vnode.h revision 99485
174298Ssos/*
2154705Ssos * Copyright (c) 1989, 1993
374298Ssos *	The Regents of the University of California.  All rights reserved.
474298Ssos *
574298Ssos * Redistribution and use in source and binary forms, with or without
674298Ssos * modification, are permitted provided that the following conditions
774298Ssos * are met:
874298Ssos * 1. Redistributions of source code must retain the above copyright
974298Ssos *    notice, this list of conditions and the following disclaimer.
1074298Ssos * 2. Redistributions in binary form must reproduce the above copyright
1174298Ssos *    notice, this list of conditions and the following disclaimer in the
1274298Ssos *    documentation and/or other materials provided with the distribution.
1374298Ssos * 3. All advertising materials mentioning features or use of this software
1474298Ssos *    must display the following acknowledgement:
1574298Ssos *	This product includes software developed by the University of
1674298Ssos *	California, Berkeley and its contributors.
1774298Ssos * 4. Neither the name of the University nor the names of its contributors
1874298Ssos *    may be used to endorse or promote products derived from this software
1974298Ssos *    without specific prior written permission.
2074298Ssos *
2174298Ssos * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
2274298Ssos * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
2374298Ssos * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2474298Ssos * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
2574298Ssos * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
2674298Ssos * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2774298Ssos * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
2874298Ssos * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2974298Ssos * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
3074298Ssos * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
3174298Ssos * SUCH DAMAGE.
3274298Ssos *
3374298Ssos *	@(#)vnode.h	8.7 (Berkeley) 2/4/94
34119404Ssos * $FreeBSD: head/sys/sys/vnode.h 99485 2002-07-06 05:21:12Z jeff $
3593881Ssos */
36146266Ssos
37146266Ssos#ifndef _SYS_VNODE_H_
38146266Ssos#define	_SYS_VNODE_H_
39146266Ssos
40146266Ssos/*
41146266Ssos * XXX - compatability until lockmgr() goes away or all the #includes are
42146266Ssos * updated.
43146266Ssos */
44146266Ssos#include <sys/lockmgr.h>
45146266Ssos
46146266Ssos#include <sys/queue.h>
47146266Ssos#include <sys/_lock.h>
48146266Ssos#include <sys/_mutex.h>
49146266Ssos#include <sys/selinfo.h>
5093881Ssos#include <sys/uio.h>
51146266Ssos#include <sys/acl.h>
52146266Ssos#include <vm/uma.h>
53146266Ssos
54146266Ssos/*
55146266Ssos * The vnode is the focus of all file activity in UNIX.  There is a
56146266Ssos * unique vnode allocated for each active file, each current directory,
57146266Ssos * each mounted-on file, text file, and the root.
58146266Ssos */
59146266Ssos
60146266Ssos/*
61146266Ssos * Vnode types.  VNON means no type.
62146266Ssos */
63146266Ssosenum vtype	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD };
64146266Ssos
65146266Ssos/*
66146266Ssos * Vnode tag types.
67146266Ssos * These are for the benefit of external programs only (e.g., pstat)
68146266Ssos * and should NEVER be inspected by the kernel.
69146266Ssos */
7093881Ssosenum vtagtype	{
71146266Ssos	VT_NON, VT_UFS, VT_NFS, VT_UNUSED, VT_PC, VT_LFS, VT_LOFS, VT_FDESC,
72146266Ssos	VT_PORTAL, VT_NULL, VT_UMAP, VT_KERNFS, VT_PROCFS, VT_AFS, VT_ISOFS,
73146266Ssos	VT_UNION, VT_MSDOSFS, VT_DEVFS, VT_TFS, VT_VFS, VT_CODA, VT_NTFS,
7474298Ssos	VT_HPFS, VT_NWFS, VT_PSEUDOFS, VT_SMBFS, VT_UDF
75146266Ssos};
76146266Ssos
7774298Ssos/*
78146266Ssos * Each underlying filesystem allocates its own private area and hangs
79146266Ssos * it from v_data.  If non-null, this area is freed in getnewvnode().
80146266Ssos */
81146266SsosTAILQ_HEAD(buflists, buf);
8274298Ssos
83146266Ssostypedef	int	vop_t(void *);
84146266Ssosstruct namecache;
85146266Ssos
86146266Ssosstruct vpollinfo {
87146266Ssos	struct	mtx vpi_lock;		/* lock to protect below */
88146266Ssos	struct	selinfo vpi_selinfo;	/* identity of poller(s) */
89146266Ssos	short	vpi_events;		/* what they are looking for */
9093881Ssos	short	vpi_revents;		/* what has happened */
91146266Ssos};
92146266Ssos
93146266Ssos/*
94146266Ssos * Reading or writing any of these items requires holding the appropriate lock.
95146266Ssos * v_freelist is locked by the global vnode_free_list mutex.
9674298Ssos * v_mntvnodes is locked by the global mntvnodes mutex.
97146266Ssos * v_flag, v_usecount, v_holdcount and v_writecount are
98146266Ssos *    locked by the v_interlock mutex.
99146266Ssos * v_pollinfo is locked by the lock contained inside it.
100146266Ssos */
101146266Ssosstruct vnode {
102146266Ssos	u_long	v_flag;				/* vnode flags (see below) */
103146266Ssos	int	v_usecount;			/* reference count of users */
104146266Ssos	int	v_writecount;			/* reference count of writers */
105146266Ssos	int	v_holdcnt;			/* page & buffer references */
106146266Ssos	u_long	v_id;				/* capability identifier */
107146266Ssos	struct	mount *v_mount;			/* ptr to vfs we are in */
108146266Ssos	vop_t	**v_op;				/* vnode operations vector */
10974298Ssos	TAILQ_ENTRY(vnode) v_freelist;		/* vnode freelist */
110146266Ssos	TAILQ_ENTRY(vnode) v_nmntvnodes;	/* vnodes for mount point */
111146266Ssos	struct	buflists v_cleanblkhd;		/* clean blocklist head */
112146266Ssos	struct	buflists v_dirtyblkhd;		/* dirty blocklist head */
113146266Ssos	LIST_ENTRY(vnode) v_synclist;		/* vnodes with dirty buffers */
114146266Ssos	long	v_numoutput;			/* num of writes in progress */
115127021Ssos	enum	vtype v_type;			/* vnode type */
116146266Ssos	union {
117146266Ssos		struct mount	*vu_mountedhere;/* ptr to mounted vfs (VDIR) */
118146266Ssos		struct socket	*vu_socket;	/* unix ipc (VSOCK) */
119146266Ssos		struct {
120146266Ssos			struct specinfo	*vu_specinfo; /* device (VCHR, VBLK) */
121146266Ssos			SLIST_ENTRY(vnode) vu_specnext;
122146266Ssos		} vu_spec;
123127021Ssos		struct fifoinfo	*vu_fifoinfo;	/* fifo (VFIFO) */
124146266Ssos	} v_un;
125146266Ssos	daddr_t	v_lastw;			/* last write (write cluster) */
126119404Ssos	daddr_t	v_cstart;			/* start block of cluster */
12784584Ssos	daddr_t	v_lasta;			/* last allocation (cluster) */
128146266Ssos	int	v_clen;				/* length of current cluster */
129146266Ssos	struct vm_object *v_object;		/* Place to store VM object */
130146266Ssos	struct	mtx v_interlock;		/* lock on usecount and flag */
131146266Ssos	struct	lock v_lock;			/* used if fs don't have one */
132146266Ssos	struct	lock *v_vnlock;			/* pointer to vnode lock */
133146266Ssos	enum	vtagtype v_tag;			/* type of underlying data */
134146266Ssos	void	*v_data;			/* private data for fs */
135146266Ssos	LIST_HEAD(, namecache) v_cache_src;	/* Cache entries from us */
136146266Ssos	TAILQ_HEAD(, namecache) v_cache_dst;	/* Cache entries to us */
137146266Ssos	struct	vnode *v_dd;			/* .. vnode */
138146266Ssos	u_long	v_ddid;				/* .. capability identifier */
139146266Ssos	struct vpollinfo *v_pollinfo;
140146266Ssos	struct thread *v_vxproc;		/* thread owning VXLOCK */
141146266Ssos#ifdef	DEBUG_LOCKS
142146266Ssos	const char *filename;			/* Source file doing locking */
14384584Ssos	int line;				/* Line number doing locking */
144146266Ssos#endif
145146266Ssos};
146146266Ssos#define	v_mountedhere	v_un.vu_mountedhere
147146266Ssos#define	v_socket	v_un.vu_socket
148146266Ssos#define	v_rdev		v_un.vu_spec.vu_specinfo
149146266Ssos#define	v_specnext	v_un.vu_spec.vu_specnext
150146266Ssos#define	v_fifoinfo	v_un.vu_fifoinfo
151146266Ssos
152146266Ssos#define	VN_POLLEVENT(vp, events)				\
153146266Ssos	do {							\
154146266Ssos		if ((vp)->v_pollinfo != NULL && 		\
155146266Ssos		    (vp)->v_pollinfo->vpi_events & (events))	\
156146266Ssos			vn_pollevent((vp), (events));		\
157146266Ssos	} while (0)
15884584Ssos
159146266Ssos#define VN_KNOTE(vp, b)						\
160144330Ssos	do {							\
16184584Ssos		if ((vp)->v_pollinfo != NULL)			\
162146266Ssos			KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b)); \
163146266Ssos	} while (0)
164146266Ssos
165146266Ssos/*
166146266Ssos * Vnode flags.
167146266Ssos */
168146266Ssos#define	VROOT		0x00001	/* root of its filesystem */
16984584Ssos#define	VTEXT		0x00002	/* vnode is a pure text prototype */
170146266Ssos#define	VSYSTEM		0x00004	/* vnode being used by kernel */
171146266Ssos#define	VISTTY		0x00008	/* vnode represents a tty */
172146266Ssos#define	VXLOCK		0x00100	/* vnode is locked to change underlying type */
17384584Ssos#define	VXWANT		0x00200	/* thread is waiting for vnode */
174146266Ssos#define	VBWAIT		0x00400	/* waiting for output to complete */
175146266Ssos#define	VNOSYNC		0x01000	/* unlinked, stop syncing */
176146266Ssos/* open for business    0x01000 */
177146266Ssos#define	VOBJBUF		0x02000	/* Allocate buffers in VM object */
178146266Ssos#define	VCOPYONWRITE    0x04000 /* vnode is doing copy-on-write */
179146266Ssos#define	VAGE		0x08000	/* Insert vnode at head of free list */
180146266Ssos#define	VOLOCK		0x10000	/* vnode is locked waiting for an object */
181146266Ssos#define	VOWANT		0x20000	/* a thread is waiting for VOLOCK */
182146266Ssos#define	VDOOMED		0x40000	/* This vnode is being recycled */
183146266Ssos#define	VFREE		0x80000	/* This vnode is on the freelist */
184146266Ssos/* open for business	0x100000 */
185146266Ssos#define	VONWORKLST	0x200000 /* On syncer work-list */
186146266Ssos#define	VMOUNT		0x400000 /* Mount in progress */
187146266Ssos#define	VOBJDIRTY	0x800000 /* object might be dirty */
188146266Ssos
189146266Ssos/*
190146266Ssos * Vnode attributes.  A field value of VNOVAL represents a field whose value
191144330Ssos * is unavailable (getattr) or which is not to be changed (setattr).
19274298Ssos */
193146266Ssosstruct vattr {
194119404Ssos	enum vtype	va_type;	/* vnode type (for create) */
195146266Ssos	u_short		va_mode;	/* files access mode and type */
196146266Ssos	short		va_nlink;	/* number of references to file */
197146266Ssos	uid_t		va_uid;		/* owner user id */
198146266Ssos	gid_t		va_gid;		/* owner group id */
199146266Ssos	udev_t		va_fsid;	/* filesystem id */
200146266Ssos	long		va_fileid;	/* file id */
201146266Ssos	u_quad_t	va_size;	/* file size in bytes */
202146266Ssos	long		va_blocksize;	/* blocksize preferred for i/o */
203146266Ssos	struct timespec	va_atime;	/* time of last access */
204146266Ssos	struct timespec	va_mtime;	/* time of last modification */
205146266Ssos	struct timespec	va_ctime;	/* time file changed */
206146266Ssos	struct timespec	va_createtime;	/* time file created */
207146266Ssos	u_long		va_gen;		/* generation number of file */
208146266Ssos	u_long		va_flags;	/* flags defined for file */
209146266Ssos	udev_t		va_rdev;	/* device the special file represents */
210146266Ssos	u_quad_t	va_bytes;	/* bytes of disk space held by file */
211146266Ssos	u_quad_t	va_filerev;	/* file modification number */
212146266Ssos	u_int		va_vaflags;	/* operations flags, see below */
213146266Ssos	long		va_spare;	/* remain quad aligned */
214146266Ssos};
215146266Ssos
216148991Ssos/*
217146266Ssos * Flags for va_vaflags.
218156325Ssos */
219156325Ssos#define	VA_UTIMES_NULL	0x01		/* utimes argument was NULL */
220156325Ssos#define	VA_EXCLUSIVE	0x02		/* exclusive create request */
22174298Ssos
222146266Ssos/*
223119404Ssos * Flags for ioflag. (high 16 bits used to ask for read-ahead and
224146266Ssos * help with write clustering)
225146266Ssos */
226146266Ssos#define	IO_UNIT		0x01		/* do I/O as atomic unit */
227146266Ssos#define	IO_APPEND	0x02		/* append write to end */
228146266Ssos#define	IO_SYNC		0x04		/* do I/O synchronously */
229146266Ssos#define	IO_NODELOCKED	0x08		/* underlying node already locked */
230146266Ssos#define	IO_NDELAY	0x10		/* FNDELAY flag set in file table */
231146266Ssos#define	IO_VMIO		0x20		/* data already in VMIO space */
232152270Ssos#define	IO_INVAL	0x40		/* invalidate after I/O */
233146266Ssos#define	IO_ASYNC	0x80		/* bawrite rather then bdwrite */
234146266Ssos#define IO_DIRECT	0x100		/* attempt to bypass buffer cache */
235146266Ssos#define IO_NOWDRAIN	0x200		/* do not block on wdrain */
236146266Ssos
237146266Ssos/*
238152270Ssos *  Modes.  Some values same as Ixxx entries from inode.h for now.
239146266Ssos */
240146266Ssos#define	VADMIN	010000		/* permission to administer vnode */
241146266Ssos#define	VSUID	004000		/* set user id on execution */
242146266Ssos#define	VSGID	002000		/* set group id on execution */
243146266Ssos#define	VSVTX	001000		/* save swapped text even after use */
244146266Ssos#define	VREAD	000400		/* read, write, execute permissions */
245146266Ssos#define	VWRITE	000200
246146266Ssos#define	VEXEC	000100
247146266Ssos
248146266Ssos/*
249146266Ssos * Token indicating no attribute value yet assigned.
250146266Ssos */
251146266Ssos#define	VNOVAL	(-1)
252146266Ssos
253146266Ssos/*
254146266Ssos * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon)
255146266Ssos */
256146266Ssos#define VLKTIMEOUT	(hz / 20 + 1)
257146266Ssos
258146266Ssos#ifdef _KERNEL
259146266Ssos
260146266Ssos#ifdef MALLOC_DECLARE
261146266SsosMALLOC_DECLARE(M_VNODE);
262146266Ssos#endif
263146266Ssos
264146266Ssos/*
265146266Ssos * Convert between vnode types and inode formats (since POSIX.1
266146266Ssos * defines mode word of stat structure in terms of inode formats).
267146266Ssos */
268146266Ssosextern enum vtype	iftovt_tab[];
269146266Ssosextern int		vttoif_tab[];
270146266Ssos#define	IFTOVT(mode)	(iftovt_tab[((mode) & S_IFMT) >> 12])
271146266Ssos#define	VTTOIF(indx)	(vttoif_tab[(int)(indx)])
272146266Ssos#define	MAKEIMODE(indx, mode)	(int)(VTTOIF(indx) | (mode))
273146266Ssos
274146266Ssos/*
275146266Ssos * Flags to various vnode functions.
276119404Ssos */
277146266Ssos#define	SKIPSYSTEM	0x0001	/* vflush: skip vnodes marked VSYSTEM */
278119404Ssos#define	FORCECLOSE	0x0002	/* vflush: force file closure */
279146266Ssos#define	WRITECLOSE	0x0004	/* vflush: only close writable files */
280146266Ssos#define	DOCLOSE		0x0008	/* vclean: close active files */
281146266Ssos#define	V_SAVE		0x0001	/* vinvalbuf: sync file first */
282146266Ssos#define	REVOKEALL	0x0001	/* vop_revoke: revoke all aliases */
283146266Ssos#define	V_WAIT		0x0001	/* vn_start_write: sleep for suspend */
284146266Ssos#define	V_NOWAIT	0x0002	/* vn_start_write: don't sleep for suspend */
285146266Ssos#define	V_XSLEEP	0x0004	/* vn_start_write: just return after sleep */
286146266Ssos
287146266Ssos#define	VREF(vp)	vref(vp)
288146266Ssos
289146266Ssos
290156317Ssos#ifdef DIAGNOSTIC
291146266Ssos#define	VATTR_NULL(vap)	vattr_null(vap)
292146266Ssos#else
293146266Ssos#define	VATTR_NULL(vap)	(*(vap) = va_null)	/* initialize a vattr */
294146266Ssos#endif /* DIAGNOSTIC */
295146266Ssos
296146266Ssos#define	NULLVP	((struct vnode *)NULL)
297146266Ssos
298146266Ssos#define	VNODEOP_SET(f) \
299146266Ssos	C_SYSINIT(f##init, SI_SUB_VFS, SI_ORDER_SECOND, vfs_add_vnodeops, &f); \
300146266Ssos	C_SYSUNINIT(f##uninit, SI_SUB_VFS, SI_ORDER_SECOND, vfs_rm_vnodeops, &f);
301146266Ssos
302146266Ssos/*
303146266Ssos * Global vnode data.
304146266Ssos */
305146266Ssosextern	struct vnode *rootvnode;	/* root (i.e. "/") vnode */
306146266Ssosextern	int desiredvnodes;		/* number of vnodes desired */
307146266Ssosextern	uma_zone_t namei_zone;
308146266Ssosextern	int prtactive;			/* nonzero to call vprint() */
309146266Ssosextern	struct vattr va_null;		/* predefined null vattr structure */
310146266Ssosextern	int vfs_ioopt;
311146266Ssos
312146266Ssos/*
313146266Ssos * Macro/function to check for client cache inconsistency w.r.t. leasing.
314146266Ssos */
315146266Ssos#define	LEASE_READ	0x1		/* Check lease for readers */
316146266Ssos#define	LEASE_WRITE	0x2		/* Check lease for modifiers */
317146266Ssos
318146266Ssos
319146266Ssosextern void	(*lease_updatetime)(int deltat);
320146266Ssos
321146266Ssos#define	VSHOULDFREE(vp)	\
322146266Ssos	(!((vp)->v_flag & (VFREE|VDOOMED)) && \
323146266Ssos	 !(vp)->v_holdcnt && !(vp)->v_usecount && \
324146266Ssos	 (!(vp)->v_object || \
325156317Ssos	  !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
326146266Ssos
327146266Ssos#define VMIGHTFREE(vp) \
328146266Ssos	(!((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) &&	\
329146266Ssos	 LIST_EMPTY(&(vp)->v_cache_src) && !(vp)->v_usecount)
330146266Ssos
331146266Ssos#define	VSHOULDBUSY(vp)	\
332146266Ssos	(((vp)->v_flag & VFREE) && \
333146266Ssos	 ((vp)->v_holdcnt || (vp)->v_usecount))
334146266Ssos
335146266Ssos#define	VI_LOCK(vp)	mtx_lock(&(vp)->v_interlock)
336146266Ssos#define	VI_TRYLOCK(vp)	mtx_trylock(&(vp)->v_interlock)
337119404Ssos#define	VI_UNLOCK(vp)	mtx_unlock(&(vp)->v_interlock)
33874298Ssos
339146266Ssos#endif /* _KERNEL */
340146266Ssos
341146266Ssos
342146266Ssos/*
343146266Ssos * Mods for extensibility.
344119404Ssos */
345146266Ssos
346146266Ssos/*
347146266Ssos * Flags for vdesc_flags:
348146266Ssos */
349146266Ssos#define	VDESC_MAX_VPS		16
350146266Ssos/* Low order 16 flag bits are reserved for willrele flags for vp arguments. */
351119404Ssos#define	VDESC_VP0_WILLRELE	0x0001
352157329Ssos#define	VDESC_VP1_WILLRELE	0x0002
353157329Ssos#define	VDESC_VP2_WILLRELE	0x0004
354157329Ssos#define	VDESC_VP3_WILLRELE	0x0008
355157329Ssos#define	VDESC_NOMAP_VPP		0x0100
356157329Ssos#define	VDESC_VPP_WILLRELE	0x0200
357157329Ssos
358157329Ssos/*
359157329Ssos * VDESC_NO_OFFSET is used to identify the end of the offset list
360157329Ssos * and in places where no such field exists.
361157329Ssos */
362157329Ssos#define VDESC_NO_OFFSET -1
363157329Ssos
364157329Ssos/*
365157329Ssos * This structure describes the vnode operation taking place.
366157329Ssos */
367157329Ssosstruct vnodeop_desc {
368157329Ssos	int	 vdesc_offset;		/* offset in vector,first for speed */
369157329Ssos	char	*vdesc_name;		/* a readable name for debugging */
370157329Ssos	int	 vdesc_flags;		/* VDESC_* flags */
371157329Ssos
372157329Ssos	/*
373157329Ssos	 * These ops are used by bypass routines to map and locate arguments.
374157329Ssos	 * Creds and procs are not needed in bypass routines, but sometimes
375157329Ssos	 * they are useful to (for example) transport layers.
376157329Ssos	 * Nameidata is useful because it has a cred in it.
377157329Ssos	 */
378157329Ssos	int	*vdesc_vp_offsets;	/* list ended by VDESC_NO_OFFSET */
379157329Ssos	int	vdesc_vpp_offset;	/* return vpp location */
380157329Ssos	int	vdesc_cred_offset;	/* cred location, if any */
381157329Ssos	int	vdesc_thread_offset;	/* thread location, if any */
382157329Ssos	int	vdesc_componentname_offset; /* if any */
383157329Ssos	/*
384157329Ssos	 * Finally, we've got a list of private data (about each operation)
385157329Ssos	 * for each transport layer.  (Support to manage this list is not
386157329Ssos	 * yet part of BSD.)
387157329Ssos	 */
388157329Ssos	caddr_t	*vdesc_transports;
389157329Ssos};
390157329Ssos
391157329Ssos#ifdef _KERNEL
392157329Ssos/*
393157329Ssos * A list of all the operation descs.
394146266Ssos */
395146266Ssosextern struct vnodeop_desc *vnodeop_descs[];
396119404Ssos
397146266Ssos/*
398146266Ssos * Interlock for scanning list of vnodes attached to a mountpoint
399146266Ssos */
400146266Ssosextern struct mtx mntvnode_mtx;
401146266Ssos
402119404Ssos/*
403146266Ssos * This macro is very helpful in defining those offsets in the vdesc struct.
404157329Ssos *
405146266Ssos * This is stolen from X11R4.  I ignored all the fancy stuff for
406146266Ssos * Crays, so if you decide to port this to such a serious machine,
407146266Ssos * you might want to consult Intrinsic.h's XtOffset{,Of,To}.
408146266Ssos */
409146266Ssos#define	VOPARG_OFFSET(p_type,field) \
410146266Ssos	((int) (((char *) (&(((p_type)NULL)->field))) - ((char *) NULL)))
411146266Ssos#define	VOPARG_OFFSETOF(s_type,field) \
412146266Ssos	VOPARG_OFFSET(s_type*,field)
413146266Ssos#define	VOPARG_OFFSETTO(S_TYPE,S_OFFSET,STRUCT_P) \
414119404Ssos	((S_TYPE)(((char*)(STRUCT_P))+(S_OFFSET)))
415146266Ssos
416146266Ssos
417146266Ssos/*
418119404Ssos * This structure is used to configure the new vnodeops vector.
419146266Ssos */
420146266Ssosstruct vnodeopv_entry_desc {
421146266Ssos	struct vnodeop_desc *opve_op;   /* which operation this is */
422146266Ssos	vop_t *opve_impl;		/* code implementing this operation */
423146266Ssos};
42493662Ssosstruct vnodeopv_desc {
425119404Ssos			/* ptr to the ptr to the vector where op should go */
426146266Ssos	vop_t ***opv_desc_vector_p;
427146266Ssos	struct vnodeopv_entry_desc *opv_desc_ops;   /* null terminated list */
428146266Ssos};
429146266Ssos
430146266Ssos/*
431146266Ssos * A generic structure.
432146266Ssos * This can be used by bypass routines to identify generic arguments.
433146266Ssos */
434146266Ssosstruct vop_generic_args {
435146266Ssos	struct vnodeop_desc *a_desc;
436146266Ssos	/* other random data follows, presumably */
43793662Ssos};
438146266Ssos
439146266Ssos
440146266Ssos#ifdef DEBUG_VFS_LOCKS
441146266Ssos/*
442146266Ssos * Macros to aid in tracing VFS locking problems.  Not totally
443119404Ssos * reliable since if the thread sleeps between changing the lock
444146266Ssos * state and checking it with the assert, some other thread could
445146266Ssos * change the state.  They are good enough for debugging a single
446146266Ssos * filesystem using a single-threaded test.  I find that 'cvs co src'
44774298Ssos * is a pretty good test.
44874298Ssos */
449146266Ssos
450148737Ssosextern int vfs_badlock_panic;
451146266Ssosextern int vfs_badlock_print;
452148729Ssos
453146266Ssos/*
454146266Ssos * [dfr] Kludge until I get around to fixing all the vfs locking.
45574298Ssos */
45674298Ssos#define IS_LOCKING_VFS(vp)	(  ((vp)->v_tag == VT_UFS		\
457				 || (vp)->v_tag == VT_NFS		\
458				 || (vp)->v_tag == VT_LFS		\
459				 || (vp)->v_tag == VT_ISOFS		\
460				 || (vp)->v_tag == VT_MSDOSFS		\
461				 || (vp)->v_tag == VT_DEVFS		\
462				 || (vp)->v_tag == VT_UDF)		\
463				 && ((vp)->v_type != VBLK		\
464				 && (vp)->v_type != VCHR) )
465
466#define ASSERT_VOP_LOCKED(vp, str)					\
467do {									\
468	struct vnode *_vp = (vp);					\
469									\
470	if (_vp && IS_LOCKING_VFS(_vp) && !VOP_ISLOCKED(_vp, NULL)) {	\
471		if (vfs_badlock_print)					\
472			printf("%s: %p is not locked but should be\n",	\
473			    str, _vp);					\
474		if (vfs_badlock_panic)					\
475			Debugger("Lock violation.\n");			\
476	}								\
477} while (0)
478
479#define ASSERT_VOP_UNLOCKED(vp, str)					\
480do {									\
481	struct vnode *_vp = (vp);					\
482	int lockstate;							\
483									\
484	if (_vp && IS_LOCKING_VFS(_vp)) {				\
485		lockstate = VOP_ISLOCKED(_vp, curthread);		\
486		if (lockstate == LK_EXCLUSIVE) {			\
487			if (vfs_badlock_print)				\
488				printf("%s: %p is locked but should not be\n",	\
489				    str, _vp);				\
490			if (vfs_badlock_panic)				\
491				Debugger("Lock Violation.\n");		\
492		}							\
493	}								\
494} while (0)
495
496#define ASSERT_VOP_ELOCKED(vp, str)					\
497do {									\
498	struct vnode *_vp = (vp);					\
499									\
500	if (_vp && IS_LOCKING_VFS(_vp) &&				\
501	    VOP_ISLOCKED(_vp, curthread) != LK_EXCLUSIVE)			\
502		panic("%s: %p is not exclusive locked but should be",	\
503		    str, _vp);						\
504} while (0)
505
506#define ASSERT_VOP_ELOCKED_OTHER(vp, str)				\
507do {									\
508	struct vnode *_vp = (vp);					\
509									\
510	if (_vp && IS_LOCKING_VFS(_vp) &&				\
511	    VOP_ISLOCKED(_vp, curthread) != LK_EXCLOTHER)			\
512		panic("%s: %p is not exclusive locked by another thread",	\
513		    str, _vp);						\
514} while (0)
515
516#define ASSERT_VOP_SLOCKED(vp, str)					\
517do {									\
518	struct vnode *_vp = (vp);					\
519									\
520	if (_vp && IS_LOCKING_VFS(_vp) &&				\
521	    VOP_ISLOCKED(_vp, NULL) != LK_SHARED)			\
522		panic("%s: %p is not locked shared but should be",	\
523		    str, _vp);						\
524} while (0)
525
526void vop_rename_pre(void *a);
527void vop_strategy_pre(void *a);
528
529#else
530
531#define ASSERT_VOP_LOCKED(vp, str)
532#define ASSERT_VOP_UNLOCKED(vp, str)
533
534#endif
535
536/*
537 * VOCALL calls an op given an ops vector.  We break it out because BSD's
538 * vclean changes the ops vector and then wants to call ops with the old
539 * vector.
540 */
541#define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP))
542
543/*
544 * This call works for vnodes in the kernel.
545 */
546#define VCALL(VP,OFF,AP) VOCALL((VP)->v_op,(OFF),(AP))
547#define VDESC(OP) (& __CONCAT(OP,_desc))
548#define VOFFSET(OP) (VDESC(OP)->vdesc_offset)
549
550/*
551 * VMIO support inline
552 */
553
554extern int vmiodirenable;
555
556static __inline int
557vn_canvmio(struct vnode *vp)
558{
559      if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR)))
560		return(TRUE);
561	return(FALSE);
562}
563
564/*
565 * Finally, include the default set of vnode operations.
566 */
567#include "vnode_if.h"
568
569/*
570 * Public vnode manipulation functions.
571 */
572struct componentname;
573struct file;
574struct mount;
575struct nameidata;
576struct ostat;
577struct thread;
578struct proc;
579struct stat;
580struct nstat;
581struct ucred;
582struct uio;
583struct vattr;
584struct vnode;
585
586extern int	(*lease_check_hook)(struct vop_lease_args *);
587extern int	(*softdep_fsync_hook)(struct vnode *);
588extern int	(*softdep_process_worklist_hook)(struct mount *);
589
590struct	vnode *addaliasu(struct vnode *vp, udev_t nvp_rdev);
591int	bdevvp(dev_t dev, struct vnode **vpp);
592/* cache_* may belong in namei.h. */
593void	cache_enter(struct vnode *dvp, struct vnode *vp,
594	    struct componentname *cnp);
595int	cache_lookup(struct vnode *dvp, struct vnode **vpp,
596	    struct componentname *cnp);
597void	cache_purge(struct vnode *vp);
598void	cache_purgevfs(struct mount *mp);
599int	cache_leaf_test(struct vnode *vp);
600void	cvtstat(struct stat *st, struct ostat *ost);
601void	cvtnstat(struct stat *sb, struct nstat *nsb);
602int	getnewvnode(enum vtagtype tag,
603	    struct mount *mp, vop_t **vops, struct vnode **vpp);
604int	lease_check(struct vop_lease_args *ap);
605int	spec_vnoperate(struct vop_generic_args *);
606int	speedup_syncer(void);
607#define textvp_fullpath(p, rb, rfb) \
608	vn_fullpath(FIRST_THREAD_IN_PROC(p), (p)->p_textvp, rb, rfb)
609int	vn_fullpath(struct thread *td, struct vnode *vn,
610	    char **retbuf, char **freebuf);
611int	vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
612	    mode_t acc_mode, struct ucred *cred, int *privused);
613int	vaccess_acl_posix1e(enum vtype type, uid_t file_uid,
614	    gid_t file_gid, struct acl *acl, mode_t acc_mode,
615	    struct ucred *cred, int *privused);
616void	vattr_null(struct vattr *vap);
617int	vcount(struct vnode *vp);
618void	vdrop(struct vnode *);
619int	vfinddev(dev_t dev, enum vtype type, struct vnode **vpp);
620void	vfs_add_vnodeops(const void *);
621void	vfs_rm_vnodeops(const void *);
622int	vflush(struct mount *mp, int rootrefs, int flags);
623int	vget(struct vnode *vp, int lockflag, struct thread *td);
624void	vgone(struct vnode *vp);
625void	vgonel(struct vnode *vp, struct thread *td);
626void	vhold(struct vnode *);
627int	vinvalbuf(struct vnode *vp, int save, struct ucred *cred,
628	    struct thread *td, int slpflag, int slptimeo);
629int	vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
630	    off_t length, int blksize);
631void	vprint(char *label, struct vnode *vp);
632int	vrecycle(struct vnode *vp, struct mtx *inter_lkp,
633	    struct thread *td);
634int	vn_close(struct vnode *vp,
635	    int flags, struct ucred *cred, struct thread *td);
636void	vn_finished_write(struct mount *mp);
637int	vn_isdisk(struct vnode *vp, int *errp);
638int	vn_lock(struct vnode *vp, int flags, struct thread *td);
639#ifdef	DEBUG_LOCKS
640int	debug_vn_lock(struct vnode *vp, int flags, struct thread *p,
641	    const char *filename, int line);
642#define vn_lock(vp,flags,p) debug_vn_lock(vp,flags,p,__FILE__,__LINE__)
643#endif
644int	vn_mkdir(char *path, int mode, enum uio_seg segflg, struct thread *td);
645int	vn_open(struct nameidata *ndp, int *flagp, int cmode);
646int	vn_open_cred(struct nameidata *ndp, int *flagp, int cmode,
647	    struct ucred *cred);
648void	vn_pollevent(struct vnode *vp, int events);
649void	vn_pollgone(struct vnode *vp);
650int	vn_pollrecord(struct vnode *vp, struct thread *p, int events);
651int	vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base,
652	    int len, off_t offset, enum uio_seg segflg, int ioflg,
653	    struct ucred *cred, int *aresid, struct thread *td);
654int	vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base,
655	    int len, off_t offset, enum uio_seg segflg, int ioflg,
656	    struct ucred *cred, int *aresid, struct thread *td);
657int	vn_stat(struct vnode *vp, struct stat *sb, struct thread *td);
658int	vn_start_write(struct vnode *vp, struct mount **mpp, int flags);
659dev_t	vn_todev(struct vnode *vp);
660int	vn_write_suspend_wait(struct vnode *vp, struct mount *mp,
661	    int flags);
662int	vn_writechk(struct vnode *vp);
663int	vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
664	    const char *attrname, int *buflen, char *buf, struct thread *td);
665int	vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
666	    const char *attrname, int buflen, char *buf, struct thread *td);
667int	vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
668	    const char *attrname, struct thread *td);
669int	vfs_cache_lookup(struct vop_lookup_args *ap);
670int	vfs_object_create(struct vnode *vp, struct thread *td,
671	    struct ucred *cred);
672void	vfs_timestamp(struct timespec *);
673void	vfs_write_resume(struct mount *mp);
674void	vfs_write_suspend(struct mount *mp);
675int	vop_stdbmap(struct vop_bmap_args *);
676int	vop_stdgetwritemount(struct vop_getwritemount_args *);
677int	vop_stdgetpages(struct vop_getpages_args *);
678int	vop_stdinactive(struct vop_inactive_args *);
679int	vop_stdislocked(struct vop_islocked_args *);
680int	vop_stdlock(struct vop_lock_args *);
681int	vop_stdputpages(struct vop_putpages_args *);
682int	vop_stdunlock(struct vop_unlock_args *);
683int	vop_noislocked(struct vop_islocked_args *);
684int	vop_nolock(struct vop_lock_args *);
685int	vop_nopoll(struct vop_poll_args *);
686int	vop_nounlock(struct vop_unlock_args *);
687int	vop_stdpathconf(struct vop_pathconf_args *);
688int	vop_stdpoll(struct vop_poll_args *);
689int	vop_revoke(struct vop_revoke_args *);
690int	vop_sharedlock(struct vop_lock_args *);
691int	vop_eopnotsupp(struct vop_generic_args *ap);
692int	vop_ebadf(struct vop_generic_args *ap);
693int	vop_einval(struct vop_generic_args *ap);
694int	vop_enotty(struct vop_generic_args *ap);
695int	vop_defaultop(struct vop_generic_args *ap);
696int	vop_null(struct vop_generic_args *ap);
697int	vop_panic(struct vop_generic_args *ap);
698int	vop_stdcreatevobject(struct vop_createvobject_args *ap);
699int	vop_stddestroyvobject(struct vop_destroyvobject_args *ap);
700int	vop_stdgetvobject(struct vop_getvobject_args *ap);
701
702void	vfree(struct vnode *);
703void	vput(struct vnode *vp);
704void	vrele(struct vnode *vp);
705void	vref(struct vnode *vp);
706void	vbusy(struct vnode *vp);
707void 	v_addpollinfo(struct vnode *vp);
708
709extern	vop_t **default_vnodeop_p;
710extern	vop_t **spec_vnodeop_p;
711extern	vop_t **dead_vnodeop_p;
712
713#endif /* _KERNEL */
714
715#endif /* !_SYS_VNODE_H_ */
716