/* VOP_INACTIVE */
484 	kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64);
485 	/* VOP_FID */
486 	kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64);
487 	/* VOP_RWLOCK */
488 	kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64);
489 	/* VOP_RWUNLOCK */
490 	kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64);
491 	/* VOP_SEEK */
492 	kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64);
493 	/* VOP_CMP */
494 	kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64);
495 	/* VOP_FRLOCK */
496 	kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64);
497 	/* VOP_SPACE */
498 	kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64);
499 	/* VOP_REALVP */
500 	kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64);
501 	/* VOP_GETPAGE */
502 	kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64);
503 	/* VOP_PUTPAGE */
504 	kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64);
505 	/* VOP_MAP */
506 	kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64);
507 	/* VOP_ADDMAP */
508 	kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64);
509 	/* VOP_DELMAP */
510 	kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64);
511 	/* VOP_POLL */
512 	kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64);
513 	/* VOP_DUMP */
514 	kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64);
515 	/* VOP_PATHCONF */
516 	kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64);
517 	/* VOP_PAGEIO */
518 	kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64);
519 	/* VOP_DUMPCTL */
520 	kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64);
521 	/* VOP_DISPOSE */
522 	kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64);
523 	/* VOP_SETSECATTR */
524 	kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64);
525 	/* VOP_GETSECATTR */
526 	kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64);
527 	/* VOP_SHRLOCK */
528 	kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64);
529 	/* VOP_VNEVENT */
530 	kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64);
531 	/* VOP_REQZCBUF */
532 	kstat_named_init(&vsp->nreqzcbuf, "nreqzcbuf", KSTAT_DATA_UINT64);
533 	/* VOP_RETZCBUF */
534 	kstat_named_init(&vsp->nretzcbuf, "nretzcbuf", KSTAT_DATA_UINT64);
535 
536 	return (vsp);
537 }
538 
539 /*
540  * Creates a kstat structure associated with a vopstats structure.
541  */
542 kstat_t *
543 new_vskstat(char *ksname, vopstats_t *vsp)
544 {
545 	kstat_t		*ksp;
546 
547 	if (!vopstats_enabled) {
548 		return (NULL);
549 	}
550 
551 	ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED,
552 	    sizeof (vopstats_t)/sizeof (kstat_named_t),
553 	    KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE);
554 	if (ksp) {
555 		ksp->ks_data = vsp;
556 		kstat_install(ksp);
557 	}
558 
559 	return (ksp);
560 }
561 
562 /*
563  * Called from vfsinit() to initialize the support mechanisms for vopstats
564  */
565 void
566 vopstats_startup()
567 {
568 	if (!vopstats_enabled)
569 		return;
570 
571 	/*
572 	 * Creates the AVL tree which holds per-vfs vopstat anchors.  This
573 	 * is necessary since we need to check if a kstat exists before we
574 	 * attempt to create it.  Also, initialize its lock.
575 	 */
576 	avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t),
577 	    offsetof(vsk_anchor_t, vsk_node));
578 	mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL);
579 
580 	vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache",
581 	    sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL,
582 	    NULL, NULL, 0);
583 
584 	/*
585 	 * Set up the array of pointers for the vopstats-by-FS-type.
586 	 * The entries will be allocated/initialized as each file system
587 	 * goes through modload/mod_installfs.
588 	 */
589 	vopstats_fstype = (vopstats_t **)kmem_zalloc(
590 	    (sizeof (vopstats_t *) * nfstype), KM_SLEEP);
591 
592 	/* Set up the global vopstats initialization template */
593 	vs_templatep = create_vopstats_template();
594 }
595 
596 /*
597  * We need to have the all of the counters zeroed.
598  * The initialization of the vopstats_t includes on the order of
599  * 50 calls to kstat_named_init().  Rather that do that on every call,
600  * we do it once in a template (vs_templatep) then bcopy it over.
601  */
602 void
603 initialize_vopstats(vopstats_t *vsp)
604 {
605 	if (vsp == NULL)
606 		return;
607 
608 	bcopy(vs_templatep, vsp, sizeof (vopstats_t));
609 }
610 
611 /*
612  * If possible, determine which vopstats by fstype to use and
613  * return a pointer to the caller.
614  */
615 vopstats_t *
616 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp)
617 {
618 	int		fstype = 0;	/* Index into vfssw[] */
619 	vopstats_t	*vsp = NULL;
620 
621 	if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 ||
622 	    !vopstats_enabled)
623 		return (NULL);
624 	/*
625 	 * Set up the fstype.  We go to so much trouble because all versions
626 	 * of NFS use the same fstype in their vfs even though they have
627 	 * distinct entries in the vfssw[] table.
628 	 * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry.
629 	 */
630 	if (vswp) {
631 		fstype = vswp - vfssw;	/* Gets us the index */
632 	} else {
633 		fstype = vfsp->vfs_fstype;
634 	}
635 
636 	/*
637 	 * Point to the per-fstype vopstats. The only valid values are
638 	 * non-zero positive values less than the number of vfssw[] table
639 	 * entries.
640 	 */
641 	if (fstype > 0 && fstype < nfstype) {
642 		vsp = vopstats_fstype[fstype];
643 	}
644 
645 	return (vsp);
646 }
647 
648 /*
649  * Generate a kstat name, create the kstat structure, and allocate a
650  * vsk_anchor_t to hold it together.  Return the pointer to the vsk_anchor_t
651  * to the caller.  This must only be called from a mount.
652  */
653 vsk_anchor_t *
654 get_vskstat_anchor(vfs_t *vfsp)
655 {
656 	char		kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */
657 	statvfs64_t	statvfsbuf;		/* Needed to find f_fsid */
658 	vsk_anchor_t	*vskp = NULL;		/* vfs <--> kstat anchor */
659 	kstat_t		*ksp;			/* Ptr to new kstat */
660 	avl_index_t	where;			/* Location in the AVL tree */
661 
662 	if (vfsp == NULL || vfsp->vfs_implp == NULL ||
663 	    (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
664 		return (NULL);
665 
666 	/* Need to get the fsid to build a kstat name */
667 	if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) {
668 		/* Create a name for our kstats based on fsid */
669 		(void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx",
670 		    VOPSTATS_STR, statvfsbuf.f_fsid);
671 
672 		/* Allocate and initialize the vsk_anchor_t */
673 		vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP);
674 		bzero(vskp, sizeof (*vskp));
675 		vskp->vsk_fsid = statvfsbuf.f_fsid;
676 
677 		mutex_enter(&vskstat_tree_lock);
678 		if (avl_find(&vskstat_tree, vskp, &where) == NULL) {
679 			avl_insert(&vskstat_tree, vskp, where);
680 			mutex_exit(&vskstat_tree_lock);
681 
682 			/*
683 			 * Now that we've got the anchor in the AVL
684 			 * tree, we can create the kstat.
685 			 */
686 			ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats);
687 			if (ksp) {
688 				vskp->vsk_ksp = ksp;
689 			}
690 		} else {
691 			/* Oops, found one! Release memory and lock. */
692 			mutex_exit(&vskstat_tree_lock);
693 			kmem_cache_free(vsk_anchor_cache, vskp);
694 			vskp = NULL;
695 		}
696 	}
697 	return (vskp);
698 }
699 
700 /*
701  * We're in the process of tearing down the vfs and need to cleanup
702  * the data structures associated with the vopstats. Must only be called
703  * from dounmount().
704  */
705 void
706 teardown_vopstats(vfs_t *vfsp)
707 {
708 	vsk_anchor_t	*vskap;
709 	avl_index_t	where;
710 
711 	if (vfsp == NULL || vfsp->vfs_implp == NULL ||
712 	    (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
713 		return;
714 
715 	/* This is a safe check since VFS_STATS must be set (see above) */
716 	if ((vskap = vfsp->vfs_vskap) == NULL)
717 		return;
718 
719 	/* Whack the pointer right away */
720 	vfsp->vfs_vskap = NULL;
721 
722 	/* Lock the tree, remove the node, and delete the kstat */
723 	mutex_enter(&vskstat_tree_lock);
724 	if (avl_find(&vskstat_tree, vskap, &where)) {
725 		avl_remove(&vskstat_tree, vskap);
726 	}
727 
728 	if (vskap->vsk_ksp) {
729 		kstat_delete(vskap->vsk_ksp);
730 	}
731 	mutex_exit(&vskstat_tree_lock);
732 
733 	kmem_cache_free(vsk_anchor_cache, vskap);
734 }
735 
736 /*
737  * Read or write a vnode.  Called from kernel code.
738  */
739 int
740 vn_rdwr(
741 	enum uio_rw rw,
742 	struct vnode *vp,
743 	caddr_t base,
744 	ssize_t len,
745 	offset_t offset,
746 	enum uio_seg seg,
747 	int ioflag,
748 	rlim64_t ulimit,	/* meaningful only if rw is UIO_WRITE */
749 	cred_t *cr,
750 	ssize_t *residp)
751 {
752 	struct uio uio;
753 	struct iovec iov;
754 	int error;
755 	int in_crit = 0;
756 
757 	if (rw == UIO_WRITE && ISROFILE(vp))
758 		return (EROFS);
759 
760 	if (len < 0)
761 		return (EIO);
762 
763 	VOPXID_MAP_CR(vp, cr);
764 
765 	iov.iov_base = base;
766 	iov.iov_len = len;
767 	uio.uio_iov = &iov;
768 	uio.uio_iovcnt = 1;
769 	uio.uio_loffset = offset;
770 	uio.uio_segflg = (short)seg;
771 	uio.uio_resid = len;
772 	uio.uio_llimit = ulimit;
773 
774 	/*
775 	 * We have to enter the critical region before calling VOP_RWLOCK
776 	 * to avoid a deadlock with ufs.
777 	 */
778 	if (nbl_need_check(vp)) {
779 		int svmand;
780 
781 		nbl_start_crit(vp, RW_READER);
782 		in_crit = 1;
783 		error = nbl_svmand(vp, cr, &svmand);
784 		if (error != 0)
785 			goto done;
786 		if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ,
787 		    uio.uio_offset, uio.uio_resid, svmand, NULL)) {
788 			error = EACCES;
789 			goto done;
790 		}
791 	}
792 
793 	(void) VOP_RWLOCK(vp,
794 	    rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
795 	if (rw == UIO_WRITE) {
796 		uio.uio_fmode = FWRITE;
797 		uio.uio_extflg = UIO_COPY_DEFAULT;
798 		error = VOP_WRITE(vp, &uio, ioflag, cr, NULL);
799 	} else {
800 		uio.uio_fmode = FREAD;
801 		uio.uio_extflg = UIO_COPY_CACHED;
802 		error = VOP_READ(vp, &uio, ioflag, cr, NULL);
803 	}
804 	VOP_RWUNLOCK(vp,
805 	    rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
806 	if (residp)
807 		*residp = uio.uio_resid;
808 	else if (uio.uio_resid)
809 		error = EIO;
810 
811 done:
812 	if (in_crit)
813 		nbl_end_crit(vp);
814 	return (error);
815 }
816 
817 /*
818  * Release a vnode.  Call VOP_INACTIVE on last reference or
819  * decrement reference count.
820  *
821  * To avoid race conditions, the v_count is left at 1 for
822  * the call to VOP_INACTIVE. This prevents another thread
823  * from reclaiming and releasing the vnode *before* the
824  * VOP_INACTIVE routine has a chance to destroy the vnode.
825  * We can't have more than 1 thread calling VOP_INACTIVE
826  * on a vnode.
827  */
828 void
829 vn_rele(vnode_t *vp)
830 {
831 	VERIFY(vp->v_count > 0);
832 	mutex_enter(&vp->v_lock);
833 	if (vp->v_count == 1) {
834 		mutex_exit(&vp->v_lock);
835 		VOP_INACTIVE(vp, CRED(), NULL);
836 		return;
837 	}
838 	vp->v_count--;
839 	mutex_exit(&vp->v_lock);
840 }
841 
842 /*
843  * Release a vnode referenced by the DNLC. Multiple DNLC references are treated
844  * as a single reference, so v_count is not decremented until the last DNLC hold
845  * is released. This makes it possible to distinguish vnodes that are referenced
846  * only by the DNLC.
847  */
848 void
849 vn_rele_dnlc(vnode_t *vp)
850 {
851 	VERIFY((vp->v_count > 0) && (vp->v_count_dnlc > 0));
852 	mutex_enter(&vp->v_lock);
853 	if (--vp->v_count_dnlc == 0) {
854 		if (vp->v_count == 1) {
855 			mutex_exit(&vp->v_lock);
856 			VOP_INACTIVE(vp, CRED(), NULL);
857 			return;
858 		}
859 		vp->v_count--;
860 	}
861 	mutex_exit(&vp->v_lock);
862 }
863 
864 /*
865  * Like vn_rele() except that it clears v_stream under v_lock.
866  * This is used by sockfs when it dismantels the association between
867  * the sockfs node and the vnode in the underlaying file system.
868  * v_lock has to be held to prevent a thread coming through the lookupname
869  * path from accessing a stream head that is going away.
870  */
871 void
872 vn_rele_stream(vnode_t *vp)
873 {
874 	VERIFY(vp->v_count > 0);
875 	mutex_enter(&vp->v_lock);
876 	vp->v_stream = NULL;
877 	if (vp->v_count == 1) {
878 		mutex_exit(&vp->v_lock);
879 		VOP_INACTIVE(vp, CRED(), NULL);
880 		return;
881 	}
882 	vp->v_count--;
883 	mutex_exit(&vp->v_lock);
884 }
885 
886 static void
887 vn_rele_inactive(vnode_t *vp)
888 {
889 	VOP_INACTIVE(vp, CRED(), NULL);
890 }
891 
892 /*
893  * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it
894  * asynchronously using a taskq. This can avoid deadlocks caused by re-entering
895  * the file system as a result of releasing the vnode. Note, file systems
896  * already have to handle the race where the vnode is incremented before the
897  * inactive routine is called and does its locking.
898  *
899  * Warning: Excessive use of this routine can lead to performance problems.
900  * This is because taskqs throttle back allocation if too many are created.
901  */
902 void
903 vn_rele_async(vnode_t *vp, taskq_t *taskq)
904 {
905 	VERIFY(vp->v_count > 0);
906 	mutex_enter(&vp->v_lock);
907 	if (vp->v_count == 1) {
908 		mutex_exit(&vp->v_lock);
909 		VERIFY(taskq_dispatch(taskq, (task_func_t *)vn_rele_inactive,
910 		    vp, TQ_SLEEP) != NULL);
911 		return;
912 	}
913 	vp->v_count--;
914 	mutex_exit(&vp->v_lock);
915 }
916 
917 int
918 vn_open(
919 	char *pnamep,
920 	enum uio_seg seg,
921 	int filemode,
922 	int createmode,
923 	struct vnode **vpp,
924 	enum create crwhy,
925 	mode_t umask)
926 {
927 	return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy,
928 	    umask, NULL, -1));
929 }
930 
931 
932 /*
933  * Open/create a vnode.
934  * This may be callable by the kernel, the only known use
935  * of user context being that the current user credentials
936  * are used for permissions.  crwhy is defined iff filemode & FCREAT.
937  */
938 int
939 vn_openat(
940 	char *pnamep,
941 	enum uio_seg seg,
942 	int filemode,
943 	int createmode,
944 	struct vnode **vpp,
945 	enum create crwhy,
946 	mode_t umask,
947 	struct vnode *startvp,
948 	int fd)
949 {
950 	struct vnode *vp;
951 	int mode;
952 	int accessflags;
953 	int error;
954 	int in_crit = 0;
955 	int open_done = 0;
956 	int shrlock_done = 0;
957 	struct vattr vattr;
958 	enum symfollow follow;
959 	int estale_retry = 0;
960 	struct shrlock shr;
961 	struct shr_locowner shr_own;
962 
963 	mode = 0;
964 	accessflags = 0;
965 	if (filemode & FREAD)
966 		mode |= VREAD;
967 	if (filemode & (FWRITE|FTRUNC))
968 		mode |= VWRITE;
969 	if (filemode & FXATTRDIROPEN)
970 		mode |= VEXEC;
971 
972 	/* symlink interpretation */
973 	if (filemode & FNOFOLLOW)
974 		follow = NO_FOLLOW;
975 	else
976 		follow = FOLLOW;
977 
978 	if (filemode & FAPPEND)
979 		accessflags |= V_APPEND;
980 
981 top:
982 	if (filemode & FCREAT) {
983 		enum vcexcl excl;
984 
985 		/*
986 		 * Wish to create a file.
987 		 */
988 		vattr.va_type = VREG;
989 		vattr.va_mode = createmode;
990 		vattr.va_mask = AT_TYPE|AT_MODE;
991 		if (filemode & FTRUNC) {
992 			vattr.va_size = 0;
993 			vattr.va_mask |= AT_SIZE;
994 		}
995 		if (filemode & FEXCL)
996 			excl = EXCL;
997 		else
998 			excl = NONEXCL;
999 
1000 		if (error =
1001 		    vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy,
1002 		    (filemode & ~(FTRUNC|FEXCL)), umask, startvp))
1003 			return (error);
1004 	} else {
1005 		/*
1006 		 * Wish to open a file.  Just look it up.
1007 		 */
1008 		if (error = lookupnameat(pnamep, seg, follow,
1009 		    NULLVPP, &vp, startvp)) {
1010 			if ((error == ESTALE) &&
1011 			    fs_need_estale_retry(estale_retry++))
1012 				goto top;
1013 			return (error);
1014 		}
1015 
1016 		/*
1017 		 * Get the attributes to check whether file is large.
1018 		 * We do this only if the FOFFMAX flag is not set and
1019 		 * only for regular files.
1020 		 */
1021 
1022 		if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) {
1023 			vattr.va_mask = AT_SIZE;
1024 			if ((error = VOP_GETATTR(vp, &vattr, 0,
1025 			    CRED(), NULL))) {
1026 				goto out;
1027 			}
1028 			if (vattr.va_size > (u_offset_t)MAXOFF32_T) {
1029 				/*
1030 				 * Large File API - regular open fails
1031 				 * if FOFFMAX flag is set in file mode
1032 				 */
1033 				error = EOVERFLOW;
1034 				goto out;
1035 			}
1036 		}
1037 		/*
1038 		 * Can't write directories, active texts, or
1039 		 * read-only filesystems.  Can't truncate files
1040 		 * on which mandatory locking is in effect.
1041 		 */
1042 		if (filemode & (FWRITE|FTRUNC)) {
1043 			/*
1044 			 * Allow writable directory if VDIROPEN flag is set.
1045 			 */
1046 			if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) {
1047 				error = EISDIR;
1048 				goto out;
1049 			}
1050 			if (ISROFILE(vp)) {
1051 				error = EROFS;
1052 				goto out;
1053 			}
1054 			/*
1055 			 * Can't truncate files on which
1056 			 * sysv mandatory locking is in effect.
1057 			 */
1058 			if (filemode & FTRUNC) {
1059 				vnode_t *rvp;
1060 
1061 				if (VOP_REALVP(vp, &rvp, NULL) != 0)
1062 					rvp = vp;
1063 				if (rvp->v_filocks != NULL) {
1064 					vattr.va_mask = AT_MODE;
1065 					if ((error = VOP_GETATTR(vp,
1066 					    &vattr, 0, CRED(), NULL)) == 0 &&
1067 					    MANDLOCK(vp, vattr.va_mode))
1068 						error = EAGAIN;
1069 				}
1070 			}
1071 			if (error)
1072 				goto out;
1073 		}
1074 		/*
1075 		 * Check permissions.
1076 		 */
1077 		if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL))
1078 			goto out;
1079 	}
1080 
1081 	/*
1082 	 * Do remaining checks for FNOFOLLOW and FNOLINKS.
1083 	 */
1084 	if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) {
1085 		error = ELOOP;
1086 		goto out;
1087 	}
1088 	if (filemode & FNOLINKS) {
1089 		vattr.va_mask = AT_NLINK;
1090 		if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) {
1091 			goto out;
1092 		}
1093 		if (vattr.va_nlink != 1) {
1094 			error = EMLINK;
1095 			goto out;
1096 		}
1097 	}
1098 
1099 	/*
1100 	 * Opening a socket corresponding to the AF_UNIX pathname
1101 	 * in the filesystem name space is not supported.
1102 	 * However, VSOCK nodes in namefs are supported in order
1103 	 * to make fattach work for sockets.
1104 	 *
1105 	 * XXX This uses VOP_REALVP to distinguish between
1106 	 * an unopened namefs node (where VOP_REALVP returns a
1107 	 * different VSOCK vnode) and a VSOCK created by vn_create
1108 	 * in some file system (where VOP_REALVP would never return
1109 	 * a different vnode).
1110 	 */
1111 	if (vp->v_type == VSOCK) {
1112 		struct vnode *nvp;
1113 
1114 		error = VOP_REALVP(vp, &nvp, NULL);
1115 		if (error != 0 || nvp == NULL || nvp == vp ||
1116 		    nvp->v_type != VSOCK) {
1117 			error = EOPNOTSUPP;
1118 			goto out;
1119 		}
1120 	}
1121 
1122 	if ((vp->v_type == VREG) && nbl_need_check(vp)) {
1123 		/* get share reservation */
1124 		shr.s_access = 0;
1125 		if (filemode & FWRITE)
1126 			shr.s_access |= F_WRACC;
1127 		if (filemode & FREAD)
1128 			shr.s_access |= F_RDACC;
1129 		shr.s_deny = 0;
1130 		shr.s_sysid = 0;
1131 		shr.s_pid = ttoproc(curthread)->p_pid;
1132 		shr_own.sl_pid = shr.s_pid;
1133 		shr_own.sl_id = fd;
1134 		shr.s_own_len = sizeof (shr_own);
1135 		shr.s_owner = (caddr_t)&shr_own;
1136 		error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(),
1137 		    NULL);
1138 		if (error)
1139 			goto out;
1140 		shrlock_done = 1;
1141 
1142 		/* nbmand conflict check if truncating file */
1143 		if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1144 			nbl_start_crit(vp, RW_READER);
1145 			in_crit = 1;
1146 
1147 			vattr.va_mask = AT_SIZE;
1148 			if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
1149 				goto out;
1150 			if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0,
1151 			    NULL)) {
1152 				error = EACCES;
1153 				goto out;
1154 			}
1155 		}
1156 	}
1157 
1158 	/*
1159 	 * Do opening protocol.
1160 	 */
1161 	error = VOP_OPEN(&vp, filemode, CRED(), NULL);
1162 	if (error)
1163 		goto out;
1164 	open_done = 1;
1165 
1166 	/*
1167 	 * Truncate if required.
1168 	 */
1169 	if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1170 		vattr.va_size = 0;
1171 		vattr.va_mask = AT_SIZE;
1172 		if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
1173 			goto out;
1174 	}
1175 out:
1176 	ASSERT(vp->v_count > 0);
1177 
1178 	if (in_crit) {
1179 		nbl_end_crit(vp);
1180 		in_crit = 0;
1181 	}
1182 	if (error) {
1183 		if (open_done) {
1184 			(void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(),
1185 			    NULL);
1186 			open_done = 0;
1187 			shrlock_done = 0;
1188 		}
1189 		if (shrlock_done) {
1190 			(void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(),
1191 			    NULL);
1192 			shrlock_done = 0;
1193 		}
1194 
1195 		/*
1196 		 * The following clause was added to handle a problem
1197 		 * with NFS consistency.  It is possible that a lookup
1198 		 * of the file to be opened succeeded, but the file
1199 		 * itself doesn't actually exist on the server.  This
1200 		 * is chiefly due to the DNLC containing an entry for
1201 		 * the file which has been removed on the server.  In
1202 		 * this case, we just start over.  If there was some
1203 		 * other cause for the ESTALE error, then the lookup
1204 		 * of the file will fail and the error will be returned
1205 		 * above instead of looping around from here.
1206 		 */
1207 		VN_RELE(vp);
1208 		if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1209 			goto top;
1210 	} else
1211 		*vpp = vp;
1212 	return (error);
1213 }
1214 
1215 /*
1216  * The following two accessor functions are for the NFSv4 server.  Since there
1217  * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the
1218  * vnode open counts correct when a client "upgrades" an open or does an
1219  * open_downgrade.  In NFS, an upgrade or downgrade can not only change the
1220  * open mode (add or subtract read or write), but also change the share/deny
1221  * modes.  However, share reservations are not integrated with OPEN, yet, so
1222  * we need to handle each separately.  These functions are cleaner than having
1223  * the NFS server manipulate the counts directly, however, nobody else should
1224  * use these functions.
1225  */
1226 void
1227 vn_open_upgrade(
1228 	vnode_t *vp,
1229 	int filemode)
1230 {
1231 	ASSERT(vp->v_type == VREG);
1232 
1233 	if (filemode & FREAD)
1234 		atomic_add_32(&(vp->v_rdcnt), 1);
1235 	if (filemode & FWRITE)
1236 		atomic_add_32(&(vp->v_wrcnt), 1);
1237 
1238 }
1239 
1240 void
1241 vn_open_downgrade(
1242 	vnode_t *vp,
1243 	int filemode)
1244 {
1245 	ASSERT(vp->v_type == VREG);
1246 
1247 	if (filemode & FREAD) {
1248 		ASSERT(vp->v_rdcnt > 0);
1249 		atomic_add_32(&(vp->v_rdcnt), -1);
1250 	}
1251 	if (filemode & FWRITE) {
1252 		ASSERT(vp->v_wrcnt > 0);
1253 		atomic_add_32(&(vp->v_wrcnt), -1);
1254 	}
1255 
1256 }
1257 
1258 int
1259 vn_create(
1260 	char *pnamep,
1261 	enum uio_seg seg,
1262 	struct vattr *vap,
1263 	enum vcexcl excl,
1264 	int mode,
1265 	struct vnode **vpp,
1266 	enum create why,
1267 	int flag,
1268 	mode_t umask)
1269 {
1270 	return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag,
1271 	    umask, NULL));
1272 }
1273 
1274 /*
1275  * Create a vnode (makenode).
1276  */
1277 int
1278 vn_createat(
1279 	char *pnamep,
1280 	enum uio_seg seg,
1281 	struct vattr *vap,
1282 	enum vcexcl excl,
1283 	int mode,
1284 	struct vnode **vpp,
1285 	enum create why,
1286 	int flag,
1287 	mode_t umask,
1288 	struct vnode *startvp)
1289 {
1290 	struct vnode *dvp;	/* ptr to parent dir vnode */
1291 	struct vnode *vp = NULL;
1292 	struct pathname pn;
1293 	int error;
1294 	int in_crit = 0;
1295 	struct vattr vattr;
1296 	enum symfollow follow;
1297 	int estale_retry = 0;
1298 	uint32_t auditing = AU_AUDITING();
1299 
1300 	ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
1301 
1302 	/* symlink interpretation */
1303 	if ((flag & FNOFOLLOW) || excl == EXCL)
1304 		follow = NO_FOLLOW;
1305 	else
1306 		follow = FOLLOW;
1307 	flag &= ~(FNOFOLLOW|FNOLINKS);
1308 
1309 top:
1310 	/*
1311 	 * Lookup directory.
1312 	 * If new object is a file, call lower level to create it.
1313 	 * Note that it is up to the lower level to enforce exclusive
1314 	 * creation, if the file is already there.
1315 	 * This allows the lower level to do whatever
1316 	 * locking or protocol that is needed to prevent races.
1317 	 * If the new object is directory call lower level to make
1318 	 * the new directory, with "." and "..".
1319 	 */
1320 	if (error = pn_get(pnamep, seg, &pn))
1321 		return (error);
1322 	if (auditing)
1323 		audit_vncreate_start();
1324 	dvp = NULL;
1325 	*vpp = NULL;
1326 	/*
1327 	 * lookup will find the parent directory for the vnode.
1328 	 * When it is done the pn holds the name of the entry
1329 	 * in the directory.
1330 	 * If this is a non-exclusive create we also find the node itself.
1331 	 */
1332 	error = lookuppnat(&pn, NULL, follow, &dvp,
1333 	    (excl == EXCL) ? NULLVPP : vpp, startvp);
1334 	if (error) {
1335 		pn_free(&pn);
1336 		if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1337 			goto top;
1338 		if (why == CRMKDIR && error == EINVAL)
1339 			error = EEXIST;		/* SVID */
1340 		return (error);
1341 	}
1342 
1343 	if (why != CRMKNOD)
1344 		vap->va_mode &= ~VSVTX;
1345 
1346 	/*
1347 	 * If default ACLs are defined for the directory don't apply the
1348 	 * umask if umask is passed.
1349 	 */
1350 
1351 	if (umask) {
1352 
1353 		vsecattr_t vsec;
1354 
1355 		vsec.vsa_aclcnt = 0;
1356 		vsec.vsa_aclentp = NULL;
1357 		vsec.vsa_dfaclcnt = 0;
1358 		vsec.vsa_dfaclentp = NULL;
1359 		vsec.vsa_mask = VSA_DFACLCNT;
1360 		error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL);
1361 		/*
1362 		 * If error is ENOSYS then treat it as no error
1363 		 * Don't want to force all file systems to support
1364 		 * aclent_t style of ACL's.
1365 		 */
1366 		if (error == ENOSYS)
1367 			error = 0;
1368 		if (error) {
1369 			if (*vpp != NULL)
1370 				VN_RELE(*vpp);
1371 			goto out;
1372 		} else {
1373 			/*
1374 			 * Apply the umask if no default ACLs.
1375 			 */
1376 			if (vsec.vsa_dfaclcnt == 0)
1377 				vap->va_mode &= ~umask;
1378 
1379 			/*
1380 			 * VOP_GETSECATTR() may have allocated memory for
1381 			 * ACLs we didn't request, so double-check and
1382 			 * free it if necessary.
1383 			 */
1384 			if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL)
1385 				kmem_free((caddr_t)vsec.vsa_aclentp,
1386 				    vsec.vsa_aclcnt * sizeof (aclent_t));
1387 			if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL)
1388 				kmem_free((caddr_t)vsec.vsa_dfaclentp,
1389 				    vsec.vsa_dfaclcnt * sizeof (aclent_t));
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * In general we want to generate EROFS if the file system is
1395 	 * readonly.  However, POSIX (IEEE Std. 1003.1) section 5.3.1
1396 	 * documents the open system call, and it says that O_CREAT has no
1397 	 * effect if the file already exists.  Bug 1119649 states
1398 	 * that open(path, O_CREAT, ...) fails when attempting to open an
1399 	 * existing file on a read only file system.  Thus, the first part
1400 	 * of the following if statement has 3 checks:
1401 	 *	if the file exists &&
1402 	 *		it is being open with write access &&
1403 	 *		the file system is read only
1404 	 *	then generate EROFS
1405 	 */
1406 	if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) ||
1407 	    (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) {
1408 		if (*vpp)
1409 			VN_RELE(*vpp);
1410 		error = EROFS;
1411 	} else if (excl == NONEXCL && *vpp != NULL) {
1412 		vnode_t *rvp;
1413 
1414 		/*
1415 		 * File already exists.  If a mandatory lock has been
1416 		 * applied, return error.
1417 		 */
1418 		vp = *vpp;
1419 		if (VOP_REALVP(vp, &rvp, NULL) != 0)
1420 			rvp = vp;
1421 		if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) {
1422 			nbl_start_crit(vp, RW_READER);
1423 			in_crit = 1;
1424 		}
1425 		if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) {
1426 			vattr.va_mask = AT_MODE|AT_SIZE;
1427 			if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) {
1428 				goto out;
1429 			}
1430 			if (MANDLOCK(vp, vattr.va_mode)) {
1431 				error = EAGAIN;
1432 				goto out;
1433 			}
1434 			/*
1435 			 * File cannot be truncated if non-blocking mandatory
1436 			 * locks are currently on the file.
1437 			 */
1438 			if ((vap->va_mask & AT_SIZE) && in_crit) {
1439 				u_offset_t offset;
1440 				ssize_t length;
1441 
1442 				offset = vap->va_size > vattr.va_size ?
1443 				    vattr.va_size : vap->va_size;
1444 				length = vap->va_size > vattr.va_size ?
1445 				    vap->va_size - vattr.va_size :
1446 				    vattr.va_size - vap->va_size;
1447 				if (nbl_conflict(vp, NBL_WRITE, offset,
1448 				    length, 0, NULL)) {
1449 					error = EACCES;
1450 					goto out;
1451 				}
1452 			}
1453 		}
1454 
1455 		/*
1456 		 * If the file is the root of a VFS, we've crossed a
1457 		 * mount point and the "containing" directory that we
1458 		 * acquired above (dvp) is irrelevant because it's in
1459 		 * a different file system.  We apply VOP_CREATE to the
1460 		 * target itself instead of to the containing directory
1461 		 * and supply a null path name to indicate (conventionally)
1462 		 * the node itself as the "component" of interest.
1463 		 *
1464 		 * The intercession of the file system is necessary to
1465 		 * ensure that the appropriate permission checks are
1466 		 * done.
1467 		 */
1468 		if (vp->v_flag & VROOT) {
1469 			ASSERT(why != CRMKDIR);
1470 			error = VOP_CREATE(vp, "", vap, excl, mode, vpp,
1471 			    CRED(), flag, NULL, NULL);
1472 			/*
1473 			 * If the create succeeded, it will have created
1474 			 * a new reference to the vnode.  Give up the
1475 			 * original reference.  The assertion should not
1476 			 * get triggered because NBMAND locks only apply to
1477 			 * VREG files.  And if in_crit is non-zero for some
1478 			 * reason, detect that here, rather than when we
1479 			 * deference a null vp.
1480 			 */
1481 			ASSERT(in_crit == 0);
1482 			VN_RELE(vp);
1483 			vp = NULL;
1484 			goto out;
1485 		}
1486 
1487 		/*
1488 		 * Large File API - non-large open (FOFFMAX flag not set)
1489 		 * of regular file fails if the file size exceeds MAXOFF32_T.
1490 		 */
1491 		if (why != CRMKDIR &&
1492 		    !(flag & FOFFMAX) &&
1493 		    (vp->v_type == VREG)) {
1494 			vattr.va_mask = AT_SIZE;
1495 			if ((error = VOP_GETATTR(vp, &vattr, 0,
1496 			    CRED(), NULL))) {
1497 				goto out;
1498 			}
1499 			if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) {
1500 				error = EOVERFLOW;
1501 				goto out;
1502 			}
1503 		}
1504 	}
1505 
1506 	if (error == 0) {
1507 		/*
1508 		 * Call mkdir() if specified, otherwise create().
1509 		 */
1510 		int must_be_dir = pn_fixslash(&pn);	/* trailing '/'? */
1511 
1512 		if (why == CRMKDIR)
1513 			/*
1514 			 * N.B., if vn_createat() ever requests
1515 			 * case-insensitive behavior then it will need
1516 			 * to be passed to VOP_MKDIR().  VOP_CREATE()
1517 			 * will already get it via "flag"
1518 			 */
1519 			error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(),
1520 			    NULL, 0, NULL);
1521 		else if (!must_be_dir)
1522 			error = VOP_CREATE(dvp, pn.pn_path, vap,
1523 			    excl, mode, vpp, CRED(), flag, NULL, NULL);
1524 		else
1525 			error = ENOTDIR;
1526 	}
1527 
1528 out:
1529 
1530 	if (auditing)
1531 		audit_vncreate_finish(*vpp, error);
1532 	if (in_crit) {
1533 		nbl_end_crit(vp);
1534 		in_crit = 0;
1535 	}
1536 	if (vp != NULL) {
1537 		VN_RELE(vp);
1538 		vp = NULL;
1539 	}
1540 	pn_free(&pn);
1541 	VN_RELE(dvp);
1542 	/*
1543 	 * The following clause was added to handle a problem
1544 	 * with NFS consistency.  It is possible that a lookup
1545 	 * of the file to be created succeeded, but the file
1546 	 * itself doesn't actually exist on the server.  This
1547 	 * is chiefly due to the DNLC containing an entry for
1548 	 * the file which has been removed on the server.  In
1549 	 * this case, we just start over.  If there was some
1550 	 * other cause for the ESTALE error, then the lookup
1551 	 * of the file will fail and the error will be returned
1552 	 * above instead of looping around from here.
1553 	 */
1554 	if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1555 		goto top;
1556 	return (error);
1557 }
1558 
1559 int
1560 vn_link(char *from, char *to, enum uio_seg seg)
1561 {
1562 	struct vnode *fvp;		/* from vnode ptr */
1563 	struct vnode *tdvp;		/* to directory vnode ptr */
1564 	struct pathname pn;
1565 	int error;
1566 	struct vattr vattr;
1567 	dev_t fsid;
1568 	int estale_retry = 0;
1569 
1570 top:
1571 	fvp = tdvp = NULL;
1572 	if (error = pn_get(to, seg, &pn))
1573 		return (error);
1574 	if (error = lookupname(from, seg, NO_FOLLOW, NULLVPP, &fvp))
1575 		goto out;
1576 	if (error = lookuppn(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP))
1577 		goto out;
1578 	/*
1579 	 * Make sure both source vnode and target directory vnode are
1580 	 * in the same vfs and that it is writeable.
1581 	 */
1582 	vattr.va_mask = AT_FSID;
1583 	if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL))
1584 		goto out;
1585 	fsid = vattr.va_fsid;
1586 	vattr.va_mask = AT_FSID;
1587 	if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL))
1588 		goto out;
1589 	if (fsid != vattr.va_fsid) {
1590 		error = EXDEV;
1591 		goto out;
1592 	}
1593 	if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) {
1594 		error = EROFS;
1595 		goto out;
1596 	}
1597 	/*
1598 	 * Do the link.
1599 	 */
1600 	(void) pn_fixslash(&pn);
1601 	error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0);
1602 out:
1603 	pn_free(&pn);
1604 	if (fvp)
1605 		VN_RELE(fvp);
1606 	if (tdvp)
1607 		VN_RELE(tdvp);
1608 	if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1609 		goto top;
1610 	return (error);
1611 }
1612 
1613 int
1614 vn_rename(char *from, char *to, enum uio_seg seg)
1615 {
1616 	return (vn_renameat(NULL, from, NULL, to, seg));
1617 }
1618 
1619 int
1620 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp,
1621 		char *tname, enum uio_seg seg)
1622 {
1623 	int error;
1624 	struct vattr vattr;
1625 	struct pathname fpn;		/* from pathname */
1626 	struct pathname tpn;		/* to pathname */
1627 	dev_t fsid;
1628 	int in_crit_src, in_crit_targ;
1629 	vnode_t *fromvp, *fvp;
1630 	vnode_t *tovp, *targvp;
1631 	int estale_retry = 0;
1632 	uint32_t auditing = AU_AUDITING();
1633 
1634 top:
1635 	fvp = fromvp = tovp = targvp = NULL;
1636 	in_crit_src = in_crit_targ = 0;
1637 	/*
1638 	 * Get to and from pathnames.
1639 	 */
1640 	if (error = pn_get(fname, seg, &fpn))
1641 		return (error);
1642 	if (error = pn_get(tname, seg, &tpn)) {
1643 		pn_free(&fpn);
1644 		return (error);
1645 	}
1646 
1647 	/*
1648 	 * First we need to resolve the correct directories
1649 	 * The passed in directories may only be a starting point,
1650 	 * but we need the real directories the file(s) live in.
1651 	 * For example the fname may be something like usr/lib/sparc
1652 	 * and we were passed in the / directory, but we need to
1653 	 * use the lib directory for the rename.
1654 	 */
1655 
1656 	if (auditing)
1657 		audit_setfsat_path(1);
1658 	/*
1659 	 * Lookup to and from directories.
1660 	 */
1661 	if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) {
1662 		goto out;
1663 	}
1664 
1665 	/*
1666 	 * Make sure there is an entry.
1667 	 */
1668 	if (fvp == NULL) {
1669 		error = ENOENT;
1670 		goto out;
1671 	}
1672 
1673 	if (auditing)
1674 		audit_setfsat_path(3);
1675 	if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) {
1676 		goto out;
1677 	}
1678 
1679 	/*
1680 	 * Make sure both the from vnode directory and the to directory
1681 	 * are in the same vfs and the to directory is writable.
1682 	 * We check fsid's, not vfs pointers, so loopback fs works.
1683 	 */
1684 	if (fromvp != tovp) {
1685 		vattr.va_mask = AT_FSID;
1686 		if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL))
1687 			goto out;
1688 		fsid = vattr.va_fsid;
1689 		vattr.va_mask = AT_FSID;
1690 		if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL))
1691 			goto out;
1692 		if (fsid != vattr.va_fsid) {
1693 			error = EXDEV;
1694 			goto out;
1695 		}
1696 	}
1697 
1698 	if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) {
1699 		error = EROFS;
1700 		goto out;
1701 	}
1702 
1703 	if (targvp && (fvp != targvp)) {
1704 		nbl_start_crit(targvp, RW_READER);
1705 		in_crit_targ = 1;
1706 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
1707 			error = EACCES;
1708 			goto out;
1709 		}
1710 	}
1711 
1712 	if (nbl_need_check(fvp)) {
1713 		nbl_start_crit(fvp, RW_READER);
1714 		in_crit_src = 1;
1715 		if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) {
1716 			error = EACCES;
1717 			goto out;
1718 		}
1719 	}
1720 
1721 	/*
1722 	 * Do the rename.
1723 	 */
1724 	(void) pn_fixslash(&tpn);
1725 	error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(),
1726 	    NULL, 0);
1727 
1728 out:
1729 	pn_free(&fpn);
1730 	pn_free(&tpn);
1731 	if (in_crit_src)
1732 		nbl_end_crit(fvp);
1733 	if (in_crit_targ)
1734 		nbl_end_crit(targvp);
1735 	if (fromvp)
1736 		VN_RELE(fromvp);
1737 	if (tovp)
1738 		VN_RELE(tovp);
1739 	if (targvp)
1740 		VN_RELE(targvp);
1741 	if (fvp)
1742 		VN_RELE(fvp);
1743 	if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1744 		goto top;
1745 	return (error);
1746 }
1747 
1748 /*
1749  * Remove a file or directory.
1750  */
1751 int
1752 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag)
1753 {
1754 	return (vn_removeat(NULL, fnamep, seg, dirflag));
1755 }
1756 
1757 int
1758 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag)
1759 {
1760 	struct vnode *vp;		/* entry vnode */
1761 	struct vnode *dvp;		/* ptr to parent dir vnode */
1762 	struct vnode *coveredvp;
1763 	struct pathname pn;		/* name of entry */
1764 	enum vtype vtype;
1765 	int error;
1766 	struct vfs *vfsp;
1767 	struct vfs *dvfsp;	/* ptr to parent dir vfs */
1768 	int in_crit = 0;
1769 	int estale_retry = 0;
1770 
1771 top:
1772 	if (error = pn_get(fnamep, seg, &pn))
1773 		return (error);
1774 	dvp = vp = NULL;
1775 	if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) {
1776 		pn_free(&pn);
1777 		if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1778 			goto top;
1779 		return (error);
1780 	}
1781 
1782 	/*
1783 	 * Make sure there is an entry.
1784 	 */
1785 	if (vp == NULL) {
1786 		error = ENOENT;
1787 		goto out;
1788 	}
1789 
1790 	vfsp = vp->v_vfsp;
1791 	dvfsp = dvp->v_vfsp;
1792 
1793 	/*
1794 	 * If the named file is the root of a mounted filesystem, fail,
1795 	 * unless it's marked unlinkable.  In that case, unmount the
1796 	 * filesystem and proceed to unlink the covered vnode.  (If the
1797 	 * covered vnode is a directory, use rmdir instead of unlink,
1798 	 * to avoid file system corruption.)
1799 	 */
1800 	if (vp->v_flag & VROOT) {
1801 		if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) {
1802 			error = EBUSY;
1803 			goto out;
1804 		}
1805 
1806 		/*
1807 		 * Namefs specific code starts here.
1808 		 */
1809 
1810 		if (dirflag == RMDIRECTORY) {
1811 			/*
1812 			 * User called rmdir(2) on a file that has
1813 			 * been namefs mounted on top of.  Since
1814 			 * namefs doesn't allow directories to
1815 			 * be mounted on other files we know
1816 			 * vp is not of type VDIR so fail to operation.
1817 			 */
1818 			error = ENOTDIR;
1819 			goto out;
1820 		}
1821 
1822 		/*
1823 		 * If VROOT is still set after grabbing vp->v_lock,
1824 		 * noone has finished nm_unmount so far and coveredvp
1825 		 * is valid.
1826 		 * If we manage to grab vn_vfswlock(coveredvp) before releasing
1827 		 * vp->v_lock, any race window is eliminated.
1828 		 */
1829 
1830 		mutex_enter(&vp->v_lock);
1831 		if ((vp->v_flag & VROOT) == 0) {
1832 			/* Someone beat us to the unmount */
1833 			mutex_exit(&vp->v_lock);
1834 			error = EBUSY;
1835 			goto out;
1836 		}
1837 		vfsp = vp->v_vfsp;
1838 		coveredvp = vfsp->vfs_vnodecovered;
1839 		ASSERT(coveredvp);
1840 		/*
1841 		 * Note: Implementation of vn_vfswlock shows that ordering of
1842 		 * v_lock / vn_vfswlock is not an issue here.
1843 		 */
1844 		error = vn_vfswlock(coveredvp);
1845 		mutex_exit(&vp->v_lock);
1846 
1847 		if (error)
1848 			goto out;
1849 
1850 		VN_HOLD(coveredvp);
1851 		VN_RELE(vp);
1852 		error = dounmount(vfsp, 0, CRED());
1853 
1854 		/*
1855 		 * Unmounted the namefs file system; now get
1856 		 * the object it was mounted over.
1857 		 */
1858 		vp = coveredvp;
1859 		/*
1860 		 * If namefs was mounted over a directory, then
1861 		 * we want to use rmdir() instead of unlink().
1862 		 */
1863 		if (vp->v_type == VDIR)
1864 			dirflag = RMDIRECTORY;
1865 
1866 		if (error)
1867 			goto out;
1868 	}
1869 
1870 	/*
1871 	 * Make sure filesystem is writeable.
1872 	 * We check the parent directory's vfs in case this is an lofs vnode.
1873 	 */
1874 	if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) {
1875 		error = EROFS;
1876 		goto out;
1877 	}
1878 
1879 	vtype = vp->v_type;
1880 
1881 	/*
1882 	 * If there is the possibility of an nbmand share reservation, make
1883 	 * sure it's okay to remove the file.  Keep a reference to the
1884 	 * vnode, so that we can exit the nbl critical region after
1885 	 * calling VOP_REMOVE.
1886 	 * If there is no possibility of an nbmand share reservation,
1887 	 * release the vnode reference now.  Filesystems like NFS may
1888 	 * behave differently if there is an extra reference, so get rid of
1889 	 * this one.  Fortunately, we can't have nbmand mounts on NFS
1890 	 * filesystems.
1891 	 */
1892 	if (nbl_need_check(vp)) {
1893 		nbl_start_crit(vp, RW_READER);
1894 		in_crit = 1;
1895 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
1896 			error = EACCES;
1897 			goto out;
1898 		}
1899 	} else {
1900 		VN_RELE(vp);
1901 		vp = NULL;
1902 	}
1903 
1904 	if (dirflag == RMDIRECTORY) {
1905 		/*
1906 		 * Caller is using rmdir(2), which can only be applied to
1907 		 * directories.
1908 		 */
1909 		if (vtype != VDIR) {
1910 			error = ENOTDIR;
1911 		} else {
1912 			vnode_t *cwd;
1913 			proc_t *pp = curproc;
1914 
1915 			mutex_enter(&pp->p_lock);
1916 			cwd = PTOU(pp)->u_cdir;
1917 			VN_HOLD(cwd);
1918 			mutex_exit(&pp->p_lock);
1919 			error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(),
1920 			    NULL, 0);
1921 			VN_RELE(cwd);
1922 		}
1923 	} else {
1924 		/*
1925 		 * Unlink(2) can be applied to anything.
1926 		 */
1927 		error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
1928 	}
1929 
1930 out:
1931 	pn_free(&pn);
1932 	if (in_crit) {
1933 		nbl_end_crit(vp);
1934 		in_crit = 0;
1935 	}
1936 	if (vp != NULL)
1937 		VN_RELE(vp);
1938 	if (dvp != NULL)
1939 		VN_RELE(dvp);
1940 	if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1941 		goto top;
1942 	return (error);
1943 }
1944 
1945 /*
1946  * Utility function to compare equality of vnodes.
1947  * Compare the underlying real vnodes, if there are underlying vnodes.
1948  * This is a more thorough comparison than the VN_CMP() macro provides.
1949  */
1950 int
1951 vn_compare(vnode_t *vp1, vnode_t *vp2)
1952 {
1953 	vnode_t *realvp;
1954 
1955 	if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0)
1956 		vp1 = realvp;
1957 	if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0)
1958 		vp2 = realvp;
1959 	return (VN_CMP(vp1, vp2));
1960 }
1961 
1962 /*
1963  * The number of locks to hash into.  This value must be a power
1964  * of 2 minus 1 and should probably also be prime.
1965  */
1966 #define	NUM_BUCKETS	1023
1967 
1968 struct  vn_vfslocks_bucket {
1969 	kmutex_t vb_lock;
1970 	vn_vfslocks_entry_t *vb_list;
1971 	char pad[64 - sizeof (kmutex_t) - sizeof (void *)];
1972 };
1973 
1974 /*
1975  * Total number of buckets will be NUM_BUCKETS + 1 .
1976  */
1977 
1978 #pragma	align	64(vn_vfslocks_buckets)
1979 static	struct vn_vfslocks_bucket	vn_vfslocks_buckets[NUM_BUCKETS + 1];
1980 
1981 #define	VN_VFSLOCKS_SHIFT	9
1982 
1983 #define	VN_VFSLOCKS_HASH(vfsvpptr)	\
1984 	((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS)
1985 
1986 /*
1987  * vn_vfslocks_getlock() uses an HASH scheme to generate
1988  * rwstlock using vfs/vnode pointer passed to it.
1989  *
1990  * vn_vfslocks_rele() releases a reference in the
1991  * HASH table which allows the entry allocated by
1992  * vn_vfslocks_getlock() to be freed at a later
1993  * stage when the refcount drops to zero.
1994  */
1995 
1996 vn_vfslocks_entry_t *
1997 vn_vfslocks_getlock(void *vfsvpptr)
1998 {
1999 	struct vn_vfslocks_bucket *bp;
2000 	vn_vfslocks_entry_t *vep;
2001 	vn_vfslocks_entry_t *tvep;
2002 
2003 	ASSERT(vfsvpptr != NULL);
2004 	bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)];
2005 
2006 	mutex_enter(&bp->vb_lock);
2007 	for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2008 		if (vep->ve_vpvfs == vfsvpptr) {
2009 			vep->ve_refcnt++;
2010 			mutex_exit(&bp->vb_lock);
2011 			return (vep);
2012 		}
2013 	}
2014 	mutex_exit(&bp->vb_lock);
2015 	vep = kmem_alloc(sizeof (*vep), KM_SLEEP);
2016 	rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL);
2017 	vep->ve_vpvfs = (char *)vfsvpptr;
2018 	vep->ve_refcnt = 1;
2019 	mutex_enter(&bp->vb_lock);
2020 	for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) {
2021 		if (tvep->ve_vpvfs == vfsvpptr) {
2022 			tvep->ve_refcnt++;
2023 			mutex_exit(&bp->vb_lock);
2024 
2025 			/*
2026 			 * There is already an entry in the hash
2027 			 * destroy what we just allocated.
2028 			 */
2029 			rwst_destroy(&vep->ve_lock);
2030 			kmem_free(vep, sizeof (*vep));
2031 			return (tvep);
2032 		}
2033 	}
2034 	vep->ve_next = bp->vb_list;
2035 	bp->vb_list = vep;
2036 	mutex_exit(&bp->vb_lock);
2037 	return (vep);
2038 }
2039 
2040 void
2041 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent)
2042 {
2043 	struct vn_vfslocks_bucket *bp;
2044 	vn_vfslocks_entry_t *vep;
2045 	vn_vfslocks_entry_t *pvep;
2046 
2047 	ASSERT(vepent != NULL);
2048 	ASSERT(vepent->ve_vpvfs != NULL);
2049 
2050 	bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)];
2051 
2052 	mutex_enter(&bp->vb_lock);
2053 	vepent->ve_refcnt--;
2054 
2055 	if ((int32_t)vepent->ve_refcnt < 0)
2056 		cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative");
2057 
2058 	if (vepent->ve_refcnt == 0) {
2059 		for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2060 			if (vep->ve_vpvfs == vepent->ve_vpvfs) {
2061 				if (bp->vb_list == vep)
2062 					bp->vb_list = vep->ve_next;
2063 				else {
2064 					/* LINTED */
2065 					pvep->ve_next = vep->ve_next;
2066 				}
2067 				mutex_exit(&bp->vb_lock);
2068 				rwst_destroy(&vep->ve_lock);
2069 				kmem_free(vep, sizeof (*vep));
2070 				return;
2071 			}
2072 			pvep = vep;
2073 		}
2074 		cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found");
2075 	}
2076 	mutex_exit(&bp->vb_lock);
2077 }
2078 
2079 /*
2080  * vn_vfswlock_wait is used to implement a lock which is logically a writers
2081  * lock protecting the v_vfsmountedhere field.
2082  * vn_vfswlock_wait has been modified to be similar to vn_vfswlock,
2083  * except that it blocks to acquire the lock VVFSLOCK.
2084  *
2085  * traverse() and routines re-implementing part of traverse (e.g. autofs)
2086  * need to hold this lock. mount(), vn_rename(), vn_remove() and so on
2087  * need the non-blocking version of the writers lock i.e. vn_vfswlock
2088  */
2089 int
2090 vn_vfswlock_wait(vnode_t *vp)
2091 {
2092 	int retval;
2093 	vn_vfslocks_entry_t *vpvfsentry;
2094 	ASSERT(vp != NULL);
2095 
2096 	vpvfsentry = vn_vfslocks_getlock(vp);
2097 	retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER);
2098 
2099 	if (retval == EINTR) {
2100 		vn_vfslocks_rele(vpvfsentry);
2101 		return (EINTR);
2102 	}
2103 	return (retval);
2104 }
2105 
2106 int
2107 vn_vfsrlock_wait(vnode_t *vp)
2108 {
2109 	int retval;
2110 	vn_vfslocks_entry_t *vpvfsentry;
2111 	ASSERT(vp != NULL);
2112 
2113 	vpvfsentry = vn_vfslocks_getlock(vp);
2114 	retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER);
2115 
2116 	if (retval == EINTR) {
2117 		vn_vfslocks_rele(vpvfsentry);
2118 		return (EINTR);
2119 	}
2120 
2121 	return (retval);
2122 }
2123 
2124 
2125 /*
2126  * vn_vfswlock is used to implement a lock which is logically a writers lock
2127  * protecting the v_vfsmountedhere field.
2128  */
2129 int
2130 vn_vfswlock(vnode_t *vp)
2131 {
2132 	vn_vfslocks_entry_t *vpvfsentry;
2133 
2134 	/*
2135 	 * If vp is NULL then somebody is trying to lock the covered vnode
2136 	 * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2137 	 * only happen when unmounting /.  Since that operation will fail
2138 	 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2139 	 */
2140 	if (vp == NULL)
2141 		return (EBUSY);
2142 
2143 	vpvfsentry = vn_vfslocks_getlock(vp);
2144 
2145 	if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER))
2146 		return (0);
2147 
2148 	vn_vfslocks_rele(vpvfsentry);
2149 	return (EBUSY);
2150 }
2151 
2152 int
2153 vn_vfsrlock(vnode_t *vp)
2154 {
2155 	vn_vfslocks_entry_t *vpvfsentry;
2156 
2157 	/*
2158 	 * If vp is NULL then somebody is trying to lock the covered vnode
2159 	 * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2160 	 * only happen when unmounting /.  Since that operation will fail
2161 	 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2162 	 */
2163 	if (vp == NULL)
2164 		return (EBUSY);
2165 
2166 	vpvfsentry = vn_vfslocks_getlock(vp);
2167 
2168 	if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER))
2169 		return (0);
2170 
2171 	vn_vfslocks_rele(vpvfsentry);
2172 	return (EBUSY);
2173 }
2174 
2175 void
2176 vn_vfsunlock(vnode_t *vp)
2177 {
2178 	vn_vfslocks_entry_t *vpvfsentry;
2179 
2180 	/*
2181 	 * ve_refcnt needs to be decremented twice.
2182 	 * 1. To release refernce after a call to vn_vfslocks_getlock()
2183 	 * 2. To release the reference from the locking routines like
2184 	 *    vn_vfsrlock/vn_vfswlock etc,.
2185 	 */
2186 	vpvfsentry = vn_vfslocks_getlock(vp);
2187 	vn_vfslocks_rele(vpvfsentry);
2188 
2189 	rwst_exit(&vpvfsentry->ve_lock);
2190 	vn_vfslocks_rele(vpvfsentry);
2191 }
2192 
2193 int
2194 vn_vfswlock_held(vnode_t *vp)
2195 {
2196 	int held;
2197 	vn_vfslocks_entry_t *vpvfsentry;
2198 
2199 	ASSERT(vp != NULL);
2200 
2201 	vpvfsentry = vn_vfslocks_getlock(vp);
2202 	held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER);
2203 
2204 	vn_vfslocks_rele(vpvfsentry);
2205 	return (held);
2206 }
2207 
2208 
2209 int
2210 vn_make_ops(
2211 	const char *name,			/* Name of file system */
2212 	const fs_operation_def_t *templ,	/* Operation specification */
2213 	vnodeops_t **actual)			/* Return the vnodeops */
2214 {
2215 	int unused_ops;
2216 	int error;
2217 
2218 	*actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP);
2219 
2220 	(*actual)->vnop_name = name;
2221 
2222 	error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ);
2223 	if (error) {
2224 		kmem_free(*actual, sizeof (vnodeops_t));
2225 	}
2226 
2227 #if DEBUG
2228 	if (unused_ops != 0)
2229 		cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied "
2230 		    "but not used", name, unused_ops);
2231 #endif
2232 
2233 	return (error);
2234 }
2235 
2236 /*
2237  * Free the vnodeops created as a result of vn_make_ops()
2238  */
2239 void
2240 vn_freevnodeops(vnodeops_t *vnops)
2241 {
2242 	kmem_free(vnops, sizeof (vnodeops_t));
2243 }
2244 
2245 /*
2246  * Vnode cache.
2247  */
2248 
2249 /* ARGSUSED */
2250 static int
2251 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
2252 {
2253 	struct vnode *vp;
2254 
2255 	vp = buf;
2256 
2257 	mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
2258 	mutex_init(&vp->v_vsd_lock, NULL, MUTEX_DEFAULT, NULL);
2259 	cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
2260 	rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
2261 	vp->v_femhead = NULL;	/* Must be done before vn_reinit() */
2262 	vp->v_path = NULL;
2263 	vp->v_mpssdata = NULL;
2264 	vp->v_vsd = NULL;
2265 	vp->v_fopdata = NULL;
2266 
2267 	return (0);
2268 }
2269 
2270 /* ARGSUSED */
2271 static void
2272 vn_cache_destructor(void *buf, void *cdrarg)
2273 {
2274 	struct vnode *vp;
2275 
2276 	vp = buf;
2277 
2278 	rw_destroy(&vp->v_nbllock);
2279 	cv_destroy(&vp->v_cv);
2280 	mutex_destroy(&vp->v_vsd_lock);
2281 	mutex_destroy(&vp->v_lock);
2282 }
2283 
2284 void
2285 vn_create_cache(void)
2286 {
2287 	vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode), 64,
2288 	    vn_cache_constructor, vn_cache_destructor, NULL, NULL,
2289 	    NULL, 0);
2290 }
2291 
2292 void
2293 vn_destroy_cache(void)
2294 {
2295 	kmem_cache_destroy(vn_cache);
2296 }
2297 
2298 /*
2299  * Used by file systems when fs-specific nodes (e.g., ufs inodes) are
2300  * cached by the file system and vnodes remain associated.
2301  */
2302 void
2303 vn_recycle(vnode_t *vp)
2304 {
2305 	ASSERT(vp->v_pages == NULL);
2306 
2307 	/*
2308 	 * XXX - This really belongs in vn_reinit(), but we have some issues
2309 	 * with the counts.  Best to have it here for clean initialization.
2310 	 */
2311 	vp->v_rdcnt = 0;
2312 	vp->v_wrcnt = 0;
2313 	vp->v_mmap_read = 0;
2314 	vp->v_mmap_write = 0;
2315 
2316 	/*
2317 	 * If FEM was in use, make sure everything gets cleaned up
2318 	 * NOTE: vp->v_femhead is initialized to NULL in the vnode
2319 	 * constructor.
2320 	 */
2321 	if (vp->v_femhead) {
2322 		/* XXX - There should be a free_femhead() that does all this */
2323 		ASSERT(vp->v_femhead->femh_list == NULL);
2324 		mutex_destroy(&vp->v_femhead->femh_lock);
2325 		kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2326 		vp->v_femhead = NULL;
2327 	}
2328 	if (vp->v_path) {
2329 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2330 		vp->v_path = NULL;
2331 	}
2332 
2333 	if (vp->v_fopdata != NULL) {
2334 		free_fopdata(vp);
2335 	}
2336 	vp->v_mpssdata = NULL;
2337 	vsd_free(vp);
2338 }
2339 
2340 /*
2341  * Used to reset the vnode fields including those that are directly accessible
2342  * as well as those which require an accessor function.
2343  *
2344  * Does not initialize:
2345  *	synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv
2346  *	v_data (since FS-nodes and vnodes point to each other and should
2347  *		be updated simultaneously)
2348  *	v_op (in case someone needs to make a VOP call on this object)
2349  */
2350 void
2351 vn_reinit(vnode_t *vp)
2352 {
2353 	vp->v_count = 1;
2354 	vp->v_count_dnlc = 0;
2355 	vp->v_vfsp = NULL;
2356 	vp->v_stream = NULL;
2357 	vp->v_vfsmountedhere = NULL;
2358 	vp->v_flag = 0;
2359 	vp->v_type = VNON;
2360 	vp->v_rdev = NODEV;
2361 
2362 	vp->v_filocks = NULL;
2363 	vp->v_shrlocks = NULL;
2364 	vp->v_pages = NULL;
2365 
2366 	vp->v_locality = NULL;
2367 	vp->v_xattrdir = NULL;
2368 
2369 	/* Handles v_femhead, v_path, and the r/w/map counts */
2370 	vn_recycle(vp);
2371 }
2372 
2373 vnode_t *
2374 vn_alloc(int kmflag)
2375 {
2376 	vnode_t *vp;
2377 
2378 	vp = kmem_cache_alloc(vn_cache, kmflag);
2379 
2380 	if (vp != NULL) {
2381 		vp->v_femhead = NULL;	/* Must be done before vn_reinit() */
2382 		vp->v_fopdata = NULL;
2383 		vn_reinit(vp);
2384 	}
2385 
2386 	return (vp);
2387 }
2388 
2389 void
2390 vn_free(vnode_t *vp)
2391 {
2392 	ASSERT(vp->v_shrlocks == NULL);
2393 	ASSERT(vp->v_filocks == NULL);
2394 
2395 	/*
2396 	 * Some file systems call vn_free() with v_count of zero,
2397 	 * some with v_count of 1.  In any case, the value should
2398 	 * never be anything else.
2399 	 */
2400 	ASSERT((vp->v_count == 0) || (vp->v_count == 1));
2401 	ASSERT(vp->v_count_dnlc == 0);
2402 	if (vp->v_path != NULL) {
2403 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2404 		vp->v_path = NULL;
2405 	}
2406 
2407 	/* If FEM was in use, make sure everything gets cleaned up */
2408 	if (vp->v_femhead) {
2409 		/* XXX - There should be a free_femhead() that does all this */
2410 		ASSERT(vp->v_femhead->femh_list == NULL);
2411 		mutex_destroy(&vp->v_femhead->femh_lock);
2412 		kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2413 		vp->v_femhead = NULL;
2414 	}
2415 
2416 	if (vp->v_fopdata != NULL) {
2417 		free_fopdata(vp);
2418 	}
2419 	vp->v_mpssdata = NULL;
2420 	vsd_free(vp);
2421 	kmem_cache_free(vn_cache, vp);
2422 }
2423 
2424 /*
2425  * vnode status changes, should define better states than 1, 0.
2426  */
2427 void
2428 vn_reclaim(vnode_t *vp)
2429 {
2430 	vfs_t   *vfsp = vp->v_vfsp;
2431 
2432 	if (vfsp == NULL ||
2433 	    vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2434 		return;
2435 	}
2436 	(void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED);
2437 }
2438 
2439 void
2440 vn_idle(vnode_t *vp)
2441 {
2442 	vfs_t   *vfsp = vp->v_vfsp;
2443 
2444 	if (vfsp == NULL ||
2445 	    vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2446 		return;
2447 	}
2448 	(void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED);
2449 }
2450 void
2451 vn_exists(vnode_t *vp)
2452 {
2453 	vfs_t   *vfsp = vp->v_vfsp;
2454 
2455 	if (vfsp == NULL ||
2456 	    vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2457 		return;
2458 	}
2459 	(void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS);
2460 }
2461 
2462 void
2463 vn_invalid(vnode_t *vp)
2464 {
2465 	vfs_t   *vfsp = vp->v_vfsp;
2466 
2467 	if (vfsp == NULL ||
2468 	    vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2469 		return;
2470 	}
2471 	(void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED);
2472 }
2473 
2474 /* Vnode event notification */
2475 
2476 int
2477 vnevent_support(vnode_t *vp, caller_context_t *ct)
2478 {
2479 	if (vp == NULL)
2480 		return (EINVAL);
2481 
2482 	return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct));
2483 }
2484 
2485 void
2486 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2487 {
2488 	if (vp == NULL || vp->v_femhead == NULL) {
2489 		return;
2490 	}
2491 	(void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
2492 }
2493 
2494 void
2495 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
2496     caller_context_t *ct)
2497 {
2498 	if (vp == NULL || vp->v_femhead == NULL) {
2499 		return;
2500 	}
2501 	(void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct);
2502 }
2503 
2504 void
2505 vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct)
2506 {
2507 	if (vp == NULL || vp->v_femhead == NULL) {
2508 		return;
2509 	}
2510 	(void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct);
2511 }
2512 
2513 void
2514 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2515 {
2516 	if (vp == NULL || vp->v_femhead == NULL) {
2517 		return;
2518 	}
2519 	(void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct);
2520 }
2521 
2522 void
2523 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2524 {
2525 	if (vp == NULL || vp->v_femhead == NULL) {
2526 		return;
2527 	}
2528 	(void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct);
2529 }
2530 
2531 void
2532 vnevent_create(vnode_t *vp, caller_context_t *ct)
2533 {
2534 	if (vp == NULL || vp->v_femhead == NULL) {
2535 		return;
2536 	}
2537 	(void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct);
2538 }
2539 
2540 void
2541 vnevent_link(vnode_t *vp, caller_context_t *ct)
2542 {
2543 	if (vp == NULL || vp->v_femhead == NULL) {
2544 		return;
2545 	}
2546 	(void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct);
2547 }
2548 
2549 void
2550 vnevent_mountedover(vnode_t *vp, caller_context_t *ct)
2551 {
2552 	if (vp == NULL || vp->v_femhead == NULL) {
2553 		return;
2554 	}
2555 	(void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct);
2556 }
2557 
2558 /*
2559  * Vnode accessors.
2560  */
2561 
2562 int
2563 vn_is_readonly(vnode_t *vp)
2564 {
2565 	return (vp->v_vfsp->vfs_flag & VFS_RDONLY);
2566 }
2567 
2568 int
2569 vn_has_flocks(vnode_t *vp)
2570 {
2571 	return (vp->v_filocks != NULL);
2572 }
2573 
2574 int
2575 vn_has_mandatory_locks(vnode_t *vp, int mode)
2576 {
2577 	return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode)));
2578 }
2579 
2580 int
2581 vn_has_cached_data(vnode_t *vp)
2582 {
2583 	return (vp->v_pages != NULL);
2584 }
2585 
2586 /*
2587  * Return 0 if the vnode in question shouldn't be permitted into a zone via
2588  * zone_enter(2).
2589  */
2590 int
2591 vn_can_change_zones(vnode_t *vp)
2592 {
2593 	struct vfssw *vswp;
2594 	int allow = 1;
2595 	vnode_t *rvp;
2596 
2597 	if (nfs_global_client_only != 0)
2598 		return (1);
2599 
2600 	/*
2601 	 * We always want to look at the underlying vnode if there is one.
2602 	 */
2603 	if (VOP_REALVP(vp, &rvp, NULL) != 0)
2604 		rvp = vp;
2605 	/*
2606 	 * Some pseudo filesystems (including doorfs) don't actually register
2607 	 * their vfsops_t, so the following may return NULL; we happily let
2608 	 * such vnodes switch zones.
2609 	 */
2610 	vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp));
2611 	if (vswp != NULL) {
2612 		if (vswp->vsw_flag & VSW_NOTZONESAFE)
2613 			allow = 0;
2614 		vfs_unrefvfssw(vswp);
2615 	}
2616 	return (allow);
2617 }
2618 
2619 /*
2620  * Return nonzero if the vnode is a mount point, zero if not.
2621  */
2622 int
2623 vn_ismntpt(vnode_t *vp)
2624 {
2625 	return (vp->v_vfsmountedhere != NULL);
2626 }
2627 
2628 /* Retrieve the vfs (if any) mounted on this vnode */
2629 vfs_t *
2630 vn_mountedvfs(vnode_t *vp)
2631 {
2632 	return (vp->v_vfsmountedhere);
2633 }
2634 
2635 /*
2636  * Return nonzero if the vnode is referenced by the dnlc, zero if not.
2637  */
2638 int
2639 vn_in_dnlc(vnode_t *vp)
2640 {
2641 	return (vp->v_count_dnlc > 0);
2642 }
2643 
2644 /*
2645  * vn_has_other_opens() checks whether a particular file is opened by more than
2646  * just the caller and whether the open is for read and/or write.
2647  * This routine is for calling after the caller has already called VOP_OPEN()
2648  * and the caller wishes to know if they are the only one with it open for
2649  * the mode(s) specified.
2650  *
2651  * Vnode counts are only kept on regular files (v_type=VREG).
2652  */
2653 int
2654 vn_has_other_opens(
2655 	vnode_t *vp,
2656 	v_mode_t mode)
2657 {
2658 
2659 	ASSERT(vp != NULL);
2660 
2661 	switch (mode) {
2662 	case V_WRITE:
2663 		if (vp->v_wrcnt > 1)
2664 			return (V_TRUE);
2665 		break;
2666 	case V_RDORWR:
2667 		if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1))
2668 			return (V_TRUE);
2669 		break;
2670 	case V_RDANDWR:
2671 		if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1))
2672 			return (V_TRUE);
2673 		break;
2674 	case V_READ:
2675 		if (vp->v_rdcnt > 1)
2676 			return (V_TRUE);
2677 		break;
2678 	}
2679 
2680 	return (V_FALSE);
2681 }
2682 
2683 /*
2684  * vn_is_opened() checks whether a particular file is opened and
2685  * whether the open is for read and/or write.
2686  *
2687  * Vnode counts are only kept on regular files (v_type=VREG).
2688  */
2689 int
2690 vn_is_opened(
2691 	vnode_t *vp,
2692 	v_mode_t mode)
2693 {
2694 
2695 	ASSERT(vp != NULL);
2696 
2697 	switch (mode) {
2698 	case V_WRITE:
2699 		if (vp->v_wrcnt)
2700 			return (V_TRUE);
2701 		break;
2702 	case V_RDANDWR:
2703 		if (vp->v_rdcnt && vp->v_wrcnt)
2704 			return (V_TRUE);
2705 		break;
2706 	case V_RDORWR:
2707 		if (vp->v_rdcnt || vp->v_wrcnt)
2708 			return (V_TRUE);
2709 		break;
2710 	case V_READ:
2711 		if (vp->v_rdcnt)
2712 			return (V_TRUE);
2713 		break;
2714 	}
2715 
2716 	return (V_FALSE);
2717 }
2718 
2719 /*
2720  * vn_is_mapped() checks whether a particular file is mapped and whether
2721  * the file is mapped read and/or write.
2722  */
2723 int
2724 vn_is_mapped(
2725 	vnode_t *vp,
2726 	v_mode_t mode)
2727 {
2728 
2729 	ASSERT(vp != NULL);
2730 
2731 #if !defined(_LP64)
2732 	switch (mode) {
2733 	/*
2734 	 * The atomic_add_64_nv functions force atomicity in the
2735 	 * case of 32 bit architectures. Otherwise the 64 bit values
2736 	 * require two fetches. The value of the fields may be
2737 	 * (potentially) changed between the first fetch and the
2738 	 * second
2739 	 */
2740 	case V_WRITE:
2741 		if (atomic_add_64_nv((&(vp->v_mmap_write)), 0))
2742 			return (V_TRUE);
2743 		break;
2744 	case V_RDANDWR:
2745 		if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) &&
2746 		    (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2747 			return (V_TRUE);
2748 		break;
2749 	case V_RDORWR:
2750 		if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) ||
2751 		    (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2752 			return (V_TRUE);
2753 		break;
2754 	case V_READ:
2755 		if (atomic_add_64_nv((&(vp->v_mmap_read)), 0))
2756 			return (V_TRUE);
2757 		break;
2758 	}
2759 #else
2760 	switch (mode) {
2761 	case V_WRITE:
2762 		if (vp->v_mmap_write)
2763 			return (V_TRUE);
2764 		break;
2765 	case V_RDANDWR:
2766 		if (vp->v_mmap_read && vp->v_mmap_write)
2767 			return (V_TRUE);
2768 		break;
2769 	case V_RDORWR:
2770 		if (vp->v_mmap_read || vp->v_mmap_write)
2771 			return (V_TRUE);
2772 		break;
2773 	case V_READ:
2774 		if (vp->v_mmap_read)
2775 			return (V_TRUE);
2776 		break;
2777 	}
2778 #endif
2779 
2780 	return (V_FALSE);
2781 }
2782 
2783 /*
2784  * Set the operations vector for a vnode.
2785  *
2786  * FEM ensures that the v_femhead pointer is filled in before the
2787  * v_op pointer is changed.  This means that if the v_femhead pointer
2788  * is NULL, and the v_op field hasn't changed since before which checked
2789  * the v_femhead pointer; then our update is ok - we are not racing with
2790  * FEM.
2791  */
2792 void
2793 vn_setops(vnode_t *vp, vnodeops_t *vnodeops)
2794 {
2795 	vnodeops_t	*op;
2796 
2797 	ASSERT(vp != NULL);
2798 	ASSERT(vnodeops != NULL);
2799 
2800 	op = vp->v_op;
2801 	membar_consumer();
2802 	/*
2803 	 * If vp->v_femhead == NULL, then we'll call casptr() to do the
2804 	 * compare-and-swap on vp->v_op.  If either fails, then FEM is
2805 	 * in effect on the vnode and we need to have FEM deal with it.
2806 	 */
2807 	if (vp->v_femhead != NULL || casptr(&vp->v_op, op, vnodeops) != op) {
2808 		fem_setvnops(vp, vnodeops);
2809 	}
2810 }
2811 
2812 /*
2813  * Retrieve the operations vector for a vnode
2814  * As with vn_setops(above); make sure we aren't racing with FEM.
2815  * FEM sets the v_op to a special, internal, vnodeops that wouldn't
2816  * make sense to the callers of this routine.
2817  */
2818 vnodeops_t *
2819 vn_getops(vnode_t *vp)
2820 {
2821 	vnodeops_t	*op;
2822 
2823 	ASSERT(vp != NULL);
2824 
2825 	op = vp->v_op;
2826 	membar_consumer();
2827 	if (vp->v_femhead == NULL && op == vp->v_op) {
2828 		return (op);
2829 	} else {
2830 		return (fem_getvnops(vp));
2831 	}
2832 }
2833 
2834 /*
2835  * Returns non-zero (1) if the vnodeops matches that of the vnode.
2836  * Returns zero (0) if not.
2837  */
2838 int
2839 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops)
2840 {
2841 	return (vn_getops(vp) == vnodeops);
2842 }
2843 
2844 /*
2845  * Returns non-zero (1) if the specified operation matches the
2846  * corresponding operation for that the vnode.
2847  * Returns zero (0) if not.
2848  */
2849 
2850 #define	MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0))
2851 
2852 int
2853 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp)
2854 {
2855 	const fs_operation_trans_def_t *otdp;
2856 	fs_generic_func_p *loc = NULL;
2857 	vnodeops_t	*vop = vn_getops(vp);
2858 
2859 	ASSERT(vopname != NULL);
2860 
2861 	for (otdp = vn_ops_table; otdp->name != NULL; otdp++) {
2862 		if (MATCHNAME(otdp->name, vopname)) {
2863 			loc = (fs_generic_func_p *)
2864 			    ((char *)(vop) + otdp->offset);
2865 			break;
2866 		}
2867 	}
2868 
2869 	return ((loc != NULL) && (*loc == funcp));
2870 }
2871 
2872 /*
2873  * fs_new_caller_id() needs to return a unique ID on a given local system.
2874  * The IDs do not need to survive across reboots.  These are primarily
2875  * used so that (FEM) monitors can detect particular callers (such as
2876  * the NFS server) to a given vnode/vfs operation.
2877  */
2878 u_longlong_t
2879 fs_new_caller_id()
2880 {
2881 	static uint64_t next_caller_id = 0LL; /* First call returns 1 */
2882 
2883 	return ((u_longlong_t)atomic_add_64_nv(&next_caller_id, 1));
2884 }
2885 
2886 /*
2887  * Given a starting vnode and a path, updates the path in the target vnode in
2888  * a safe manner.  If the vnode already has path information embedded, then the
2889  * cached path is left untouched.
2890  */
2891 
2892 size_t max_vnode_path = 4 * MAXPATHLEN;
2893 
2894 void
2895 vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
2896     const char *path, size_t plen)
2897 {
2898 	char	*rpath;
2899 	vnode_t	*base;
2900 	size_t	rpathlen, rpathalloc;
2901 	int	doslash = 1;
2902 
2903 	if (*path == '/') {
2904 		base = rootvp;
2905 		path++;
2906 		plen--;
2907 	} else {
2908 		base = startvp;
2909 	}
2910 
2911 	/*
2912 	 * We cannot grab base->v_lock while we hold vp->v_lock because of
2913 	 * the potential for deadlock.
2914 	 */
2915 	mutex_enter(&base->v_lock);
2916 	if (base->v_path == NULL) {
2917 		mutex_exit(&base->v_lock);
2918 		return;
2919 	}
2920 
2921 	rpathlen = strlen(base->v_path);
2922 	rpathalloc = rpathlen + plen + 1;
2923 	/* Avoid adding a slash if there's already one there */
2924 	if (base->v_path[rpathlen-1] == '/')
2925 		doslash = 0;
2926 	else
2927 		rpathalloc++;
2928 
2929 	/*
2930 	 * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held,
2931 	 * so we must do this dance.  If, by chance, something changes the path,
2932 	 * just give up since there is no real harm.
2933 	 */
2934 	mutex_exit(&base->v_lock);
2935 
2936 	/* Paths should stay within reason */
2937 	if (rpathalloc > max_vnode_path)
2938 		return;
2939 
2940 	rpath = kmem_alloc(rpathalloc, KM_SLEEP);
2941 
2942 	mutex_enter(&base->v_lock);
2943 	if (base->v_path == NULL || strlen(base->v_path) != rpathlen) {
2944 		mutex_exit(&base->v_lock);
2945 		kmem_free(rpath, rpathalloc);
2946 		return;
2947 	}
2948 	bcopy(base->v_path, rpath, rpathlen);
2949 	mutex_exit(&base->v_lock);
2950 
2951 	if (doslash)
2952 		rpath[rpathlen++] = '/';
2953 	bcopy(path, rpath + rpathlen, plen);
2954 	rpath[rpathlen + plen] = '\0';
2955 
2956 	mutex_enter(&vp->v_lock);
2957 	if (vp->v_path != NULL) {
2958 		mutex_exit(&vp->v_lock);
2959 		kmem_free(rpath, rpathalloc);
2960 	} else {
2961 		vp->v_path = rpath;
2962 		mutex_exit(&vp->v_lock);
2963 	}
2964 }
2965 
2966 /*
2967  * Sets the path to the vnode to be the given string, regardless of current
2968  * context.  The string must be a complete path from rootdir.  This is only used
2969  * by fsop_root() for setting the path based on the mountpoint.
2970  */
2971 void
2972 vn_setpath_str(struct vnode *vp, const char *str, size_t len)
2973 {
2974 	char *buf = kmem_alloc(len + 1, KM_SLEEP);
2975 
2976 	mutex_enter(&vp->v_lock);
2977 	if (vp->v_path != NULL) {
2978 		mutex_exit(&vp->v_lock);
2979 		kmem_free(buf, len + 1);
2980 		return;
2981 	}
2982 
2983 	vp->v_path = buf;
2984 	bcopy(str, vp->v_path, len);
2985 	vp->v_path[len] = '\0';
2986 
2987 	mutex_exit(&vp->v_lock);
2988 }
2989 
2990 /*
2991  * Called from within filesystem's vop_rename() to handle renames once the
2992  * target vnode is available.
2993  */
2994 void
2995 vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len)
2996 {
2997 	char *tmp;
2998 
2999 	mutex_enter(&vp->v_lock);
3000 	tmp = vp->v_path;
3001 	vp->v_path = NULL;
3002 	mutex_exit(&vp->v_lock);
3003 	vn_setpath(rootdir, dvp, vp, nm, len);
3004 	if (tmp != NULL)
3005 		kmem_free(tmp, strlen(tmp) + 1);
3006 }
3007 
3008 /*
3009  * Similar to vn_setpath_str(), this function sets the path of the destination
3010  * vnode to the be the same as the source vnode.
3011  */
3012 void
3013 vn_copypath(struct vnode *src, struct vnode *dst)
3014 {
3015 	char *buf;
3016 	int alloc;
3017 
3018 	mutex_enter(&src->v_lock);
3019 	if (src->v_path == NULL) {
3020 		mutex_exit(&src->v_lock);
3021 		return;
3022 	}
3023 	alloc = strlen(src->v_path) + 1;
3024 
3025 	/* avoid kmem_alloc() with lock held */
3026 	mutex_exit(&src->v_lock);
3027 	buf = kmem_alloc(alloc, KM_SLEEP);
3028 	mutex_enter(&src->v_lock);
3029 	if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) {
3030 		mutex_exit(&src->v_lock);
3031 		kmem_free(buf, alloc);
3032 		return;
3033 	}
3034 	bcopy(src->v_path, buf, alloc);
3035 	mutex_exit(&src->v_lock);
3036 
3037 	mutex_enter(&dst->v_lock);
3038 	if (dst->v_path != NULL) {
3039 		mutex_exit(&dst->v_lock);
3040 		kmem_free(buf, alloc);
3041 		return;
3042 	}
3043 	dst->v_path = buf;
3044 	mutex_exit(&dst->v_lock);
3045 }
3046 
3047 /*
3048  * XXX Private interface for segvn routines that handle vnode
3049  * large page segments.
3050  *
3051  * return 1 if vp's file system VOP_PAGEIO() implementation
3052  * can be safely used instead of VOP_GETPAGE() for handling
3053  * pagefaults against regular non swap files. VOP_PAGEIO()
3054  * interface is considered safe here if its implementation
3055  * is very close to VOP_GETPAGE() implementation.
3056  * e.g. It zero's out the part of the page beyond EOF. Doesn't
3057  * panic if there're file holes but instead returns an error.
3058  * Doesn't assume file won't be changed by user writes, etc.
3059  *
3060  * return 0 otherwise.
3061  *
3062  * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs.
3063  */
3064 int
3065 vn_vmpss_usepageio(vnode_t *vp)
3066 {
3067 	vfs_t   *vfsp = vp->v_vfsp;
3068 	char *fsname = vfssw[vfsp->vfs_fstype].vsw_name;
3069 	char *pageio_ok_fss[] = {"ufs", "nfs", NULL};
3070 	char **fsok = pageio_ok_fss;
3071 
3072 	if (fsname == NULL) {
3073 		return (0);
3074 	}
3075 
3076 	for (; *fsok; fsok++) {
3077 		if (strcmp(*fsok, fsname) == 0) {
3078 			return (1);
3079 		}
3080 	}
3081 	return (0);
3082 }
3083 
3084 /* VOP_XXX() macros call the corresponding fop_xxx() function */
3085 
3086 int
3087 fop_open(
3088 	vnode_t **vpp,
3089 	int mode,
3090 	cred_t *cr,
3091 	caller_context_t *ct)
3092 {
3093 	int ret;
3094 	vnode_t *vp = *vpp;
3095 
3096 	VN_HOLD(vp);
3097 	/*
3098 	 * Adding to the vnode counts before calling open
3099 	 * avoids the need for a mutex. It circumvents a race
3100 	 * condition where a query made on the vnode counts results in a
3101 	 * false negative. The inquirer goes away believing the file is
3102 	 * not open when there is an open on the file already under way.
3103 	 *
3104 	 * The counts are meant to prevent NFS from granting a delegation
3105 	 * when it would be dangerous to do so.
3106 	 *
3107 	 * The vnode counts are only kept on regular files
3108 	 */
3109 	if ((*vpp)->v_type == VREG) {
3110 		if (mode & FREAD)
3111 			atomic_add_32(&((*vpp)->v_rdcnt), 1);
3112 		if (mode & FWRITE)
3113 			atomic_add_32(&((*vpp)->v_wrcnt), 1);
3114 	}
3115 
3116 	VOPXID_MAP_CR(vp, cr);
3117 
3118 	ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct);
3119 
3120 	if (ret) {
3121 		/*
3122 		 * Use the saved vp just in case the vnode ptr got trashed
3123 		 * by the error.
3124 		 */
3125 		VOPSTATS_UPDATE(vp, open);
3126 		if ((vp->v_type == VREG) && (mode & FREAD))
3127 			atomic_add_32(&(vp->v_rdcnt), -1);
3128 		if ((vp->v_type == VREG) && (mode & FWRITE))
3129 			atomic_add_32(&(vp->v_wrcnt), -1);
3130 	} else {
3131 		/*
3132 		 * Some filesystems will return a different vnode,
3133 		 * but the same path was still used to open it.
3134 		 * So if we do change the vnode and need to
3135 		 * copy over the path, do so here, rather than special
3136 		 * casing each filesystem. Adjust the vnode counts to
3137 		 * reflect the vnode switch.
3138 		 */
3139 		VOPSTATS_UPDATE(*vpp, open);
3140 		if (*vpp != vp && *vpp != NULL) {
3141 			vn_copypath(vp, *vpp);
3142 			if (((*vpp)->v_type == VREG) && (mode & FREAD))
3143 				atomic_add_32(&((*vpp)->v_rdcnt), 1);
3144 			if ((vp->v_type == VREG) && (mode & FREAD))
3145 				atomic_add_32(&(vp->v_rdcnt), -1);
3146 			if (((*vpp)->v_type == VREG) && (mode & FWRITE))
3147 				atomic_add_32(&((*vpp)->v_wrcnt), 1);
3148 			if ((vp->v_type == VREG) && (mode & FWRITE))
3149 				atomic_add_32(&(vp->v_wrcnt), -1);
3150 		}
3151 	}
3152 	VN_RELE(vp);
3153 	return (ret);
3154 }
3155 
3156 int
3157 fop_close(
3158 	vnode_t *vp,
3159 	int flag,
3160 	int count,
3161 	offset_t offset,
3162 	cred_t *cr,
3163 	caller_context_t *ct)
3164 {
3165 	int err;
3166 
3167 	VOPXID_MAP_CR(vp, cr);
3168 
3169 	err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct);
3170 	VOPSTATS_UPDATE(vp, close);
3171 	/*
3172 	 * Check passed in count to handle possible dups. Vnode counts are only
3173 	 * kept on regular files
3174 	 */
3175 	if ((vp->v_type == VREG) && (count == 1))  {
3176 		if (flag & FREAD) {
3177 			ASSERT(vp->v_rdcnt > 0);
3178 			atomic_add_32(&(vp->v_rdcnt), -1);
3179 		}
3180 		if (flag & FWRITE) {
3181 			ASSERT(vp->v_wrcnt > 0);
3182 			atomic_add_32(&(vp->v_wrcnt), -1);
3183 		}
3184 	}
3185 	return (err);
3186 }
3187 
3188 int
3189 fop_read(
3190 	vnode_t *vp,
3191 	uio_t *uiop,
3192 	int ioflag,
3193 	cred_t *cr,
3194 	caller_context_t *ct)
3195 {
3196 	int	err;
3197 	ssize_t	resid_start = uiop->uio_resid;
3198 
3199 	VOPXID_MAP_CR(vp, cr);
3200 
3201 	err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
3202 	VOPSTATS_UPDATE_IO(vp, read,
3203 	    read_bytes, (resid_start - uiop->uio_resid));
3204 	return (err);
3205 }
3206 
3207 int
3208 fop_write(
3209 	vnode_t *vp,
3210 	uio_t *uiop,
3211 	int ioflag,
3212 	cred_t *cr,
3213 	caller_context_t *ct)
3214 {
3215 	int	err;
3216 	ssize_t	resid_start = uiop->uio_resid;
3217 
3218 	VOPXID_MAP_CR(vp, cr);
3219 
3220 	err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
3221 	VOPSTATS_UPDATE_IO(vp, write,
3222 	    write_bytes, (resid_start - uiop->uio_resid));
3223 	return (err);
3224 }
3225 
3226 int
3227 fop_ioctl(
3228 	vnode_t *vp,
3229 	int cmd,
3230 	intptr_t arg,
3231 	int flag,
3232 	cred_t *cr,
3233 	int *rvalp,
3234 	caller_context_t *ct)
3235 {
3236 	int	err;
3237 
3238 	VOPXID_MAP_CR(vp, cr);
3239 
3240 	err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct);
3241 	VOPSTATS_UPDATE(vp, ioctl);
3242 	return (err);
3243 }
3244 
3245 int
3246 fop_setfl(
3247 	vnode_t *vp,
3248 	int oflags,
3249 	int nflags,
3250 	cred_t *cr,
3251 	caller_context_t *ct)
3252 {
3253 	int	err;
3254 
3255 	VOPXID_MAP_CR(vp, cr);
3256 
3257 	err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct);
3258 	VOPSTATS_UPDATE(vp, setfl);
3259 	return (err);
3260 }
3261 
3262 int
3263 fop_getattr(
3264 	vnode_t *vp,
3265 	vattr_t *vap,
3266 	int flags,
3267 	cred_t *cr,
3268 	caller_context_t *ct)
3269 {
3270 	int	err;
3271 
3272 	VOPXID_MAP_CR(vp, cr);
3273 
3274 	/*
3275 	 * If this file system doesn't understand the xvattr extensions
3276 	 * then turn off the xvattr bit.
3277 	 */
3278 	if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3279 		vap->va_mask &= ~AT_XVATTR;
3280 	}
3281 
3282 	/*
3283 	 * We're only allowed to skip the ACL check iff we used a 32 bit
3284 	 * ACE mask with VOP_ACCESS() to determine permissions.
3285 	 */
3286 	if ((flags & ATTR_NOACLCHECK) &&
3287 	    vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3288 		return (EINVAL);
3289 	}
3290 	err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct);
3291 	VOPSTATS_UPDATE(vp, getattr);
3292 	return (err);
3293 }
3294 
3295 int
3296 fop_setattr(
3297 	vnode_t *vp,
3298 	vattr_t *vap,
3299 	int flags,
3300 	cred_t *cr,
3301 	caller_context_t *ct)
3302 {
3303 	int	err;
3304 
3305 	VOPXID_MAP_CR(vp, cr);
3306 
3307 	/*
3308 	 * If this file system doesn't understand the xvattr extensions
3309 	 * then turn off the xvattr bit.
3310 	 */
3311 	if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3312 		vap->va_mask &= ~AT_XVATTR;
3313 	}
3314 
3315 	/*
3316 	 * We're only allowed to skip the ACL check iff we used a 32 bit
3317 	 * ACE mask with VOP_ACCESS() to determine permissions.
3318 	 */
3319 	if ((flags & ATTR_NOACLCHECK) &&
3320 	    vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3321 		return (EINVAL);
3322 	}
3323 	err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct);
3324 	VOPSTATS_UPDATE(vp, setattr);
3325 	return (err);
3326 }
3327 
3328 int
3329 fop_access(
3330 	vnode_t *vp,
3331 	int mode,
3332 	int flags,
3333 	cred_t *cr,
3334 	caller_context_t *ct)
3335 {
3336 	int	err;
3337 
3338 	if ((flags & V_ACE_MASK) &&
3339 	    vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3340 		return (EINVAL);
3341 	}
3342 
3343 	VOPXID_MAP_CR(vp, cr);
3344 
3345 	err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct);
3346 	VOPSTATS_UPDATE(vp, access);
3347 	return (err);
3348 }
3349 
3350 int
3351 fop_lookup(
3352 	vnode_t *dvp,
3353 	char *nm,
3354 	vnode_t **vpp,
3355 	pathname_t *pnp,
3356 	int flags,
3357 	vnode_t *rdir,
3358 	cred_t *cr,
3359 	caller_context_t *ct,
3360 	int *deflags,		/* Returned per-dirent flags */
3361 	pathname_t *ppnp)	/* Returned case-preserved name in directory */
3362 {
3363 	int ret;
3364 
3365 	/*
3366 	 * If this file system doesn't support case-insensitive access
3367 	 * and said access is requested, fail quickly.  It is required
3368 	 * that if the vfs supports case-insensitive lookup, it also
3369 	 * supports extended dirent flags.
3370 	 */
3371 	if (flags & FIGNORECASE &&
3372 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3373 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3374 		return (EINVAL);
3375 
3376 	VOPXID_MAP_CR(dvp, cr);
3377 
3378 	if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) {
3379 		ret = xattr_dir_lookup(dvp, vpp, flags, cr);
3380 	} else {
3381 		ret = (*(dvp)->v_op->vop_lookup)
3382 		    (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp);
3383 	}
3384 	if (ret == 0 && *vpp) {
3385 		VOPSTATS_UPDATE(*vpp, lookup);
3386 		if ((*vpp)->v_path == NULL) {
3387 			vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm));
3388 		}
3389 	}
3390 
3391 	return (ret);
3392 }
3393 
3394 int
3395 fop_create(
3396 	vnode_t *dvp,
3397 	char *name,
3398 	vattr_t *vap,
3399 	vcexcl_t excl,
3400 	int mode,
3401 	vnode_t **vpp,
3402 	cred_t *cr,
3403 	int flags,
3404 	caller_context_t *ct,
3405 	vsecattr_t *vsecp)	/* ACL to set during create */
3406 {
3407 	int ret;
3408 
3409 	if (vsecp != NULL &&
3410 	    vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3411 		return (EINVAL);
3412 	}
3413 	/*
3414 	 * If this file system doesn't support case-insensitive access
3415 	 * and said access is requested, fail quickly.
3416 	 */
3417 	if (flags & FIGNORECASE &&
3418 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3419 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3420 		return (EINVAL);
3421 
3422 	VOPXID_MAP_CR(dvp, cr);
3423 
3424 	ret = (*(dvp)->v_op->vop_create)
3425 	    (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
3426 	if (ret == 0 && *vpp) {
3427 		VOPSTATS_UPDATE(*vpp, create);
3428 		if ((*vpp)->v_path == NULL) {
3429 			vn_setpath(rootdir, dvp, *vpp, name, strlen(name));
3430 		}
3431 	}
3432 
3433 	return (ret);
3434 }
3435 
3436 int
3437 fop_remove(
3438 	vnode_t *dvp,
3439 	char *nm,
3440 	cred_t *cr,
3441 	caller_context_t *ct,
3442 	int flags)
3443 {
3444 	int	err;
3445 
3446 	/*
3447 	 * If this file system doesn't support case-insensitive access
3448 	 * and said access is requested, fail quickly.
3449 	 */
3450 	if (flags & FIGNORECASE &&
3451 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3452 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3453 		return (EINVAL);
3454 
3455 	VOPXID_MAP_CR(dvp, cr);
3456 
3457 	err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags);
3458 	VOPSTATS_UPDATE(dvp, remove);
3459 	return (err);
3460 }
3461 
3462 int
3463 fop_link(
3464 	vnode_t *tdvp,
3465 	vnode_t *svp,
3466 	char *tnm,
3467 	cred_t *cr,
3468 	caller_context_t *ct,
3469 	int flags)
3470 {
3471 	int	err;
3472 
3473 	/*
3474 	 * If the target file system doesn't support case-insensitive access
3475 	 * and said access is requested, fail quickly.
3476 	 */
3477 	if (flags & FIGNORECASE &&
3478 	    (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3479 	    vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3480 		return (EINVAL);
3481 
3482 	VOPXID_MAP_CR(tdvp, cr);
3483 
3484 	err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags);
3485 	VOPSTATS_UPDATE(tdvp, link);
3486 	return (err);
3487 }
3488 
3489 int
3490 fop_rename(
3491 	vnode_t *sdvp,
3492 	char *snm,
3493 	vnode_t *tdvp,
3494 	char *tnm,
3495 	cred_t *cr,
3496 	caller_context_t *ct,
3497 	int flags)
3498 {
3499 	int	err;
3500 
3501 	/*
3502 	 * If the file system involved does not support
3503 	 * case-insensitive access and said access is requested, fail
3504 	 * quickly.
3505 	 */
3506 	if (flags & FIGNORECASE &&
3507 	    ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3508 	    vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)))
3509 		return (EINVAL);
3510 
3511 	VOPXID_MAP_CR(tdvp, cr);
3512 
3513 	err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags);
3514 	VOPSTATS_UPDATE(sdvp, rename);
3515 	return (err);
3516 }
3517 
3518 int
3519 fop_mkdir(
3520 	vnode_t *dvp,
3521 	char *dirname,
3522 	vattr_t *vap,
3523 	vnode_t **vpp,
3524 	cred_t *cr,
3525 	caller_context_t *ct,
3526 	int flags,
3527 	vsecattr_t *vsecp)	/* ACL to set during create */
3528 {
3529 	int ret;
3530 
3531 	if (vsecp != NULL &&
3532 	    vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3533 		return (EINVAL);
3534 	}
3535 	/*
3536 	 * If this file system doesn't support case-insensitive access
3537 	 * and said access is requested, fail quickly.
3538 	 */
3539 	if (flags & FIGNORECASE &&
3540 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3541 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3542 		return (EINVAL);
3543 
3544 	VOPXID_MAP_CR(dvp, cr);
3545 
3546 	ret = (*(dvp)->v_op->vop_mkdir)
3547 	    (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
3548 	if (ret == 0 && *vpp) {
3549 		VOPSTATS_UPDATE(*vpp, mkdir);
3550 		if ((*vpp)->v_path == NULL) {
3551 			vn_setpath(rootdir, dvp, *vpp, dirname,
3552 			    strlen(dirname));
3553 		}
3554 	}
3555 
3556 	return (ret);
3557 }
3558 
3559 int
3560 fop_rmdir(
3561 	vnode_t *dvp,
3562 	char *nm,
3563 	vnode_t *cdir,
3564 	cred_t *cr,
3565 	caller_context_t *ct,
3566 	int flags)
3567 {
3568 	int	err;
3569 
3570 	/*
3571 	 * If this file system doesn't support case-insensitive access
3572 	 * and said access is requested, fail quickly.
3573 	 */
3574 	if (flags & FIGNORECASE &&
3575 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3576 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3577 		return (EINVAL);
3578 
3579 	VOPXID_MAP_CR(dvp, cr);
3580 
3581 	err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags);
3582 	VOPSTATS_UPDATE(dvp, rmdir);
3583 	return (err);
3584 }
3585 
3586 int
3587 fop_readdir(
3588 	vnode_t *vp,
3589 	uio_t *uiop,
3590 	cred_t *cr,
3591 	int *eofp,
3592 	caller_context_t *ct,
3593 	int flags)
3594 {
3595 	int	err;
3596 	ssize_t	resid_start = uiop->uio_resid;
3597 
3598 	/*
3599 	 * If this file system doesn't support retrieving directory
3600 	 * entry flags and said access is requested, fail quickly.
3601 	 */
3602 	if (flags & V_RDDIR_ENTFLAGS &&
3603 	    vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0)
3604 		return (EINVAL);
3605 
3606 	VOPXID_MAP_CR(vp, cr);
3607 
3608 	err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags);
3609 	VOPSTATS_UPDATE_IO(vp, readdir,
3610 	    readdir_bytes, (resid_start - uiop->uio_resid));
3611 	return (err);
3612 }
3613 
3614 int
3615 fop_symlink(
3616 	vnode_t *dvp,
3617 	char *linkname,
3618 	vattr_t *vap,
3619 	char *target,
3620 	cred_t *cr,
3621 	caller_context_t *ct,
3622 	int flags)
3623 {
3624 	int	err;
3625 	xvattr_t xvattr;
3626 
3627 	/*
3628 	 * If this file system doesn't support case-insensitive access
3629 	 * and said access is requested, fail quickly.
3630 	 */
3631 	if (flags & FIGNORECASE &&
3632 	    (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3633 	    vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3634 		return (EINVAL);
3635 
3636 	VOPXID_MAP_CR(dvp, cr);
3637 
3638 	/* check for reparse point */
3639 	if ((vfs_has_feature(dvp->v_vfsp, VFSFT_REPARSE)) &&
3640 	    (strncmp(target, FS_REPARSE_TAG_STR,
3641 	    strlen(FS_REPARSE_TAG_STR)) == 0)) {
3642 		if (!fs_reparse_mark(target, vap, &xvattr))
3643 			vap = (vattr_t *)&xvattr;
3644 	}
3645 
3646 	err = (*(dvp)->v_op->vop_symlink)
3647 	    (dvp, linkname, vap, target, cr, ct, flags);
3648 	VOPSTATS_UPDATE(dvp, symlink);
3649 	return (err);
3650 }
3651 
3652 int
3653 fop_readlink(
3654 	vnode_t *vp,
3655 	uio_t *uiop,
3656 	cred_t *cr,
3657 	caller_context_t *ct)
3658 {
3659 	int	err;
3660 
3661 	VOPXID_MAP_CR(vp, cr);
3662 
3663 	err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct);
3664 	VOPSTATS_UPDATE(vp, readlink);
3665 	return (err);
3666 }
3667 
3668 int
3669 fop_fsync(
3670 	vnode_t *vp,
3671 	int syncflag,
3672 	cred_t *cr,
3673 	caller_context_t *ct)
3674 {
3675 	int	err;
3676 
3677 	VOPXID_MAP_CR(vp, cr);
3678 
3679 	err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct);
3680 	VOPSTATS_UPDATE(vp, fsync);
3681 	return (err);
3682 }
3683 
3684 void
3685 fop_inactive(
3686 	vnode_t *vp,
3687 	cred_t *cr,
3688 	caller_context_t *ct)
3689 {
3690 	/* Need to update stats before vop call since we may lose the vnode */
3691 	VOPSTATS_UPDATE(vp, inactive);
3692 
3693 	VOPXID_MAP_CR(vp, cr);
3694 
3695 	(*(vp)->v_op->vop_inactive)(vp, cr, ct);
3696 }
3697 
3698 int
3699 fop_fid(
3700 	vnode_t *vp,
3701 	fid_t *fidp,
3702 	caller_context_t *ct)
3703 {
3704 	int	err;
3705 
3706 	err = (*(vp)->v_op->vop_fid)(vp, fidp, ct);
3707 	VOPSTATS_UPDATE(vp, fid);
3708 	return (err);
3709 }
3710 
3711 int
3712 fop_rwlock(
3713 	vnode_t *vp,
3714 	int write_lock,
3715 	caller_context_t *ct)
3716 {
3717 	int	ret;
3718 
3719 	ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct));
3720 	VOPSTATS_UPDATE(vp, rwlock);
3721 	return (ret);
3722 }
3723 
3724 void
3725 fop_rwunlock(
3726 	vnode_t *vp,
3727 	int write_lock,
3728 	caller_context_t *ct)
3729 {
3730 	(*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct);
3731 	VOPSTATS_UPDATE(vp, rwunlock);
3732 }
3733 
3734 int
3735 fop_seek(
3736 	vnode_t *vp,
3737 	offset_t ooff,
3738 	offset_t *noffp,
3739 	caller_context_t *ct)
3740 {
3741 	int	err;
3742 
3743 	err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct);
3744 	VOPSTATS_UPDATE(vp, seek);
3745 	return (err);
3746 }
3747 
3748 int
3749 fop_cmp(
3750 	vnode_t *vp1,
3751 	vnode_t *vp2,
3752 	caller_context_t *ct)
3753 {
3754 	int	err;
3755 
3756 	err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct);
3757 	VOPSTATS_UPDATE(vp1, cmp);
3758 	return (err);
3759 }
3760 
3761 int
3762 fop_frlock(
3763 	vnode_t *vp,
3764 	int cmd,
3765 	flock64_t *bfp,
3766 	int flag,
3767 	offset_t offset,
3768 	struct flk_callback *flk_cbp,
3769 	cred_t *cr,
3770 	caller_context_t *ct)
3771 {
3772 	int	err;
3773 
3774 	VOPXID_MAP_CR(vp, cr);
3775 
3776 	err = (*(vp)->v_op->vop_frlock)
3777 	    (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3778 	VOPSTATS_UPDATE(vp, frlock);
3779 	return (err);
3780 }
3781 
3782 int
3783 fop_space(
3784 	vnode_t *vp,
3785 	int cmd,
3786 	flock64_t *bfp,
3787 	int flag,
3788 	offset_t offset,
3789 	cred_t *cr,
3790 	caller_context_t *ct)
3791 {
3792 	int	err;
3793 
3794 	VOPXID_MAP_CR(vp, cr);
3795 
3796 	err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct);
3797 	VOPSTATS_UPDATE(vp, space);
3798 	return (err);
3799 }
3800 
3801 int
3802 fop_realvp(
3803 	vnode_t *vp,
3804 	vnode_t **vpp,
3805 	caller_context_t *ct)
3806 {
3807 	int	err;
3808 
3809 	err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct);
3810 	VOPSTATS_UPDATE(vp, realvp);
3811 	return (err);
3812 }
3813 
3814 int
3815 fop_getpage(
3816 	vnode_t *vp,
3817 	offset_t off,
3818 	size_t len,
3819 	uint_t *protp,
3820 	page_t **plarr,
3821 	size_t plsz,
3822 	struct seg *seg,
3823 	caddr_t addr,
3824 	enum seg_rw rw,
3825 	cred_t *cr,
3826 	caller_context_t *ct)
3827 {
3828 	int	err;
3829 
3830 	VOPXID_MAP_CR(vp, cr);
3831 
3832 	err = (*(vp)->v_op->vop_getpage)
3833 	    (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct);
3834 	VOPSTATS_UPDATE(vp, getpage);
3835 	return (err);
3836 }
3837 
3838 int
3839 fop_putpage(
3840 	vnode_t *vp,
3841 	offset_t off,
3842 	size_t len,
3843 	int flags,
3844 	cred_t *cr,
3845 	caller_context_t *ct)
3846 {
3847 	int	err;
3848 
3849 	VOPXID_MAP_CR(vp, cr);
3850 
3851 	err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct);
3852 	VOPSTATS_UPDATE(vp, putpage);
3853 	return (err);
3854 }
3855 
3856 int
3857 fop_map(
3858 	vnode_t *vp,
3859 	offset_t off,
3860 	struct as *as,
3861 	caddr_t *addrp,
3862 	size_t len,
3863 	uchar_t prot,
3864 	uchar_t maxprot,
3865 	uint_t flags,
3866 	cred_t *cr,
3867 	caller_context_t *ct)
3868 {
3869 	int	err;
3870 
3871 	VOPXID_MAP_CR(vp, cr);
3872 
3873 	err = (*(vp)->v_op->vop_map)
3874 	    (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct);
3875 	VOPSTATS_UPDATE(vp, map);
3876 	return (err);
3877 }
3878 
3879 int
3880 fop_addmap(
3881 	vnode_t *vp,
3882 	offset_t off,
3883 	struct as *as,
3884 	caddr_t addr,
3885 	size_t len,
3886 	uchar_t prot,
3887 	uchar_t maxprot,
3888 	uint_t flags,
3889 	cred_t *cr,
3890 	caller_context_t *ct)
3891 {
3892 	int error;
3893 	u_longlong_t delta;
3894 
3895 	VOPXID_MAP_CR(vp, cr);
3896 
3897 	error = (*(vp)->v_op->vop_addmap)
3898 	    (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3899 
3900 	if ((!error) && (vp->v_type == VREG)) {
3901 		delta = (u_longlong_t)btopr(len);
3902 		/*
3903 		 * If file is declared MAP_PRIVATE, it can't be written back
3904 		 * even if open for write. Handle as read.
3905 		 */
3906 		if (flags & MAP_PRIVATE) {
3907 			atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3908 			    (int64_t)delta);
3909 		} else {
3910 			/*
3911 			 * atomic_add_64 forces the fetch of a 64 bit value to
3912 			 * be atomic on 32 bit machines
3913 			 */
3914 			if (maxprot & PROT_WRITE)
3915 				atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
3916 				    (int64_t)delta);
3917 			if (maxprot & PROT_READ)
3918 				atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3919 				    (int64_t)delta);
3920 			if (maxprot & PROT_EXEC)
3921 				atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3922 				    (int64_t)delta);
3923 		}
3924 	}
3925 	VOPSTATS_UPDATE(vp, addmap);
3926 	return (error);
3927 }
3928 
3929 int
3930 fop_delmap(
3931 	vnode_t *vp,
3932 	offset_t off,
3933 	struct as *as,
3934 	caddr_t addr,
3935 	size_t len,
3936 	uint_t prot,
3937 	uint_t maxprot,
3938 	uint_t flags,
3939 	cred_t *cr,
3940 	caller_context_t *ct)
3941 {
3942 	int error;
3943 	u_longlong_t delta;
3944 
3945 	VOPXID_MAP_CR(vp, cr);
3946 
3947 	error = (*(vp)->v_op->vop_delmap)
3948 	    (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3949 
3950 	/*
3951 	 * NFS calls into delmap twice, the first time
3952 	 * it simply establishes a callback mechanism and returns EAGAIN
3953 	 * while the real work is being done upon the second invocation.
3954 	 * We have to detect this here and only decrement the counts upon
3955 	 * the second delmap request.
3956 	 */
3957 	if ((error != EAGAIN) && (vp->v_type == VREG)) {
3958 
3959 		delta = (u_longlong_t)btopr(len);
3960 
3961 		if (flags & MAP_PRIVATE) {
3962 			atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3963 			    (int64_t)(-delta));
3964 		} else {
3965 			/*
3966 			 * atomic_add_64 forces the fetch of a 64 bit value
3967 			 * to be atomic on 32 bit machines
3968 			 */
3969 			if (maxprot & PROT_WRITE)
3970 				atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
3971 				    (int64_t)(-delta));
3972 			if (maxprot & PROT_READ)
3973 				atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3974 				    (int64_t)(-delta));
3975 			if (maxprot & PROT_EXEC)
3976 				atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3977 				    (int64_t)(-delta));
3978 		}
3979 	}
3980 	VOPSTATS_UPDATE(vp, delmap);
3981 	return (error);
3982 }
3983 
3984 
3985 int
3986 fop_poll(
3987 	vnode_t *vp,
3988 	short events,
3989 	int anyyet,
3990 	short *reventsp,
3991 	struct pollhead **phpp,
3992 	caller_context_t *ct)
3993 {
3994 	int	err;
3995 
3996 	err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct);
3997 	VOPSTATS_UPDATE(vp, poll);
3998 	return (err);
3999 }
4000 
4001 int
4002 fop_dump(
4003 	vnode_t *vp,
4004 	caddr_t addr,
4005 	offset_t lbdn,
4006 	offset_t dblks,
4007 	caller_context_t *ct)
4008 {
4009 	int	err;
4010 
4011 	/* ensure lbdn and dblks can be passed safely to bdev_dump */
4012 	if ((lbdn != (daddr_t)lbdn) || (dblks != (int)dblks))
4013 		return (EIO);
4014 
4015 	err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct);
4016 	VOPSTATS_UPDATE(vp, dump);
4017 	return (err);
4018 }
4019 
4020 int
4021 fop_pathconf(
4022 	vnode_t *vp,
4023 	int cmd,
4024 	ulong_t *valp,
4025 	cred_t *cr,
4026 	caller_context_t *ct)
4027 {
4028 	int	err;
4029 
4030 	VOPXID_MAP_CR(vp, cr);
4031 
4032 	err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct);
4033 	VOPSTATS_UPDATE(vp, pathconf);
4034 	return (err);
4035 }
4036 
4037 int
4038 fop_pageio(
4039 	vnode_t *vp,
4040 	struct page *pp,
4041 	u_offset_t io_off,
4042 	size_t io_len,
4043 	int flags,
4044 	cred_t *cr,
4045 	caller_context_t *ct)
4046 {
4047 	int	err;
4048 
4049 	VOPXID_MAP_CR(vp, cr);
4050 
4051 	err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct);
4052 	VOPSTATS_UPDATE(vp, pageio);
4053 	return (err);
4054 }
4055 
4056 int
4057 fop_dumpctl(
4058 	vnode_t *vp,
4059 	int action,
4060 	offset_t *blkp,
4061 	caller_context_t *ct)
4062 {
4063 	int	err;
4064 	err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct);
4065 	VOPSTATS_UPDATE(vp, dumpctl);
4066 	return (err);
4067 }
4068 
4069 void
4070 fop_dispose(
4071 	vnode_t *vp,
4072 	page_t *pp,
4073 	int flag,
4074 	int dn,
4075 	cred_t *cr,
4076 	caller_context_t *ct)
4077 {
4078 	/* Must do stats first since it's possible to lose the vnode */
4079 	VOPSTATS_UPDATE(vp, dispose);
4080 
4081 	VOPXID_MAP_CR(vp, cr);
4082 
4083 	(*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct);
4084 }
4085 
4086 int
4087 fop_setsecattr(
4088 	vnode_t *vp,
4089 	vsecattr_t *vsap,
4090 	int flag,
4091 	cred_t *cr,
4092 	caller_context_t *ct)
4093 {
4094 	int	err;
4095 
4096 	VOPXID_MAP_CR(vp, cr);
4097 
4098 	/*
4099 	 * We're only allowed to skip the ACL check iff we used a 32 bit
4100 	 * ACE mask with VOP_ACCESS() to determine permissions.
4101 	 */
4102 	if ((flag & ATTR_NOACLCHECK) &&
4103 	    vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4104 		return (EINVAL);
4105 	}
4106 	err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct);
4107 	VOPSTATS_UPDATE(vp, setsecattr);
4108 	return (err);
4109 }
4110 
4111 int
4112 fop_getsecattr(
4113 	vnode_t *vp,
4114 	vsecattr_t *vsap,
4115 	int flag,
4116 	cred_t *cr,
4117 	caller_context_t *ct)
4118 {
4119 	int	err;
4120 
4121 	/*
4122 	 * We're only allowed to skip the ACL check iff we used a 32 bit
4123 	 * ACE mask with VOP_ACCESS() to determine permissions.
4124 	 */
4125 	if ((flag & ATTR_NOACLCHECK) &&
4126 	    vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4127 		return (EINVAL);
4128 	}
4129 
4130 	VOPXID_MAP_CR(vp, cr);
4131 
4132 	err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct);
4133 	VOPSTATS_UPDATE(vp, getsecattr);
4134 	return (err);
4135 }
4136 
4137 int
4138 fop_shrlock(
4139 	vnode_t *vp,
4140 	int cmd,
4141 	struct shrlock *shr,
4142 	int flag,
4143 	cred_t *cr,
4144 	caller_context_t *ct)
4145 {
4146 	int	err;
4147 
4148 	VOPXID_MAP_CR(vp, cr);
4149 
4150 	err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct);
4151 	VOPSTATS_UPDATE(vp, shrlock);
4152 	return (err);
4153 }
4154 
4155 int
4156 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm,
4157     caller_context_t *ct)
4158 {
4159 	int	err;
4160 
4161 	err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct);
4162 	VOPSTATS_UPDATE(vp, vnevent);
4163 	return (err);
4164 }
4165 
4166 int
4167 fop_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *uiop, cred_t *cr,
4168     caller_context_t *ct)
4169 {
4170 	int err;
4171 
4172 	if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4173 		return (ENOTSUP);
4174 	err = (*(vp)->v_op->vop_reqzcbuf)(vp, ioflag, uiop, cr, ct);
4175 	VOPSTATS_UPDATE(vp, reqzcbuf);
4176 	return (err);
4177 }
4178 
4179 int
4180 fop_retzcbuf(vnode_t *vp, xuio_t *uiop, cred_t *cr, caller_context_t *ct)
4181 {
4182 	int err;
4183 
4184 	if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4185 		return (ENOTSUP);
4186 	err = (*(vp)->v_op->vop_retzcbuf)(vp, uiop, cr, ct);
4187 	VOPSTATS_UPDATE(vp, retzcbuf);
4188 	return (err);
4189 }
4190 
4191 /*
4192  * Default destructor
4193  *	Needed because NULL destructor means that the key is unused
4194  */
4195 /* ARGSUSED */
4196 void
4197 vsd_defaultdestructor(void *value)
4198 {}
4199 
4200 /*
4201  * Create a key (index into per vnode array)
4202  *	Locks out vsd_create, vsd_destroy, and vsd_free
4203  *	May allocate memory with lock held
4204  */
4205 void
4206 vsd_create(uint_t *keyp, void (*destructor)(void *))
4207 {
4208 	int	i;
4209 	uint_t	nkeys;
4210 
4211 	/*
4212 	 * if key is allocated, do nothing
4213 	 */
4214 	mutex_enter(&vsd_lock);
4215 	if (*keyp) {
4216 		mutex_exit(&vsd_lock);
4217 		return;
4218 	}
4219 	/*
4220 	 * find an unused key
4221 	 */
4222 	if (destructor == NULL)
4223 		destructor = vsd_defaultdestructor;
4224 
4225 	for (i = 0; i < vsd_nkeys; ++i)
4226 		if (vsd_destructor[i] == NULL)
4227 			break;
4228 
4229 	/*
4230 	 * if no unused keys, increase the size of the destructor array
4231 	 */
4232 	if (i == vsd_nkeys) {
4233 		if ((nkeys = (vsd_nkeys << 1)) == 0)
4234 			nkeys = 1;
4235 		vsd_destructor =
4236 		    (void (**)(void *))vsd_realloc((void *)vsd_destructor,
4237 		    (size_t)(vsd_nkeys * sizeof (void (*)(void *))),
4238 		    (size_t)(nkeys * sizeof (void (*)(void *))));
4239 		vsd_nkeys = nkeys;
4240 	}
4241 
4242 	/*
4243 	 * allocate the next available unused key
4244 	 */
4245 	vsd_destructor[i] = destructor;
4246 	*keyp = i + 1;
4247 
4248 	/* create vsd_list, if it doesn't exist */
4249 	if (vsd_list == NULL) {
4250 		vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
4251 		list_create(vsd_list, sizeof (struct vsd_node),
4252 		    offsetof(struct vsd_node, vs_nodes));
4253 	}
4254 
4255 	mutex_exit(&vsd_lock);
4256 }
4257 
4258 /*
4259  * Destroy a key
4260  *
4261  * Assumes that the caller is preventing vsd_set and vsd_get
4262  * Locks out vsd_create, vsd_destroy, and vsd_free
4263  * May free memory with lock held
4264  */
4265 void
4266 vsd_destroy(uint_t *keyp)
4267 {
4268 	uint_t key;
4269 	struct vsd_node *vsd;
4270 
4271 	/*
4272 	 * protect the key namespace and our destructor lists
4273 	 */
4274 	mutex_enter(&vsd_lock);
4275 	key = *keyp;
4276 	*keyp = 0;
4277 
4278 	ASSERT(key <= vsd_nkeys);
4279 
4280 	/*
4281 	 * if the key is valid
4282 	 */
4283 	if (key != 0) {
4284 		uint_t k = key - 1;
4285 		/*
4286 		 * for every vnode with VSD, call key's destructor
4287 		 */
4288 		for (vsd = list_head(vsd_list); vsd != NULL;
4289 		    vsd = list_next(vsd_list, vsd)) {
4290 			/*
4291 			 * no VSD for key in this vnode
4292 			 */
4293 			if (key > vsd->vs_nkeys)
4294 				continue;
4295 			/*
4296 			 * call destructor for key
4297 			 */
4298 			if (vsd->vs_value[k] && vsd_destructor[k])
4299 				(*vsd_destructor[k])(vsd->vs_value[k]);
4300 			/*
4301 			 * reset value for key
4302 			 */
4303 			vsd->vs_value[k] = NULL;
4304 		}
4305 		/*
4306 		 * actually free the key (NULL destructor == unused)
4307 		 */
4308 		vsd_destructor[k] = NULL;
4309 	}
4310 
4311 	mutex_exit(&vsd_lock);
4312 }
4313 
4314 /*
4315  * Quickly return the per vnode value that was stored with the specified key
4316  * Assumes the caller is protecting key from vsd_create and vsd_destroy
4317  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4318  */
4319 void *
4320 vsd_get(vnode_t *vp, uint_t key)
4321 {
4322 	struct vsd_node *vsd;
4323 
4324 	ASSERT(vp != NULL);
4325 	ASSERT(mutex_owned(&vp->v_vsd_lock));
4326 
4327 	vsd = vp->v_vsd;
4328 
4329 	if (key && vsd != NULL && key <= vsd->vs_nkeys)
4330 		return (vsd->vs_value[key - 1]);
4331 	return (NULL);
4332 }
4333 
4334 /*
4335  * Set a per vnode value indexed with the specified key
4336  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4337  */
4338 int
4339 vsd_set(vnode_t *vp, uint_t key, void *value)
4340 {
4341 	struct vsd_node *vsd;
4342 
4343 	ASSERT(vp != NULL);
4344 	ASSERT(mutex_owned(&vp->v_vsd_lock));
4345 
4346 	if (key == 0)
4347 		return (EINVAL);
4348 
4349 	vsd = vp->v_vsd;
4350 	if (vsd == NULL)
4351 		vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP);
4352 
4353 	/*
4354 	 * If the vsd was just allocated, vs_nkeys will be 0, so the following
4355 	 * code won't happen and we will continue down and allocate space for
4356 	 * the vs_value array.
4357 	 * If the caller is replacing one value with another, then it is up
4358 	 * to the caller to free/rele/destroy the previous value (if needed).
4359 	 */
4360 	if (key <= vsd->vs_nkeys) {
4361 		vsd->vs_value[key - 1] = value;
4362 		return (0);
4363 	}
4364 
4365 	ASSERT(key <= vsd_nkeys);
4366 
4367 	if (vsd->vs_nkeys == 0) {
4368 		mutex_enter(&vsd_lock);	/* lock out vsd_destroy() */
4369 		/*
4370 		 * Link onto list of all VSD nodes.
4371 		 */
4372 		list_insert_head(vsd_list, vsd);
4373 		mutex_exit(&vsd_lock);
4374 	}
4375 
4376 	/*
4377 	 * Allocate vnode local storage and set the value for key
4378 	 */
4379 	vsd->vs_value = vsd_realloc(vsd->vs_value,
4380 	    vsd->vs_nkeys * sizeof (void *),
4381 	    key * sizeof (void *));
4382 	vsd->vs_nkeys = key;
4383 	vsd->vs_value[key - 1] = value;
4384 
4385 	return (0);
4386 }
4387 
4388 /*
4389  * Called from vn_free() to run the destructor function for each vsd
4390  *	Locks out vsd_create and vsd_destroy
4391  *	Assumes that the destructor *DOES NOT* use vsd
4392  */
4393 void
4394 vsd_free(vnode_t *vp)
4395 {
4396 	int i;
4397 	struct vsd_node *vsd = vp->v_vsd;
4398 
4399 	if (vsd == NULL)
4400 		return;
4401 
4402 	if (vsd->vs_nkeys == 0) {
4403 		kmem_free(vsd, sizeof (*vsd));
4404 		vp->v_vsd = NULL;
4405 		return;
4406 	}
4407 
4408 	/*
4409 	 * lock out vsd_create and vsd_destroy, call
4410 	 * the destructor, and mark the value as destroyed.
4411 	 */
4412 	mutex_enter(&vsd_lock);
4413 
4414 	for (i = 0; i < vsd->vs_nkeys; i++) {
4415 		if (vsd->vs_value[i] && vsd_destructor[i])
4416 			(*vsd_destructor[i])(vsd->vs_value[i]);
4417 		vsd->vs_value[i] = NULL;
4418 	}
4419 
4420 	/*
4421 	 * remove from linked list of VSD nodes
4422 	 */
4423 	list_remove(vsd_list, vsd);
4424 
4425 	mutex_exit(&vsd_lock);
4426 
4427 	/*
4428 	 * free up the VSD
4429 	 */
4430 	kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *));
4431 	kmem_free(vsd, sizeof (struct vsd_node));
4432 	vp->v_vsd = NULL;
4433 }
4434 
4435 /*
4436  * realloc
4437  */
4438 static void *
4439 vsd_realloc(void *old, size_t osize, size_t nsize)
4440 {
4441 	void *new;
4442 
4443 	new = kmem_zalloc(nsize, KM_SLEEP);
4444 	if (old) {
4445 		bcopy(old, new, osize);
4446 		kmem_free(old, osize);
4447 	}
4448 	return (new);
4449 }
4450 
4451 /*
4452  * Setup the extensible system attribute for creating a reparse point.
4453  * The symlink data 'target' is validated for proper format of a reparse
4454  * string and a check also made to make sure the symlink data does not
4455  * point to an existing file.
4456  *
4457  * return 0 if ok else -1.
4458  */
4459 static int
4460 fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr)
4461 {
4462 	xoptattr_t *xoap;
4463 
4464 	if ((!target) || (!vap) || (!xvattr))
4465 		return (-1);
4466 
4467 	/* validate reparse string */
4468 	if (reparse_validate((const char *)target))
4469 		return (-1);
4470 
4471 	xva_init(xvattr);
4472 	xvattr->xva_vattr = *vap;
4473 	xvattr->xva_vattr.va_mask |= AT_XVATTR;
4474 	xoap = xva_getxoptattr(xvattr);
4475 	ASSERT(xoap);
4476 	XVA_SET_REQ(xvattr, XAT_REPARSE);
4477 	xoap->xoa_reparse = 1;
4478 
4479 	return (0);
4480 }
4481 
4482 /*
4483  * Function to check whether a symlink is a reparse point.
4484  * Return B_TRUE if it is a reparse point, else return B_FALSE
4485  */
4486 boolean_t
4487 vn_is_reparse(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4488 {
4489 	xvattr_t xvattr;
4490 	xoptattr_t *xoap;
4491 
4492 	if ((vp->v_type != VLNK) ||
4493 	    !(vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR)))
4494 		return (B_FALSE);
4495 
4496 	xva_init(&xvattr);
4497 	xoap = xva_getxoptattr(&xvattr);
4498 	ASSERT(xoap);
4499 	XVA_SET_REQ(&xvattr, XAT_REPARSE);
4500 
4501 	if (VOP_GETATTR(vp, &xvattr.xva_vattr, 0, cr, ct))
4502 		return (B_FALSE);
4503 
4504 	if ((!(xvattr.xva_vattr.va_mask & AT_XVATTR)) ||
4505 	    (!(XVA_ISSET_RTN(&xvattr, XAT_REPARSE))))
4506 		return (B_FALSE);
4507 
4508 	return (xoap->xoa_reparse ? B_TRUE : B_FALSE);
4509 }
4510