xref: /illumos-gate/usr/src/uts/common/sys/vnode.h (revision ade42b55)
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 /*
23  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
26  */
27 
28 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
29 /*	  All Rights Reserved  	*/
30 
31 /*
32  * University Copyright- Copyright (c) 1982, 1986, 1988
33  * The Regents of the University of California
34  * All Rights Reserved
35  *
36  * University Acknowledgment- Portions of this document are derived from
37  * software developed by the University of California, Berkeley, and its
38  * contributors.
39  */
40 
41 #ifndef _SYS_VNODE_H
42 #define	_SYS_VNODE_H
43 
44 #include <sys/types.h>
45 #include <sys/t_lock.h>
46 #include <sys/rwstlock.h>
47 #include <sys/time_impl.h>
48 #include <sys/cred.h>
49 #include <sys/uio.h>
50 #include <sys/resource.h>
51 #include <vm/seg_enum.h>
52 #include <sys/kstat.h>
53 #include <sys/kmem.h>
54 #include <sys/list.h>
55 #ifdef	_KERNEL
56 #include <sys/buf.h>
57 #include <sys/sdt.h>
58 #endif	/* _KERNEL */
59 
60 #ifdef	__cplusplus
61 extern "C" {
62 #endif
63 
64 /*
65  * Statistics for all vnode operations.
66  * All operations record number of ops (since boot/mount/zero'ed).
67  * Certain I/O operations (read, write, readdir) also record number
68  * of bytes transferred.
69  * This appears in two places in the system: one is embedded in each
70  * vfs_t.  There is also an array of vopstats_t structures allocated
71  * on a per-fstype basis.
72  */
73 
74 #define	VOPSTATS_STR	"vopstats_"	/* Initial string for vopstat kstats */
75 
76 typedef struct vopstats {
77 	kstat_named_t	nopen;		/* VOP_OPEN */
78 	kstat_named_t	nclose;		/* VOP_CLOSE */
79 	kstat_named_t	nread;		/* VOP_READ */
80 	kstat_named_t	read_bytes;
81 	kstat_named_t	nwrite;		/* VOP_WRITE */
82 	kstat_named_t	write_bytes;
83 	kstat_named_t	nioctl;		/* VOP_IOCTL */
84 	kstat_named_t	nsetfl;		/* VOP_SETFL */
85 	kstat_named_t	ngetattr;	/* VOP_GETATTR */
86 	kstat_named_t	nsetattr;	/* VOP_SETATTR */
87 	kstat_named_t	naccess;	/* VOP_ACCESS */
88 	kstat_named_t	nlookup;	/* VOP_LOOKUP */
89 	kstat_named_t	ncreate;	/* VOP_CREATE */
90 	kstat_named_t	nremove;	/* VOP_REMOVE */
91 	kstat_named_t	nlink;		/* VOP_LINK */
92 	kstat_named_t	nrename;	/* VOP_RENAME */
93 	kstat_named_t	nmkdir;		/* VOP_MKDIR */
94 	kstat_named_t	nrmdir;		/* VOP_RMDIR */
95 	kstat_named_t	nreaddir;	/* VOP_READDIR */
96 	kstat_named_t	readdir_bytes;
97 	kstat_named_t	nsymlink;	/* VOP_SYMLINK */
98 	kstat_named_t	nreadlink;	/* VOP_READLINK */
99 	kstat_named_t	nfsync;		/* VOP_FSYNC */
100 	kstat_named_t	ninactive;	/* VOP_INACTIVE */
101 	kstat_named_t	nfid;		/* VOP_FID */
102 	kstat_named_t	nrwlock;	/* VOP_RWLOCK */
103 	kstat_named_t	nrwunlock;	/* VOP_RWUNLOCK */
104 	kstat_named_t	nseek;		/* VOP_SEEK */
105 	kstat_named_t	ncmp;		/* VOP_CMP */
106 	kstat_named_t	nfrlock;	/* VOP_FRLOCK */
107 	kstat_named_t	nspace;		/* VOP_SPACE */
108 	kstat_named_t	nrealvp;	/* VOP_REALVP */
109 	kstat_named_t	ngetpage;	/* VOP_GETPAGE */
110 	kstat_named_t	nputpage;	/* VOP_PUTPAGE */
111 	kstat_named_t	nmap;		/* VOP_MAP */
112 	kstat_named_t	naddmap;	/* VOP_ADDMAP */
113 	kstat_named_t	ndelmap;	/* VOP_DELMAP */
114 	kstat_named_t	npoll;		/* VOP_POLL */
115 	kstat_named_t	ndump;		/* VOP_DUMP */
116 	kstat_named_t	npathconf;	/* VOP_PATHCONF */
117 	kstat_named_t	npageio;	/* VOP_PAGEIO */
118 	kstat_named_t	ndumpctl;	/* VOP_DUMPCTL */
119 	kstat_named_t	ndispose;	/* VOP_DISPOSE */
120 	kstat_named_t	nsetsecattr;	/* VOP_SETSECATTR */
121 	kstat_named_t	ngetsecattr;	/* VOP_GETSECATTR */
122 	kstat_named_t	nshrlock;	/* VOP_SHRLOCK */
123 	kstat_named_t	nvnevent;	/* VOP_VNEVENT */
124 	kstat_named_t	nreqzcbuf;	/* VOP_REQZCBUF */
125 	kstat_named_t	nretzcbuf;	/* VOP_RETZCBUF */
126 } vopstats_t;
127 
128 /*
129  * The vnode is the focus of all file activity in UNIX.
130  * A vnode is allocated for each active file, each current
131  * directory, each mounted-on file, and the root.
132  *
133  * Each vnode is usually associated with a file-system-specific node (for
134  * UFS, this is the in-memory inode).  Generally, a vnode and an fs-node
135  * should be created and destroyed together as a pair.
136  *
137  * If a vnode is reused for a new file, it should be reinitialized by calling
138  * either vn_reinit() or vn_recycle().
139  *
140  * vn_reinit() resets the entire vnode as if it was returned by vn_alloc().
141  * The caller is responsible for setting up the entire vnode after calling
142  * vn_reinit().  This is important when using kmem caching where the vnode is
143  * allocated by a constructor, for instance.
144  *
145  * vn_recycle() is used when the file system keeps some state around in both
146  * the vnode and the associated FS-node.  In UFS, for example, the inode of
147  * a deleted file can be reused immediately.  The v_data, v_vfsp, v_op, etc.
148  * remains the same but certain fields related to the previous instance need
149  * to be reset.  In particular:
150  *	v_femhead
151  *	v_path
152  *	v_rdcnt, v_wrcnt
153  *	v_mmap_read, v_mmap_write
154  */
155 
156 /*
157  * vnode types.  VNON means no type.  These values are unrelated to
158  * values in on-disk inodes.
159  */
160 typedef enum vtype {
161 	VNON	= 0,
162 	VREG	= 1,
163 	VDIR	= 2,
164 	VBLK	= 3,
165 	VCHR	= 4,
166 	VLNK	= 5,
167 	VFIFO	= 6,
168 	VDOOR	= 7,
169 	VPROC	= 8,
170 	VSOCK	= 9,
171 	VPORT	= 10,
172 	VBAD	= 11
173 } vtype_t;
174 
175 /*
176  * VSD - Vnode Specific Data
177  * Used to associate additional private data with a vnode.
178  */
179 struct vsd_node {
180 	list_node_t vs_nodes;		/* list of all VSD nodes */
181 	uint_t vs_nkeys;		/* entries in value array */
182 	void **vs_value;		/* array of value/key */
183 };
184 
185 /*
186  * Many of the fields in the vnode are read-only once they are initialized
187  * at vnode creation time.  Other fields are protected by locks.
188  *
189  * IMPORTANT: vnodes should be created ONLY by calls to vn_alloc().  They
190  * may not be embedded into the file-system specific node (inode).  The
191  * size of vnodes may change.
192  *
193  * The v_lock protects:
194  *   v_flag
195  *   v_stream
196  *   v_count
197  *   v_shrlocks
198  *   v_path
199  *   v_vsd
200  *   v_xattrdir
201  *
202  * A special lock (implemented by vn_vfswlock in vnode.c) protects:
203  *   v_vfsmountedhere
204  *
205  * The global flock_lock mutex (in flock.c) protects:
206  *   v_filocks
207  *
208  * IMPORTANT NOTE:
209  *
210  *   The following vnode fields are considered public and may safely be
211  *   accessed by file systems or other consumers:
212  *
213  *     v_lock
214  *     v_flag
215  *     v_count
216  *     v_data
217  *     v_vfsp
218  *     v_stream
219  *     v_type
220  *     v_rdev
221  *
222  * ALL OTHER FIELDS SHOULD BE ACCESSED ONLY BY THE OWNER OF THAT FIELD.
223  * In particular, file systems should not access other fields; they may
224  * change or even be removed.  The functionality which was once provided
225  * by these fields is available through vn_* functions.
226  */
227 
228 struct fem_head;	/* from fem.h */
229 
230 typedef struct vnode {
231 	kmutex_t	v_lock;		/* protects vnode fields */
232 	uint_t		v_flag;		/* vnode flags (see below) */
233 	uint_t		v_count;	/* reference count */
234 	void		*v_data;	/* private data for fs */
235 	struct vfs	*v_vfsp;	/* ptr to containing VFS */
236 	struct stdata	*v_stream;	/* associated stream */
237 	enum vtype	v_type;		/* vnode type */
238 	dev_t		v_rdev;		/* device (VCHR, VBLK) */
239 
240 	/* PRIVATE FIELDS BELOW - DO NOT USE */
241 
242 	struct vfs	*v_vfsmountedhere; /* ptr to vfs mounted here */
243 	struct vnodeops	*v_op;		/* vnode operations */
244 	struct page	*v_pages;	/* vnode pages list */
245 	struct filock	*v_filocks;	/* ptr to filock list */
246 	struct shrlocklist *v_shrlocks;	/* ptr to shrlock list */
247 	krwlock_t	v_nbllock;	/* sync for NBMAND locks */
248 	kcondvar_t	v_cv;		/* synchronize locking */
249 	void		*v_locality;	/* hook for locality info */
250 	struct fem_head	*v_femhead;	/* fs monitoring */
251 	char		*v_path;	/* cached path */
252 	uint_t		v_rdcnt;	/* open for read count  (VREG only) */
253 	uint_t		v_wrcnt;	/* open for write count (VREG only) */
254 	u_longlong_t	v_mmap_read;	/* mmap read count */
255 	u_longlong_t	v_mmap_write;	/* mmap write count */
256 	void		*v_mpssdata;	/* info for large page mappings */
257 	void		*v_fopdata;	/* list of file ops event watches */
258 	kmutex_t	v_vsd_lock;	/* protects v_vsd field */
259 	struct vsd_node *v_vsd;		/* vnode specific data */
260 	struct vnode	*v_xattrdir;	/* unnamed extended attr dir (GFS) */
261 	uint_t		v_count_dnlc;	/* dnlc reference count */
262 } vnode_t;
263 
264 #define	IS_DEVVP(vp)	\
265 	((vp)->v_type == VCHR || (vp)->v_type == VBLK || (vp)->v_type == VFIFO)
266 
267 #define	VNODE_ALIGN	64
268 /* Count of low-order 0 bits in a vnode *, based on size and alignment. */
269 #if defined(_LP64)
270 #define	VNODE_ALIGN_LOG2	8
271 #else
272 #define	VNODE_ALIGN_LOG2	7
273 #endif
274 
275 /*
276  * vnode flags.
277  */
278 #define	VROOT		0x01	/* root of its file system */
279 #define	VNOCACHE	0x02	/* don't keep cache pages on vnode */
280 #define	VNOMAP		0x04	/* file cannot be mapped/faulted */
281 #define	VDUP		0x08	/* file should be dup'ed rather then opened */
282 #define	VNOSWAP		0x10	/* file cannot be used as virtual swap device */
283 #define	VNOMOUNT	0x20	/* file cannot be covered by mount */
284 #define	VISSWAP		0x40	/* vnode is being used for swap */
285 #define	VSWAPLIKE	0x80	/* vnode acts like swap (but may not be) */
286 
287 #define	IS_SWAPVP(vp)	(((vp)->v_flag & (VISSWAP | VSWAPLIKE)) != 0)
288 
289 typedef struct vn_vfslocks_entry {
290 	rwstlock_t ve_lock;
291 	void *ve_vpvfs;
292 	struct vn_vfslocks_entry *ve_next;
293 	uint32_t ve_refcnt;
294 	char pad[64 - sizeof (rwstlock_t) - 2 * sizeof (void *) - \
295 	    sizeof (uint32_t)];
296 } vn_vfslocks_entry_t;
297 
298 /*
299  * The following two flags are used to lock the v_vfsmountedhere field
300  */
301 #define	VVFSLOCK	0x100
302 #define	VVFSWAIT	0x200
303 
304 /*
305  * Used to serialize VM operations on a vnode
306  */
307 #define	VVMLOCK		0x400
308 
309 /*
310  * Tell vn_open() not to fail a directory open for writing but
311  * to go ahead and call VOP_OPEN() to let the filesystem check.
312  */
313 #define	VDIROPEN	0x800
314 
315 /*
316  * Flag to let the VM system know that this file is most likely a binary
317  * or shared library since it has been mmap()ed EXEC at some time.
318  */
319 #define	VVMEXEC		0x1000
320 
321 #define	VPXFS		0x2000  /* clustering: global fs proxy vnode */
322 
323 #define	IS_PXFSVP(vp)	((vp)->v_flag & VPXFS)
324 
325 #define	V_XATTRDIR	0x4000	/* attribute unnamed directory */
326 
327 #define	IS_XATTRDIR(vp)	((vp)->v_flag & V_XATTRDIR)
328 
329 #define	V_LOCALITY	0x8000	/* whether locality aware */
330 
331 /*
332  * Flag that indicates the VM should maintain the v_pages list with all modified
333  * pages on one end and unmodified pages at the other. This makes finding dirty
334  * pages to write back to disk much faster at the expense of taking a minor
335  * fault on the first store instruction which touches a writable page.
336  */
337 #define	VMODSORT	(0x10000)
338 #define	IS_VMODSORT(vp) \
339 	(pvn_vmodsort_supported != 0 && ((vp)->v_flag  & VMODSORT) != 0)
340 
341 #define	VISSWAPFS	0x20000	/* vnode is being used for swapfs */
342 
343 /*
344  * The mdb memstat command assumes that IS_SWAPFSVP only uses the
345  * vnode's v_flag field.  If this changes, cache the additional
346  * fields in mdb; see vn_get in mdb/common/modules/genunix/memory.c
347  */
348 #define	IS_SWAPFSVP(vp)	(((vp)->v_flag & VISSWAPFS) != 0)
349 
350 #define	V_SYSATTR	0x40000	/* vnode is a GFS system attribute */
351 
352 /*
353  * Vnode attributes.  A bit-mask is supplied as part of the
354  * structure to indicate the attributes the caller wants to
355  * set (setattr) or extract (getattr).
356  */
357 
358 /*
359  * Note that va_nodeid and va_nblocks are 64bit data type.
360  * We support large files over NFSV3. With Solaris client and
361  * Server that generates 64bit ino's and sizes these fields
362  * will overflow if they are 32 bit sizes.
363  */
364 
365 typedef struct vattr {
366 	uint_t		va_mask;	/* bit-mask of attributes */
367 	vtype_t		va_type;	/* vnode type (for create) */
368 	mode_t		va_mode;	/* file access mode */
369 	uid_t		va_uid;		/* owner user id */
370 	gid_t		va_gid;		/* owner group id */
371 	dev_t		va_fsid;	/* file system id (dev for now) */
372 	u_longlong_t	va_nodeid;	/* node id */
373 	nlink_t		va_nlink;	/* number of references to file */
374 	u_offset_t	va_size;	/* file size in bytes */
375 	timestruc_t	va_atime;	/* time of last access */
376 	timestruc_t	va_mtime;	/* time of last modification */
377 	timestruc_t	va_ctime;	/* time of last status change */
378 	dev_t		va_rdev;	/* device the file represents */
379 	uint_t		va_blksize;	/* fundamental block size */
380 	u_longlong_t	va_nblocks;	/* # of blocks allocated */
381 	uint_t		va_seq;		/* sequence number */
382 } vattr_t;
383 
384 #define	AV_SCANSTAMP_SZ	32		/* length of anti-virus scanstamp */
385 
386 /*
387  * Structure of all optional attributes.
388  */
389 typedef struct xoptattr {
390 	timestruc_t	xoa_createtime;	/* Create time of file */
391 	uint8_t		xoa_archive;
392 	uint8_t		xoa_system;
393 	uint8_t		xoa_readonly;
394 	uint8_t		xoa_hidden;
395 	uint8_t		xoa_nounlink;
396 	uint8_t		xoa_immutable;
397 	uint8_t		xoa_appendonly;
398 	uint8_t		xoa_nodump;
399 	uint8_t		xoa_opaque;
400 	uint8_t		xoa_av_quarantined;
401 	uint8_t		xoa_av_modified;
402 	uint8_t		xoa_av_scanstamp[AV_SCANSTAMP_SZ];
403 	uint8_t		xoa_reparse;
404 	uint64_t	xoa_generation;
405 	uint8_t		xoa_offline;
406 	uint8_t		xoa_sparse;
407 } xoptattr_t;
408 
409 /*
410  * The xvattr structure is really a variable length structure that
411  * is made up of:
412  * - The classic vattr_t (xva_vattr)
413  * - a 32 bit quantity (xva_mapsize) that specifies the size of the
414  *   attribute bitmaps in 32 bit words.
415  * - A pointer to the returned attribute bitmap (needed because the
416  *   previous element, the requested attribute bitmap) is variable lenth.
417  * - The requested attribute bitmap, which is an array of 32 bit words.
418  *   Callers use the XVA_SET_REQ() macro to set the bits corresponding to
419  *   the attributes that are being requested.
420  * - The returned attribute bitmap, which is an array of 32 bit words.
421  *   File systems that support optional attributes use the XVA_SET_RTN()
422  *   macro to set the bits corresponding to the attributes that are being
423  *   returned.
424  * - The xoptattr_t structure which contains the attribute values
425  *
426  * xva_mapsize determines how many words in the attribute bitmaps.
427  * Immediately following the attribute bitmaps is the xoptattr_t.
428  * xva_getxoptattr() is used to get the pointer to the xoptattr_t
429  * section.
430  */
431 
432 #define	XVA_MAPSIZE	3		/* Size of attr bitmaps */
433 #define	XVA_MAGIC	0x78766174	/* Magic # for verification */
434 
435 /*
436  * The xvattr structure is an extensible structure which permits optional
437  * attributes to be requested/returned.  File systems may or may not support
438  * optional attributes.  They do so at their own discretion but if they do
439  * support optional attributes, they must register the VFSFT_XVATTR feature
440  * so that the optional attributes can be set/retrived.
441  *
442  * The fields of the xvattr structure are:
443  *
444  * xva_vattr - The first element of an xvattr is a legacy vattr structure
445  * which includes the common attributes.  If AT_XVATTR is set in the va_mask
446  * then the entire structure is treated as an xvattr.  If AT_XVATTR is not
447  * set, then only the xva_vattr structure can be used.
448  *
449  * xva_magic - 0x78766174 (hex for "xvat"). Magic number for verification.
450  *
451  * xva_mapsize - Size of requested and returned attribute bitmaps.
452  *
453  * xva_rtnattrmapp - Pointer to xva_rtnattrmap[].  We need this since the
454  * size of the array before it, xva_reqattrmap[], could change which means
455  * the location of xva_rtnattrmap[] could change.  This will allow unbundled
456  * file systems to find the location of xva_rtnattrmap[] when the sizes change.
457  *
458  * xva_reqattrmap[] - Array of requested attributes.  Attributes are
459  * represented by a specific bit in a specific element of the attribute
460  * map array.  Callers set the bits corresponding to the attributes
461  * that the caller wants to get/set.
462  *
463  * xva_rtnattrmap[] - Array of attributes that the file system was able to
464  * process.  Not all file systems support all optional attributes.  This map
465  * informs the caller which attributes the underlying file system was able
466  * to set/get.  (Same structure as the requested attributes array in terms
467  * of each attribute  corresponding to specific bits and array elements.)
468  *
469  * xva_xoptattrs - Structure containing values of optional attributes.
470  * These values are only valid if the corresponding bits in xva_reqattrmap
471  * are set and the underlying file system supports those attributes.
472  */
473 typedef struct xvattr {
474 	vattr_t		xva_vattr;	/* Embedded vattr structure */
475 	uint32_t	xva_magic;	/* Magic Number */
476 	uint32_t	xva_mapsize;	/* Size of attr bitmap (32-bit words) */
477 	uint32_t	*xva_rtnattrmapp;	/* Ptr to xva_rtnattrmap[] */
478 	uint32_t	xva_reqattrmap[XVA_MAPSIZE];	/* Requested attrs */
479 	uint32_t	xva_rtnattrmap[XVA_MAPSIZE];	/* Returned attrs */
480 	xoptattr_t	xva_xoptattrs;	/* Optional attributes */
481 } xvattr_t;
482 
483 #ifdef _SYSCALL32
484 /*
485  * For bigtypes time_t changed to 64 bit on the 64-bit kernel.
486  * Define an old version for user/kernel interface
487  */
488 
489 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
490 #pragma pack(4)
491 #endif
492 
493 typedef struct vattr32 {
494 	uint32_t	va_mask;	/* bit-mask of attributes */
495 	vtype_t		va_type;	/* vnode type (for create) */
496 	mode32_t	va_mode;	/* file access mode */
497 	uid32_t		va_uid;		/* owner user id */
498 	gid32_t		va_gid;		/* owner group id */
499 	dev32_t		va_fsid;	/* file system id (dev for now) */
500 	u_longlong_t	va_nodeid;	/* node id */
501 	nlink_t		va_nlink;	/* number of references to file */
502 	u_offset_t	va_size;	/* file size in bytes */
503 	timestruc32_t	va_atime;	/* time of last access */
504 	timestruc32_t	va_mtime;	/* time of last modification */
505 	timestruc32_t	va_ctime;	/* time of last status change */
506 	dev32_t		va_rdev;	/* device the file represents */
507 	uint32_t	va_blksize;	/* fundamental block size */
508 	u_longlong_t	va_nblocks;	/* # of blocks allocated */
509 	uint32_t	va_seq;		/* sequence number */
510 } vattr32_t;
511 
512 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
513 #pragma pack()
514 #endif
515 
516 #else  /* not _SYSCALL32 */
517 #define	vattr32		vattr
518 typedef vattr_t		vattr32_t;
519 #endif /* _SYSCALL32 */
520 
521 /*
522  * Attributes of interest to the caller of setattr or getattr.
523  */
524 #define	AT_TYPE		0x00001
525 #define	AT_MODE		0x00002
526 #define	AT_UID		0x00004
527 #define	AT_GID		0x00008
528 #define	AT_FSID		0x00010
529 #define	AT_NODEID	0x00020
530 #define	AT_NLINK	0x00040
531 #define	AT_SIZE		0x00080
532 #define	AT_ATIME	0x00100
533 #define	AT_MTIME	0x00200
534 #define	AT_CTIME	0x00400
535 #define	AT_RDEV		0x00800
536 #define	AT_BLKSIZE	0x01000
537 #define	AT_NBLOCKS	0x02000
538 /*			0x04000 */	/* unused */
539 #define	AT_SEQ		0x08000
540 /*
541  * If AT_XVATTR is set then there are additional bits to process in
542  * the xvattr_t's attribute bitmap.  If this is not set then the bitmap
543  * MUST be ignored.  Note that this bit must be set/cleared explicitly.
544  * That is, setting AT_ALL will NOT set AT_XVATTR.
545  */
546 #define	AT_XVATTR	0x10000
547 
548 #define	AT_ALL		(AT_TYPE|AT_MODE|AT_UID|AT_GID|AT_FSID|AT_NODEID|\
549 			AT_NLINK|AT_SIZE|AT_ATIME|AT_MTIME|AT_CTIME|\
550 			AT_RDEV|AT_BLKSIZE|AT_NBLOCKS|AT_SEQ)
551 
552 #define	AT_STAT		(AT_MODE|AT_UID|AT_GID|AT_FSID|AT_NODEID|AT_NLINK|\
553 			AT_SIZE|AT_ATIME|AT_MTIME|AT_CTIME|AT_RDEV|AT_TYPE)
554 
555 #define	AT_TIMES	(AT_ATIME|AT_MTIME|AT_CTIME)
556 
557 #define	AT_NOSET	(AT_NLINK|AT_RDEV|AT_FSID|AT_NODEID|AT_TYPE|\
558 			AT_BLKSIZE|AT_NBLOCKS|AT_SEQ)
559 
560 /*
561  * Attribute bits used in the extensible attribute's (xva's) attribute
562  * bitmaps.  Note that the bitmaps are made up of a variable length number
563  * of 32-bit words.  The convention is to use XAT{n}_{attrname} where "n"
564  * is the element in the bitmap (starting at 1).  This convention is for
565  * the convenience of the maintainer to keep track of which element each
566  * attribute belongs to.
567  *
568  * NOTE THAT CONSUMERS MUST *NOT* USE THE XATn_* DEFINES DIRECTLY.  CONSUMERS
569  * MUST USE THE XAT_* DEFINES.
570  */
571 #define	XAT0_INDEX	0LL		/* Index into bitmap for XAT0 attrs */
572 #define	XAT0_CREATETIME	0x00000001	/* Create time of file */
573 #define	XAT0_ARCHIVE	0x00000002	/* Archive */
574 #define	XAT0_SYSTEM	0x00000004	/* System */
575 #define	XAT0_READONLY	0x00000008	/* Readonly */
576 #define	XAT0_HIDDEN	0x00000010	/* Hidden */
577 #define	XAT0_NOUNLINK	0x00000020	/* Nounlink */
578 #define	XAT0_IMMUTABLE	0x00000040	/* immutable */
579 #define	XAT0_APPENDONLY	0x00000080	/* appendonly */
580 #define	XAT0_NODUMP	0x00000100	/* nodump */
581 #define	XAT0_OPAQUE	0x00000200	/* opaque */
582 #define	XAT0_AV_QUARANTINED	0x00000400	/* anti-virus quarantine */
583 #define	XAT0_AV_MODIFIED	0x00000800	/* anti-virus modified */
584 #define	XAT0_AV_SCANSTAMP	0x00001000	/* anti-virus scanstamp */
585 #define	XAT0_REPARSE	0x00002000	/* FS reparse point */
586 #define	XAT0_GEN	0x00004000	/* object generation number */
587 #define	XAT0_OFFLINE	0x00008000	/* offline */
588 #define	XAT0_SPARSE	0x00010000	/* sparse */
589 
590 #define	XAT0_ALL_ATTRS	(XAT0_CREATETIME|XAT0_ARCHIVE|XAT0_SYSTEM| \
591     XAT0_READONLY|XAT0_HIDDEN|XAT0_NOUNLINK|XAT0_IMMUTABLE|XAT0_APPENDONLY| \
592     XAT0_NODUMP|XAT0_OPAQUE|XAT0_AV_QUARANTINED|  XAT0_AV_MODIFIED| \
593     XAT0_AV_SCANSTAMP|XAT0_REPARSE|XATO_GEN|XAT0_OFFLINE|XAT0_SPARSE)
594 
595 /* Support for XAT_* optional attributes */
596 #define	XVA_MASK		0xffffffff	/* Used to mask off 32 bits */
597 #define	XVA_SHFT		32		/* Used to shift index */
598 
599 /*
600  * Used to pry out the index and attribute bits from the XAT_* attributes
601  * defined below.  Note that we're masking things down to 32 bits then
602  * casting to uint32_t.
603  */
604 #define	XVA_INDEX(attr)		((uint32_t)(((attr) >> XVA_SHFT) & XVA_MASK))
605 #define	XVA_ATTRBIT(attr)	((uint32_t)((attr) & XVA_MASK))
606 
607 /*
608  * The following defines present a "flat namespace" so that consumers don't
609  * need to keep track of which element belongs to which bitmap entry.
610  *
611  * NOTE THAT THESE MUST NEVER BE OR-ed TOGETHER
612  */
613 #define	XAT_CREATETIME		((XAT0_INDEX << XVA_SHFT) | XAT0_CREATETIME)
614 #define	XAT_ARCHIVE		((XAT0_INDEX << XVA_SHFT) | XAT0_ARCHIVE)
615 #define	XAT_SYSTEM		((XAT0_INDEX << XVA_SHFT) | XAT0_SYSTEM)
616 #define	XAT_READONLY		((XAT0_INDEX << XVA_SHFT) | XAT0_READONLY)
617 #define	XAT_HIDDEN		((XAT0_INDEX << XVA_SHFT) | XAT0_HIDDEN)
618 #define	XAT_NOUNLINK		((XAT0_INDEX << XVA_SHFT) | XAT0_NOUNLINK)
619 #define	XAT_IMMUTABLE		((XAT0_INDEX << XVA_SHFT) | XAT0_IMMUTABLE)
620 #define	XAT_APPENDONLY		((XAT0_INDEX << XVA_SHFT) | XAT0_APPENDONLY)
621 #define	XAT_NODUMP		((XAT0_INDEX << XVA_SHFT) | XAT0_NODUMP)
622 #define	XAT_OPAQUE		((XAT0_INDEX << XVA_SHFT) | XAT0_OPAQUE)
623 #define	XAT_AV_QUARANTINED	((XAT0_INDEX << XVA_SHFT) | XAT0_AV_QUARANTINED)
624 #define	XAT_AV_MODIFIED		((XAT0_INDEX << XVA_SHFT) | XAT0_AV_MODIFIED)
625 #define	XAT_AV_SCANSTAMP	((XAT0_INDEX << XVA_SHFT) | XAT0_AV_SCANSTAMP)
626 #define	XAT_REPARSE		((XAT0_INDEX << XVA_SHFT) | XAT0_REPARSE)
627 #define	XAT_GEN			((XAT0_INDEX << XVA_SHFT) | XAT0_GEN)
628 #define	XAT_OFFLINE		((XAT0_INDEX << XVA_SHFT) | XAT0_OFFLINE)
629 #define	XAT_SPARSE		((XAT0_INDEX << XVA_SHFT) | XAT0_SPARSE)
630 
631 /*
632  * The returned attribute map array (xva_rtnattrmap[]) is located past the
633  * requested attribute map array (xva_reqattrmap[]).  Its location changes
634  * when the array sizes change.  We use a separate pointer in a known location
635  * (xva_rtnattrmapp) to hold the location of xva_rtnattrmap[].  This is
636  * set in xva_init()
637  */
638 #define	XVA_RTNATTRMAP(xvap)	((xvap)->xva_rtnattrmapp)
639 
640 /*
641  * XVA_SET_REQ() sets an attribute bit in the proper element in the bitmap
642  * of requested attributes (xva_reqattrmap[]).
643  */
644 #define	XVA_SET_REQ(xvap, attr)					\
645 	ASSERT((xvap)->xva_vattr.va_mask | AT_XVATTR);		\
646 	ASSERT((xvap)->xva_magic == XVA_MAGIC);			\
647 	(xvap)->xva_reqattrmap[XVA_INDEX(attr)] |= XVA_ATTRBIT(attr)
648 /*
649  * XVA_CLR_REQ() clears an attribute bit in the proper element in the bitmap
650  * of requested attributes (xva_reqattrmap[]).
651  */
652 #define	XVA_CLR_REQ(xvap, attr)					\
653 	ASSERT((xvap)->xva_vattr.va_mask | AT_XVATTR);		\
654 	ASSERT((xvap)->xva_magic == XVA_MAGIC);			\
655 	(xvap)->xva_reqattrmap[XVA_INDEX(attr)] &= ~XVA_ATTRBIT(attr)
656 
657 /*
658  * XVA_SET_RTN() sets an attribute bit in the proper element in the bitmap
659  * of returned attributes (xva_rtnattrmap[]).
660  */
661 #define	XVA_SET_RTN(xvap, attr)					\
662 	ASSERT((xvap)->xva_vattr.va_mask | AT_XVATTR);		\
663 	ASSERT((xvap)->xva_magic == XVA_MAGIC);			\
664 	(XVA_RTNATTRMAP(xvap))[XVA_INDEX(attr)] |= XVA_ATTRBIT(attr)
665 
666 /*
667  * XVA_ISSET_REQ() checks the requested attribute bitmap (xva_reqattrmap[])
668  * to see of the corresponding attribute bit is set.  If so, returns non-zero.
669  */
670 #define	XVA_ISSET_REQ(xvap, attr)					\
671 	((((xvap)->xva_vattr.va_mask | AT_XVATTR) &&			\
672 		((xvap)->xva_magic == XVA_MAGIC) &&			\
673 		((xvap)->xva_mapsize > XVA_INDEX(attr))) ?		\
674 	((xvap)->xva_reqattrmap[XVA_INDEX(attr)] & XVA_ATTRBIT(attr)) :	0)
675 
676 /*
677  * XVA_ISSET_RTN() checks the returned attribute bitmap (xva_rtnattrmap[])
678  * to see of the corresponding attribute bit is set.  If so, returns non-zero.
679  */
680 #define	XVA_ISSET_RTN(xvap, attr)					\
681 	((((xvap)->xva_vattr.va_mask | AT_XVATTR) &&			\
682 		((xvap)->xva_magic == XVA_MAGIC) &&			\
683 		((xvap)->xva_mapsize > XVA_INDEX(attr))) ?		\
684 	((XVA_RTNATTRMAP(xvap))[XVA_INDEX(attr)] & XVA_ATTRBIT(attr)) : 0)
685 
686 /*
687  *  Modes.  Some values same as S_xxx entries from stat.h for convenience.
688  */
689 #define	VSUID		04000		/* set user id on execution */
690 #define	VSGID		02000		/* set group id on execution */
691 #define	VSVTX		01000		/* save swapped text even after use */
692 
693 /*
694  * Permissions.
695  */
696 #define	VREAD		00400
697 #define	VWRITE		00200
698 #define	VEXEC		00100
699 
700 #define	MODEMASK	07777		/* mode bits plus permission bits */
701 #define	PERMMASK	00777		/* permission bits */
702 
703 /*
704  * VOP_ACCESS flags
705  */
706 #define	V_ACE_MASK	0x1	/* mask represents  NFSv4 ACE permissions */
707 #define	V_APPEND	0x2	/* want to do append only check */
708 
709 /*
710  * Check whether mandatory file locking is enabled.
711  */
712 
713 #define	MANDMODE(mode)		(((mode) & (VSGID|(VEXEC>>3))) == VSGID)
714 #define	MANDLOCK(vp, mode)	((vp)->v_type == VREG && MANDMODE(mode))
715 
716 /*
717  * Flags for vnode operations.
718  */
719 enum rm		{ RMFILE, RMDIRECTORY };	/* rm or rmdir (remove) */
720 enum symfollow	{ NO_FOLLOW, FOLLOW };		/* follow symlinks (or not) */
721 enum vcexcl	{ NONEXCL, EXCL };		/* (non)excl create */
722 enum create	{ CRCREAT, CRMKNOD, CRMKDIR };	/* reason for create */
723 
724 typedef enum rm		rm_t;
725 typedef enum symfollow	symfollow_t;
726 typedef enum vcexcl	vcexcl_t;
727 typedef enum create	create_t;
728 
729 /*
730  * Vnode Events - Used by VOP_VNEVENT
731  * The VE_PRE_RENAME_* events fire before the rename operation and are
732  * primarily used for specialized applications, such as NFSv4 delegation, which
733  * need to know about rename before it occurs.
734  */
735 typedef enum vnevent	{
736 	VE_SUPPORT	= 0,	/* Query */
737 	VE_RENAME_SRC	= 1,	/* Rename, with vnode as source */
738 	VE_RENAME_DEST	= 2,	/* Rename, with vnode as target/destination */
739 	VE_REMOVE	= 3,	/* Remove of vnode's name */
740 	VE_RMDIR	= 4,	/* Remove of directory vnode's name */
741 	VE_CREATE	= 5,	/* Create with vnode's name which exists */
742 	VE_LINK		= 6, 	/* Link with vnode's name as source */
743 	VE_RENAME_DEST_DIR	= 7, 	/* Rename with vnode as target dir */
744 	VE_MOUNTEDOVER	= 8, 	/* File or Filesystem got mounted over vnode */
745 	VE_TRUNCATE = 9,	/* Truncate */
746 	VE_PRE_RENAME_SRC = 10,	/* Pre-rename, with vnode as source */
747 	VE_PRE_RENAME_DEST = 11, /* Pre-rename, with vnode as target/dest. */
748 	VE_PRE_RENAME_DEST_DIR = 12 /* Pre-rename with vnode as target dir */
749 } vnevent_t;
750 
751 /*
752  * Values for checking vnode open and map counts
753  */
754 enum v_mode { V_READ, V_WRITE, V_RDORWR, V_RDANDWR };
755 
756 typedef enum v_mode v_mode_t;
757 
758 #define	V_TRUE	1
759 #define	V_FALSE	0
760 
761 /*
762  * Structure used on VOP_GETSECATTR and VOP_SETSECATTR operations
763  */
764 
765 typedef struct vsecattr {
766 	uint_t		vsa_mask;	/* See below */
767 	int		vsa_aclcnt;	/* ACL entry count */
768 	void		*vsa_aclentp;	/* pointer to ACL entries */
769 	int		vsa_dfaclcnt;	/* default ACL entry count */
770 	void		*vsa_dfaclentp;	/* pointer to default ACL entries */
771 	size_t		vsa_aclentsz;	/* ACE size in bytes of vsa_aclentp */
772 	uint_t		vsa_aclflags;	/* ACE ACL flags */
773 } vsecattr_t;
774 
775 /* vsa_mask values */
776 #define	VSA_ACL			0x0001
777 #define	VSA_ACLCNT		0x0002
778 #define	VSA_DFACL		0x0004
779 #define	VSA_DFACLCNT		0x0008
780 #define	VSA_ACE			0x0010
781 #define	VSA_ACECNT		0x0020
782 #define	VSA_ACE_ALLTYPES	0x0040
783 #define	VSA_ACE_ACLFLAGS	0x0080	/* get/set ACE ACL flags */
784 
785 /*
786  * Structure used by various vnode operations to determine
787  * the context (pid, host, identity) of a caller.
788  *
789  * The cc_caller_id is used to identify one or more callers who invoke
790  * operations, possibly on behalf of others.  For example, the NFS
791  * server could have it's own cc_caller_id which can be detected by
792  * vnode/vfs operations or (FEM) monitors on those operations.  New
793  * caller IDs are generated by fs_new_caller_id().
794  */
795 typedef struct caller_context {
796 	pid_t		cc_pid;		/* Process ID of the caller */
797 	int		cc_sysid;	/* System ID, used for remote calls */
798 	u_longlong_t	cc_caller_id;	/* Identifier for (set of) caller(s) */
799 	ulong_t		cc_flags;
800 } caller_context_t;
801 
802 /*
803  * Flags for caller context.  The caller sets CC_DONTBLOCK if it does not
804  * want to block inside of a FEM monitor.  The monitor will set CC_WOULDBLOCK
805  * and return EAGAIN if the operation would have blocked.
806  */
807 #define	CC_WOULDBLOCK	0x01
808 #define	CC_DONTBLOCK	0x02
809 
810 /*
811  * Structure tags for function prototypes, defined elsewhere.
812  */
813 struct pathname;
814 struct fid;
815 struct flock64;
816 struct flk_callback;
817 struct shrlock;
818 struct page;
819 struct seg;
820 struct as;
821 struct pollhead;
822 struct taskq;
823 
824 #ifdef	_KERNEL
825 
826 /*
827  * VNODE_OPS defines all the vnode operations.  It is used to define
828  * the vnodeops structure (below) and the fs_func_p union (vfs_opreg.h).
829  */
830 #define	VNODE_OPS							\
831 	int	(*vop_open)(vnode_t **, int, cred_t *,			\
832 				caller_context_t *);			\
833 	int	(*vop_close)(vnode_t *, int, int, offset_t, cred_t *,	\
834 				caller_context_t *);			\
835 	int	(*vop_read)(vnode_t *, uio_t *, int, cred_t *,		\
836 				caller_context_t *);			\
837 	int	(*vop_write)(vnode_t *, uio_t *, int, cred_t *,		\
838 				caller_context_t *);			\
839 	int	(*vop_ioctl)(vnode_t *, int, intptr_t, int, cred_t *,	\
840 				int *, caller_context_t *);		\
841 	int	(*vop_setfl)(vnode_t *, int, int, cred_t *,		\
842 				caller_context_t *);			\
843 	int	(*vop_getattr)(vnode_t *, vattr_t *, int, cred_t *,	\
844 				caller_context_t *);			\
845 	int	(*vop_setattr)(vnode_t *, vattr_t *, int, cred_t *,	\
846 				caller_context_t *);			\
847 	int	(*vop_access)(vnode_t *, int, int, cred_t *,		\
848 				caller_context_t *);			\
849 	int	(*vop_lookup)(vnode_t *, char *, vnode_t **,		\
850 				struct pathname *,			\
851 				int, vnode_t *, cred_t *,		\
852 				caller_context_t *, int *,		\
853 				struct pathname *);			\
854 	int	(*vop_create)(vnode_t *, char *, vattr_t *, vcexcl_t,	\
855 				int, vnode_t **, cred_t *, int,		\
856 				caller_context_t *, vsecattr_t *);	\
857 	int	(*vop_remove)(vnode_t *, char *, cred_t *,		\
858 				caller_context_t *, int);		\
859 	int	(*vop_link)(vnode_t *, vnode_t *, char *, cred_t *,	\
860 				caller_context_t *, int);		\
861 	int	(*vop_rename)(vnode_t *, char *, vnode_t *, char *,	\
862 				cred_t *, caller_context_t *, int);	\
863 	int	(*vop_mkdir)(vnode_t *, char *, vattr_t *, vnode_t **,	\
864 				cred_t *, caller_context_t *, int,	\
865 				vsecattr_t *);				\
866 	int	(*vop_rmdir)(vnode_t *, char *, vnode_t *, cred_t *,	\
867 				caller_context_t *, int);		\
868 	int	(*vop_readdir)(vnode_t *, uio_t *, cred_t *, int *,	\
869 				caller_context_t *, int);		\
870 	int	(*vop_symlink)(vnode_t *, char *, vattr_t *, char *,	\
871 				cred_t *, caller_context_t *, int);	\
872 	int	(*vop_readlink)(vnode_t *, uio_t *, cred_t *,		\
873 				caller_context_t *);			\
874 	int	(*vop_fsync)(vnode_t *, int, cred_t *,			\
875 				caller_context_t *);			\
876 	void	(*vop_inactive)(vnode_t *, cred_t *,			\
877 				caller_context_t *);			\
878 	int	(*vop_fid)(vnode_t *, struct fid *,			\
879 				caller_context_t *);			\
880 	int	(*vop_rwlock)(vnode_t *, int, caller_context_t *);	\
881 	void	(*vop_rwunlock)(vnode_t *, int, caller_context_t *);	\
882 	int	(*vop_seek)(vnode_t *, offset_t, offset_t *,		\
883 				caller_context_t *);			\
884 	int	(*vop_cmp)(vnode_t *, vnode_t *, caller_context_t *);	\
885 	int	(*vop_frlock)(vnode_t *, int, struct flock64 *,		\
886 				int, offset_t,				\
887 				struct flk_callback *, cred_t *,	\
888 				caller_context_t *);			\
889 	int	(*vop_space)(vnode_t *, int, struct flock64 *,		\
890 				int, offset_t,				\
891 				cred_t *, caller_context_t *);		\
892 	int	(*vop_realvp)(vnode_t *, vnode_t **,			\
893 				caller_context_t *);			\
894 	int	(*vop_getpage)(vnode_t *, offset_t, size_t, uint_t *,	\
895 				struct page **, size_t, struct seg *,	\
896 				caddr_t, enum seg_rw, cred_t *,		\
897 				caller_context_t *);			\
898 	int	(*vop_putpage)(vnode_t *, offset_t, size_t,		\
899 				int, cred_t *, caller_context_t *);	\
900 	int	(*vop_map)(vnode_t *, offset_t, struct as *,		\
901 				caddr_t *, size_t,			\
902 				uchar_t, uchar_t, uint_t, cred_t *,	\
903 				caller_context_t *);			\
904 	int	(*vop_addmap)(vnode_t *, offset_t, struct as *,		\
905 				caddr_t, size_t,			\
906 				uchar_t, uchar_t, uint_t, cred_t *,	\
907 				caller_context_t *);			\
908 	int	(*vop_delmap)(vnode_t *, offset_t, struct as *,		\
909 				caddr_t, size_t,			\
910 				uint_t, uint_t, uint_t, cred_t *,	\
911 				caller_context_t *);			\
912 	int	(*vop_poll)(vnode_t *, short, int, short *,		\
913 				struct pollhead **,			\
914 				caller_context_t *);			\
915 	int	(*vop_dump)(vnode_t *, caddr_t, offset_t, offset_t,	\
916 				caller_context_t *);			\
917 	int	(*vop_pathconf)(vnode_t *, int, ulong_t *, cred_t *,	\
918 				caller_context_t *);			\
919 	int	(*vop_pageio)(vnode_t *, struct page *,			\
920 				u_offset_t, size_t, int, cred_t *,	\
921 				caller_context_t *);			\
922 	int	(*vop_dumpctl)(vnode_t *, int, offset_t *,		\
923 				caller_context_t *);			\
924 	void	(*vop_dispose)(vnode_t *, struct page *,		\
925 				int, int, cred_t *,			\
926 				caller_context_t *);			\
927 	int	(*vop_setsecattr)(vnode_t *, vsecattr_t *,		\
928 				int, cred_t *, caller_context_t *);	\
929 	int	(*vop_getsecattr)(vnode_t *, vsecattr_t *,		\
930 				int, cred_t *, caller_context_t *);	\
931 	int	(*vop_shrlock)(vnode_t *, int, struct shrlock *,	\
932 				int, cred_t *, caller_context_t *);	\
933 	int	(*vop_vnevent)(vnode_t *, vnevent_t, vnode_t *,		\
934 				char *, caller_context_t *);		\
935 	int	(*vop_reqzcbuf)(vnode_t *, enum uio_rw, xuio_t *,	\
936 				cred_t *, caller_context_t *);		\
937 	int	(*vop_retzcbuf)(vnode_t *, xuio_t *, cred_t *,		\
938 				caller_context_t *)
939 	/* NB: No ";" */
940 
941 /*
942  * Operations on vnodes.  Note: File systems must never operate directly
943  * on a 'vnodeops' structure -- it WILL change in future releases!  They
944  * must use vn_make_ops() to create the structure.
945  */
946 typedef struct vnodeops {
947 	const char *vnop_name;
948 	VNODE_OPS;	/* Signatures of all vnode operations (vops) */
949 } vnodeops_t;
950 
951 typedef int (*fs_generic_func_p) ();	/* Generic vop/vfsop/femop/fsemop ptr */
952 
953 extern int	fop_open(vnode_t **, int, cred_t *, caller_context_t *);
954 extern int	fop_close(vnode_t *, int, int, offset_t, cred_t *,
955 				caller_context_t *);
956 extern int	fop_read(vnode_t *, uio_t *, int, cred_t *, caller_context_t *);
957 extern int	fop_write(vnode_t *, uio_t *, int, cred_t *,
958 				caller_context_t *);
959 extern int	fop_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
960 				caller_context_t *);
961 extern int	fop_setfl(vnode_t *, int, int, cred_t *, caller_context_t *);
962 extern int	fop_getattr(vnode_t *, vattr_t *, int, cred_t *,
963 				caller_context_t *);
964 extern int	fop_setattr(vnode_t *, vattr_t *, int, cred_t *,
965 				caller_context_t *);
966 extern int	fop_access(vnode_t *, int, int, cred_t *, caller_context_t *);
967 extern int	fop_lookup(vnode_t *, char *, vnode_t **, struct pathname *,
968 				int, vnode_t *, cred_t *, caller_context_t *,
969 				int *, struct pathname *);
970 extern int	fop_create(vnode_t *, char *, vattr_t *, vcexcl_t, int,
971 				vnode_t **, cred_t *, int, caller_context_t *,
972 				vsecattr_t *);
973 extern int	fop_remove(vnode_t *vp, char *, cred_t *, caller_context_t *,
974 				int);
975 extern int	fop_link(vnode_t *, vnode_t *, char *, cred_t *,
976 				caller_context_t *, int);
977 extern int	fop_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
978 				caller_context_t *, int);
979 extern int	fop_mkdir(vnode_t *, char *, vattr_t *, vnode_t **, cred_t *,
980 				caller_context_t *, int, vsecattr_t *);
981 extern int	fop_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
982 				caller_context_t *, int);
983 extern int	fop_readdir(vnode_t *, uio_t *, cred_t *, int *,
984 				caller_context_t *, int);
985 extern int	fop_symlink(vnode_t *, char *, vattr_t *, char *, cred_t *,
986 				caller_context_t *, int);
987 extern int	fop_readlink(vnode_t *, uio_t *, cred_t *, caller_context_t *);
988 extern int	fop_fsync(vnode_t *, int, cred_t *, caller_context_t *);
989 extern void	fop_inactive(vnode_t *, cred_t *, caller_context_t *);
990 extern int	fop_fid(vnode_t *, struct fid *, caller_context_t *);
991 extern int	fop_rwlock(vnode_t *, int, caller_context_t *);
992 extern void	fop_rwunlock(vnode_t *, int, caller_context_t *);
993 extern int	fop_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
994 extern int	fop_cmp(vnode_t *, vnode_t *, caller_context_t *);
995 extern int	fop_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
996 				struct flk_callback *, cred_t *,
997 				caller_context_t *);
998 extern int	fop_space(vnode_t *, int, struct flock64 *, int, offset_t,
999 				cred_t *, caller_context_t *);
1000 extern int	fop_realvp(vnode_t *, vnode_t **, caller_context_t *);
1001 extern int	fop_getpage(vnode_t *, offset_t, size_t, uint_t *,
1002 				struct page **, size_t, struct seg *,
1003 				caddr_t, enum seg_rw, cred_t *,
1004 				caller_context_t *);
1005 extern int	fop_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
1006 				caller_context_t *);
1007 extern int	fop_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
1008 				uchar_t, uchar_t, uint_t, cred_t *cr,
1009 				caller_context_t *);
1010 extern int	fop_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
1011 				uchar_t, uchar_t, uint_t, cred_t *,
1012 				caller_context_t *);
1013 extern int	fop_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
1014 				uint_t, uint_t, uint_t, cred_t *,
1015 				caller_context_t *);
1016 extern int	fop_poll(vnode_t *, short, int, short *, struct pollhead **,
1017 				caller_context_t *);
1018 extern int	fop_dump(vnode_t *, caddr_t, offset_t, offset_t,
1019     caller_context_t *);
1020 extern int	fop_pathconf(vnode_t *, int, ulong_t *, cred_t *,
1021 				caller_context_t *);
1022 extern int	fop_pageio(vnode_t *, struct page *, u_offset_t, size_t, int,
1023 				cred_t *, caller_context_t *);
1024 extern int	fop_dumpctl(vnode_t *, int, offset_t *, caller_context_t *);
1025 extern void	fop_dispose(vnode_t *, struct page *, int, int, cred_t *,
1026 				caller_context_t *);
1027 extern int	fop_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
1028 				caller_context_t *);
1029 extern int	fop_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
1030 				caller_context_t *);
1031 extern int	fop_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
1032 				caller_context_t *);
1033 extern int	fop_vnevent(vnode_t *, vnevent_t, vnode_t *, char *,
1034 				caller_context_t *);
1035 extern int	fop_reqzcbuf(vnode_t *, enum uio_rw, xuio_t *, cred_t *,
1036 				caller_context_t *);
1037 extern int	fop_retzcbuf(vnode_t *, xuio_t *, cred_t *, caller_context_t *);
1038 
1039 #endif	/* _KERNEL */
1040 
1041 #define	VOP_OPEN(vpp, mode, cr, ct) \
1042 	fop_open(vpp, mode, cr, ct)
1043 #define	VOP_CLOSE(vp, f, c, o, cr, ct) \
1044 	fop_close(vp, f, c, o, cr, ct)
1045 #define	VOP_READ(vp, uiop, iof, cr, ct) \
1046 	fop_read(vp, uiop, iof, cr, ct)
1047 #define	VOP_WRITE(vp, uiop, iof, cr, ct) \
1048 	fop_write(vp, uiop, iof, cr, ct)
1049 #define	VOP_IOCTL(vp, cmd, a, f, cr, rvp, ct) \
1050 	fop_ioctl(vp, cmd, a, f, cr, rvp, ct)
1051 #define	VOP_SETFL(vp, f, a, cr, ct) \
1052 	fop_setfl(vp, f, a, cr, ct)
1053 #define	VOP_GETATTR(vp, vap, f, cr, ct) \
1054 	fop_getattr(vp, vap, f, cr, ct)
1055 #define	VOP_SETATTR(vp, vap, f, cr, ct) \
1056 	fop_setattr(vp, vap, f, cr, ct)
1057 #define	VOP_ACCESS(vp, mode, f, cr, ct) \
1058 	fop_access(vp, mode, f, cr, ct)
1059 #define	VOP_LOOKUP(vp, cp, vpp, pnp, f, rdir, cr, ct, defp, rpnp) \
1060 	fop_lookup(vp, cp, vpp, pnp, f, rdir, cr, ct, defp, rpnp)
1061 #define	VOP_CREATE(dvp, p, vap, ex, mode, vpp, cr, flag, ct, vsap) \
1062 	fop_create(dvp, p, vap, ex, mode, vpp, cr, flag, ct, vsap)
1063 #define	VOP_REMOVE(dvp, p, cr, ct, f) \
1064 	fop_remove(dvp, p, cr, ct, f)
1065 #define	VOP_LINK(tdvp, fvp, p, cr, ct, f) \
1066 	fop_link(tdvp, fvp, p, cr, ct, f)
1067 #define	VOP_RENAME(fvp, fnm, tdvp, tnm, cr, ct, f) \
1068 	fop_rename(fvp, fnm, tdvp, tnm, cr, ct, f)
1069 #define	VOP_MKDIR(dp, p, vap, vpp, cr, ct, f, vsap) \
1070 	fop_mkdir(dp, p, vap, vpp, cr, ct, f, vsap)
1071 #define	VOP_RMDIR(dp, p, cdir, cr, ct, f) \
1072 	fop_rmdir(dp, p, cdir, cr, ct, f)
1073 #define	VOP_READDIR(vp, uiop, cr, eofp, ct, f) \
1074 	fop_readdir(vp, uiop, cr, eofp, ct, f)
1075 #define	VOP_SYMLINK(dvp, lnm, vap, tnm, cr, ct, f) \
1076 	fop_symlink(dvp, lnm, vap, tnm, cr, ct, f)
1077 #define	VOP_READLINK(vp, uiop, cr, ct) \
1078 	fop_readlink(vp, uiop, cr, ct)
1079 #define	VOP_FSYNC(vp, syncflag, cr, ct) \
1080 	fop_fsync(vp, syncflag, cr, ct)
1081 #define	VOP_INACTIVE(vp, cr, ct) \
1082 	fop_inactive(vp, cr, ct)
1083 #define	VOP_FID(vp, fidp, ct) \
1084 	fop_fid(vp, fidp, ct)
1085 #define	VOP_RWLOCK(vp, w, ct) \
1086 	fop_rwlock(vp, w, ct)
1087 #define	VOP_RWUNLOCK(vp, w, ct) \
1088 	fop_rwunlock(vp, w, ct)
1089 #define	VOP_SEEK(vp, ooff, noffp, ct) \
1090 	fop_seek(vp, ooff, noffp, ct)
1091 #define	VOP_CMP(vp1, vp2, ct) \
1092 	fop_cmp(vp1, vp2, ct)
1093 #define	VOP_FRLOCK(vp, cmd, a, f, o, cb, cr, ct) \
1094 	fop_frlock(vp, cmd, a, f, o, cb, cr, ct)
1095 #define	VOP_SPACE(vp, cmd, a, f, o, cr, ct) \
1096 	fop_space(vp, cmd, a, f, o, cr, ct)
1097 #define	VOP_REALVP(vp1, vp2, ct) \
1098 	fop_realvp(vp1, vp2, ct)
1099 #define	VOP_GETPAGE(vp, of, sz, pr, pl, ps, sg, a, rw, cr, ct) \
1100 	fop_getpage(vp, of, sz, pr, pl, ps, sg, a, rw, cr, ct)
1101 #define	VOP_PUTPAGE(vp, of, sz, fl, cr, ct) \
1102 	fop_putpage(vp, of, sz, fl, cr, ct)
1103 #define	VOP_MAP(vp, of, as, a, sz, p, mp, fl, cr, ct) \
1104 	fop_map(vp, of, as, a, sz, p, mp, fl, cr, ct)
1105 #define	VOP_ADDMAP(vp, of, as, a, sz, p, mp, fl, cr, ct) \
1106 	fop_addmap(vp, of, as, a, sz, p, mp, fl, cr, ct)
1107 #define	VOP_DELMAP(vp, of, as, a, sz, p, mp, fl, cr, ct) \
1108 	fop_delmap(vp, of, as, a, sz, p, mp, fl, cr, ct)
1109 #define	VOP_POLL(vp, events, anyyet, reventsp, phpp, ct) \
1110 	fop_poll(vp, events, anyyet, reventsp, phpp, ct)
1111 #define	VOP_DUMP(vp, addr, bn, count, ct) \
1112 	fop_dump(vp, addr, bn, count, ct)
1113 #define	VOP_PATHCONF(vp, cmd, valp, cr, ct) \
1114 	fop_pathconf(vp, cmd, valp, cr, ct)
1115 #define	VOP_PAGEIO(vp, pp, io_off, io_len, flags, cr, ct) \
1116 	fop_pageio(vp, pp, io_off, io_len, flags, cr, ct)
1117 #define	VOP_DUMPCTL(vp, action, blkp, ct) \
1118 	fop_dumpctl(vp, action, blkp, ct)
1119 #define	VOP_DISPOSE(vp, pp, flag, dn, cr, ct) \
1120 	fop_dispose(vp, pp, flag, dn, cr, ct)
1121 #define	VOP_GETSECATTR(vp, vsap, f, cr, ct) \
1122 	fop_getsecattr(vp, vsap, f, cr, ct)
1123 #define	VOP_SETSECATTR(vp, vsap, f, cr, ct) \
1124 	fop_setsecattr(vp, vsap, f, cr, ct)
1125 #define	VOP_SHRLOCK(vp, cmd, shr, f, cr, ct) \
1126 	fop_shrlock(vp, cmd, shr, f, cr, ct)
1127 #define	VOP_VNEVENT(vp, vnevent, dvp, fnm, ct) \
1128 	fop_vnevent(vp, vnevent, dvp, fnm, ct)
1129 #define	VOP_REQZCBUF(vp, rwflag, xuiop, cr, ct) \
1130 	fop_reqzcbuf(vp, rwflag, xuiop, cr, ct)
1131 #define	VOP_RETZCBUF(vp, xuiop, cr, ct) \
1132 	fop_retzcbuf(vp, xuiop, cr, ct)
1133 
1134 #define	VOPNAME_OPEN		"open"
1135 #define	VOPNAME_CLOSE		"close"
1136 #define	VOPNAME_READ		"read"
1137 #define	VOPNAME_WRITE		"write"
1138 #define	VOPNAME_IOCTL		"ioctl"
1139 #define	VOPNAME_SETFL		"setfl"
1140 #define	VOPNAME_GETATTR		"getattr"
1141 #define	VOPNAME_SETATTR		"setattr"
1142 #define	VOPNAME_ACCESS		"access"
1143 #define	VOPNAME_LOOKUP		"lookup"
1144 #define	VOPNAME_CREATE		"create"
1145 #define	VOPNAME_REMOVE		"remove"
1146 #define	VOPNAME_LINK		"link"
1147 #define	VOPNAME_RENAME		"rename"
1148 #define	VOPNAME_MKDIR		"mkdir"
1149 #define	VOPNAME_RMDIR		"rmdir"
1150 #define	VOPNAME_READDIR		"readdir"
1151 #define	VOPNAME_SYMLINK		"symlink"
1152 #define	VOPNAME_READLINK	"readlink"
1153 #define	VOPNAME_FSYNC		"fsync"
1154 #define	VOPNAME_INACTIVE	"inactive"
1155 #define	VOPNAME_FID		"fid"
1156 #define	VOPNAME_RWLOCK		"rwlock"
1157 #define	VOPNAME_RWUNLOCK	"rwunlock"
1158 #define	VOPNAME_SEEK		"seek"
1159 #define	VOPNAME_CMP		"cmp"
1160 #define	VOPNAME_FRLOCK		"frlock"
1161 #define	VOPNAME_SPACE		"space"
1162 #define	VOPNAME_REALVP		"realvp"
1163 #define	VOPNAME_GETPAGE		"getpage"
1164 #define	VOPNAME_PUTPAGE		"putpage"
1165 #define	VOPNAME_MAP		"map"
1166 #define	VOPNAME_ADDMAP		"addmap"
1167 #define	VOPNAME_DELMAP		"delmap"
1168 #define	VOPNAME_POLL		"poll"
1169 #define	VOPNAME_DUMP		"dump"
1170 #define	VOPNAME_PATHCONF	"pathconf"
1171 #define	VOPNAME_PAGEIO		"pageio"
1172 #define	VOPNAME_DUMPCTL		"dumpctl"
1173 #define	VOPNAME_DISPOSE		"dispose"
1174 #define	VOPNAME_GETSECATTR	"getsecattr"
1175 #define	VOPNAME_SETSECATTR	"setsecattr"
1176 #define	VOPNAME_SHRLOCK		"shrlock"
1177 #define	VOPNAME_VNEVENT		"vnevent"
1178 #define	VOPNAME_REQZCBUF	"reqzcbuf"
1179 #define	VOPNAME_RETZCBUF	"retzcbuf"
1180 
1181 /*
1182  * Flags for VOP_LOOKUP
1183  *
1184  * Defined in file.h, but also possible, FIGNORECASE and FSEARCH
1185  *
1186  */
1187 #define	LOOKUP_DIR		0x01	/* want parent dir vp */
1188 #define	LOOKUP_XATTR		0x02	/* lookup up extended attr dir */
1189 #define	CREATE_XATTR_DIR	0x04	/* Create extended attr dir */
1190 #define	LOOKUP_HAVE_SYSATTR_DIR	0x08	/* Already created virtual GFS dir */
1191 
1192 /*
1193  * Flags for VOP_READDIR
1194  */
1195 #define	V_RDDIR_ENTFLAGS	0x01	/* request dirent flags */
1196 #define	V_RDDIR_ACCFILTER	0x02	/* filter out inaccessible dirents */
1197 
1198 /*
1199  * Flags for VOP_RWLOCK/VOP_RWUNLOCK
1200  * VOP_RWLOCK will return the flag that was actually set, or -1 if none.
1201  */
1202 #define	V_WRITELOCK_TRUE	(1)	/* Request write-lock on the vnode */
1203 #define	V_WRITELOCK_FALSE	(0)	/* Request read-lock on the vnode */
1204 
1205 /*
1206  * Flags for VOP_DUMPCTL
1207  */
1208 #define	DUMP_ALLOC	0
1209 #define	DUMP_FREE	1
1210 #define	DUMP_SCAN	2
1211 
1212 /*
1213  * Public vnode manipulation functions.
1214  */
1215 #ifdef	_KERNEL
1216 
1217 vnode_t *vn_alloc(int);
1218 void	vn_reinit(vnode_t *);
1219 void	vn_recycle(vnode_t *);
1220 void	vn_free(vnode_t *);
1221 
1222 int	vn_is_readonly(vnode_t *);
1223 int   	vn_is_opened(vnode_t *, v_mode_t);
1224 int   	vn_is_mapped(vnode_t *, v_mode_t);
1225 int   	vn_has_other_opens(vnode_t *, v_mode_t);
1226 void	vn_open_upgrade(vnode_t *, int);
1227 void	vn_open_downgrade(vnode_t *, int);
1228 
1229 int	vn_can_change_zones(vnode_t *vp);
1230 
1231 int	vn_has_flocks(vnode_t *);
1232 int	vn_has_mandatory_locks(vnode_t *, int);
1233 int	vn_has_cached_data(vnode_t *);
1234 
1235 void	vn_setops(vnode_t *, vnodeops_t *);
1236 vnodeops_t *vn_getops(vnode_t *);
1237 int	vn_matchops(vnode_t *, vnodeops_t *);
1238 int	vn_matchopval(vnode_t *, char *, fs_generic_func_p);
1239 int	vn_ismntpt(vnode_t *);
1240 
1241 struct vfs *vn_mountedvfs(vnode_t *);
1242 
1243 int	vn_in_dnlc(vnode_t *);
1244 
1245 void	vn_create_cache(void);
1246 void	vn_destroy_cache(void);
1247 
1248 void	vn_freevnodeops(vnodeops_t *);
1249 
1250 int	vn_open(char *pnamep, enum uio_seg seg, int filemode, int createmode,
1251 		struct vnode **vpp, enum create crwhy, mode_t umask);
1252 int	vn_openat(char *pnamep, enum uio_seg seg, int filemode, int createmode,
1253 		struct vnode **vpp, enum create crwhy,
1254 		mode_t umask, struct vnode *startvp, int fd);
1255 int	vn_create(char *pnamep, enum uio_seg seg, struct vattr *vap,
1256 		enum vcexcl excl, int mode, struct vnode **vpp,
1257 		enum create why, int flag, mode_t umask);
1258 int	vn_createat(char *pnamep, enum uio_seg seg, struct vattr *vap,
1259 		enum vcexcl excl, int mode, struct vnode **vpp,
1260 		enum create why, int flag, mode_t umask, struct vnode *startvp);
1261 int	vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, ssize_t len,
1262 		offset_t offset, enum uio_seg seg, int ioflag, rlim64_t ulimit,
1263 		cred_t *cr, ssize_t *residp);
1264 void	vn_rele(struct vnode *vp);
1265 void	vn_rele_async(struct vnode *vp, struct taskq *taskq);
1266 void	vn_rele_dnlc(struct vnode *vp);
1267 void	vn_rele_stream(struct vnode *vp);
1268 int	vn_link(char *from, char *to, enum uio_seg seg);
1269 int	vn_linkat(vnode_t *fstartvp, char *from, enum symfollow follow,
1270 		vnode_t *tstartvp, char *to, enum uio_seg seg);
1271 int	vn_rename(char *from, char *to, enum uio_seg seg);
1272 int	vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp, char *tname,
1273 		enum uio_seg seg);
1274 int	vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag);
1275 int	vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg,
1276 		enum rm dirflag);
1277 int	vn_compare(vnode_t *vp1, vnode_t *vp2);
1278 int	vn_vfswlock(struct vnode *vp);
1279 int	vn_vfswlock_wait(struct vnode *vp);
1280 int	vn_vfsrlock(struct vnode *vp);
1281 int	vn_vfsrlock_wait(struct vnode *vp);
1282 void	vn_vfsunlock(struct vnode *vp);
1283 int	vn_vfswlock_held(struct vnode *vp);
1284 vnode_t *specvp(struct vnode *vp, dev_t dev, vtype_t type, struct cred *cr);
1285 vnode_t *makespecvp(dev_t dev, vtype_t type);
1286 vn_vfslocks_entry_t *vn_vfslocks_getlock(void *);
1287 void	vn_vfslocks_rele(vn_vfslocks_entry_t *);
1288 boolean_t vn_is_reparse(vnode_t *, cred_t *, caller_context_t *);
1289 
1290 void vn_copypath(struct vnode *src, struct vnode *dst);
1291 void vn_setpath_str(struct vnode *vp, const char *str, size_t len);
1292 void vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
1293     const char *path, size_t plen);
1294 void vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len);
1295 
1296 /* Vnode event notification */
1297 void	vnevent_rename_src(vnode_t *, vnode_t *, char *, caller_context_t *);
1298 void	vnevent_rename_dest(vnode_t *, vnode_t *, char *, caller_context_t *);
1299 void	vnevent_remove(vnode_t *, vnode_t *, char *, caller_context_t *);
1300 void	vnevent_rmdir(vnode_t *, vnode_t *, char *, caller_context_t *);
1301 void	vnevent_create(vnode_t *, caller_context_t *);
1302 void	vnevent_link(vnode_t *, caller_context_t *);
1303 void	vnevent_rename_dest_dir(vnode_t *, caller_context_t *ct);
1304 void	vnevent_mountedover(vnode_t *, caller_context_t *);
1305 void	vnevent_truncate(vnode_t *, caller_context_t *);
1306 int	vnevent_support(vnode_t *, caller_context_t *);
1307 void	vnevent_pre_rename_src(vnode_t *, vnode_t *, char *,
1308 	    caller_context_t *);
1309 void	vnevent_pre_rename_dest(vnode_t *, vnode_t *, char *,
1310 	    caller_context_t *);
1311 void	vnevent_pre_rename_dest_dir(vnode_t *, vnode_t *, char *,
1312 	    caller_context_t *);
1313 
1314 /* Vnode specific data */
1315 void vsd_create(uint_t *, void (*)(void *));
1316 void vsd_destroy(uint_t *);
1317 void *vsd_get(vnode_t *, uint_t);
1318 int vsd_set(vnode_t *, uint_t, void *);
1319 void vsd_free(vnode_t *);
1320 
1321 /*
1322  * Extensible vnode attribute (xva) routines:
1323  * xva_init() initializes an xvattr_t (zero struct, init mapsize, set AT_XATTR)
1324  * xva_getxoptattr() returns a ponter to the xoptattr_t section of xvattr_t
1325  */
1326 void		xva_init(xvattr_t *);
1327 xoptattr_t	*xva_getxoptattr(xvattr_t *);	/* Get ptr to xoptattr_t */
1328 
1329 void xattr_init(void);		/* Initialize vnodeops for xattrs */
1330 
1331 /* GFS tunnel for xattrs */
1332 int xattr_dir_lookup(vnode_t *, vnode_t **, int, cred_t *);
1333 
1334 /* Reparse Point */
1335 void reparse_point_init(void);
1336 
1337 /* Context identification */
1338 u_longlong_t	fs_new_caller_id();
1339 
1340 int	vn_vmpss_usepageio(vnode_t *);
1341 
1342 /*
1343  * Needed for use of IS_VMODSORT() in kernel.
1344  */
1345 extern uint_t pvn_vmodsort_supported;
1346 
1347 /*
1348  * All changes to v_count should be done through VN_HOLD() or VN_RELE(), or
1349  * one of their variants. This makes it possible to ensure proper locking,
1350  * and to guarantee that all modifications are accompanied by a firing of
1351  * the vn-hold or vn-rele SDT DTrace probe.
1352  *
1353  * Example DTrace command for tracing vnode references using these probes:
1354  *
1355  * dtrace -q -n 'sdt:::vn-hold,sdt:::vn-rele
1356  * {
1357  *	this->vp = (vnode_t *)arg0;
1358  *	printf("%s %s(%p[%s]) %d\n", execname, probename, this->vp,
1359  *	    this->vp->v_path == NULL ? "NULL" : stringof(this->vp->v_path),
1360  *	    this->vp->v_count)
1361  * }'
1362  */
1363 #define	VN_HOLD_LOCKED(vp) {			\
1364 	ASSERT(mutex_owned(&(vp)->v_lock));	\
1365 	(vp)->v_count++;			\
1366 	DTRACE_PROBE1(vn__hold, vnode_t *, vp);	\
1367 }
1368 
1369 #define	VN_HOLD(vp)	{		\
1370 	mutex_enter(&(vp)->v_lock);	\
1371 	VN_HOLD_LOCKED(vp);		\
1372 	mutex_exit(&(vp)->v_lock);	\
1373 }
1374 
1375 #define	VN_RELE(vp)	{ \
1376 	vn_rele(vp); \
1377 }
1378 
1379 #define	VN_RELE_ASYNC(vp, taskq)	{ \
1380 	vn_rele_async(vp, taskq); \
1381 }
1382 
1383 #define	VN_RELE_LOCKED(vp) {			\
1384 	ASSERT(mutex_owned(&(vp)->v_lock));	\
1385 	ASSERT((vp)->v_count >= 1);		\
1386 	(vp)->v_count--;			\
1387 	DTRACE_PROBE1(vn__rele, vnode_t *, vp);	\
1388 }
1389 
1390 #define	VN_SET_VFS_TYPE_DEV(vp, vfsp, type, dev)	{ \
1391 	(vp)->v_vfsp = (vfsp); \
1392 	(vp)->v_type = (type); \
1393 	(vp)->v_rdev = (dev); \
1394 }
1395 
1396 /*
1397  * Compare two vnodes for equality.  In general this macro should be used
1398  * in preference to calling VOP_CMP directly.
1399  */
1400 #define	VN_CMP(VP1, VP2)	((VP1) == (VP2) ? 1 : 	\
1401 	((VP1) && (VP2) && (vn_getops(VP1) == vn_getops(VP2)) ? \
1402 	VOP_CMP(VP1, VP2, NULL) : 0))
1403 
1404 /*
1405  * Some well-known global vnodes used by the VM system to name pages.
1406  */
1407 extern struct vnode kvps[];
1408 
1409 typedef enum {
1410 	KV_KVP,		/* vnode for all segkmem pages */
1411 	KV_ZVP,		/* vnode for all ZFS pages */
1412 #if defined(__sparc)
1413 	KV_MPVP,	/* vnode for all page_t meta-pages */
1414 	KV_PROMVP,	/* vnode for all PROM pages */
1415 #endif	/* __sparc */
1416 	KV_MAX		/* total number of vnodes in kvps[] */
1417 } kvps_index_t;
1418 
1419 #define	VN_ISKAS(vp)	((vp) >= &kvps[0] && (vp) < &kvps[KV_MAX])
1420 
1421 #endif	/* _KERNEL */
1422 
1423 /*
1424  * Flags to VOP_SETATTR/VOP_GETATTR.
1425  */
1426 #define	ATTR_UTIME	0x01	/* non-default utime(2) request */
1427 #define	ATTR_EXEC	0x02	/* invocation from exec(2) */
1428 #define	ATTR_COMM	0x04	/* yield common vp attributes */
1429 #define	ATTR_HINT	0x08	/* information returned will be `hint' */
1430 #define	ATTR_REAL	0x10	/* yield attributes of the real vp */
1431 #define	ATTR_NOACLCHECK	0x20	/* Don't check ACL when checking permissions */
1432 #define	ATTR_TRIGGER	0x40	/* Mount first if vnode is a trigger mount */
1433 /*
1434  * Generally useful macros.
1435  */
1436 #define	VBSIZE(vp)	((vp)->v_vfsp->vfs_bsize)
1437 
1438 #define	VTOZONE(vp)	((vp)->v_vfsp->vfs_zone)
1439 
1440 #define	NULLVP		((struct vnode *)0)
1441 #define	NULLVPP		((struct vnode **)0)
1442 
1443 #ifdef	_KERNEL
1444 
1445 /*
1446  * Structure used while handling asynchronous VOP_PUTPAGE operations.
1447  */
1448 struct async_reqs {
1449 	struct async_reqs *a_next;	/* pointer to next arg struct */
1450 	struct vnode *a_vp;		/* vnode pointer */
1451 	u_offset_t a_off;			/* offset in file */
1452 	uint_t a_len;			/* size of i/o request */
1453 	int a_flags;			/* flags to indicate operation type */
1454 	struct cred *a_cred;		/* cred pointer	*/
1455 	ushort_t a_prealloced;		/* set if struct is pre-allocated */
1456 };
1457 
1458 /*
1459  * VN_DISPOSE() -- given a page pointer, safely invoke VOP_DISPOSE().
1460  * Note that there is no guarantee that the page passed in will be
1461  * freed.  If that is required, then a check after calling VN_DISPOSE would
1462  * be necessary to ensure the page was freed.
1463  */
1464 #define	VN_DISPOSE(pp, flag, dn, cr)	{ \
1465 	if ((pp)->p_vnode != NULL && !VN_ISKAS((pp)->p_vnode)) \
1466 		VOP_DISPOSE((pp)->p_vnode, (pp), (flag), (dn), (cr), NULL); \
1467 	else if ((flag) == B_FREE) \
1468 		page_free((pp), (dn)); \
1469 	else \
1470 		page_destroy((pp), (dn)); \
1471 	}
1472 
1473 #endif	/* _KERNEL */
1474 
1475 #ifdef	__cplusplus
1476 }
1477 #endif
1478 
1479 #endif	/* _SYS_VNODE_H */
1480