vnode.h revision 133741
162587Sitojun/*
278064Sume * Copyright (c) 1989, 1993
362587Sitojun *	The Regents of the University of California.  All rights reserved.
452904Sshin *
552904Sshin * Redistribution and use in source and binary forms, with or without
652904Sshin * modification, are permitted provided that the following conditions
753541Sshin * are met:
852904Sshin * 1. Redistributions of source code must retain the above copyright
952904Sshin *    notice, this list of conditions and the following disclaimer.
1052904Sshin * 2. Redistributions in binary form must reproduce the above copyright
1152904Sshin *    notice, this list of conditions and the following disclaimer in the
1252904Sshin *    documentation and/or other materials provided with the distribution.
1352904Sshin * 4. Neither the name of the University nor the names of its contributors
1452904Sshin *    may be used to endorse or promote products derived from this software
1552904Sshin *    without specific prior written permission.
1652904Sshin *
1752904Sshin * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
1852904Sshin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1953541Sshin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2052904Sshin * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
2152904Sshin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
2252904Sshin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2352904Sshin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
2452904Sshin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2552904Sshin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
2652904Sshin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
2752904Sshin * SUCH DAMAGE.
2852904Sshin *
2952904Sshin *	@(#)vnode.h	8.7 (Berkeley) 2/4/94
3052904Sshin * $FreeBSD: head/sys/sys/vnode.h 133741 2004-08-15 06:24:42Z jmg $
3152904Sshin */
3252904Sshin
3352904Sshin#ifndef _SYS_VNODE_H_
3452904Sshin#define	_SYS_VNODE_H_
3552904Sshin
3652904Sshin/*
3752904Sshin * XXX - compatability until lockmgr() goes away or all the #includes are
3852904Sshin * updated.
3952904Sshin */
4052904Sshin#include <sys/lockmgr.h>
4152904Sshin
4252904Sshin#include <sys/queue.h>
4352904Sshin#include <sys/_lock.h>
4452904Sshin#include <sys/lock.h>
4552904Sshin#include <sys/_mutex.h>
4652904Sshin#include <sys/mutex.h>
4752904Sshin#include <sys/selinfo.h>
4852904Sshin#include <sys/uio.h>
4952904Sshin#include <sys/acl.h>
5052904Sshin#include <sys/ktr.h>
5152904Sshin
5252904Sshin/*
5352904Sshin * The vnode is the focus of all file activity in UNIX.  There is a
5452904Sshin * unique vnode allocated for each active file, each current directory,
5552904Sshin * each mounted-on file, text file, and the root.
5652904Sshin */
5752904Sshin
5852904Sshin/*
5952904Sshin * Vnode types.  VNON means no type.
6052904Sshin */
6152904Sshinenum vtype	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD };
6252904Sshin
6352904Sshin/*
6452904Sshin * Each underlying filesystem allocates its own private area and hangs
6552904Sshin * it from v_data.  If non-null, this area is freed in getnewvnode().
6652904Sshin */
6752904SshinTAILQ_HEAD(buflists, buf);
6862587Sitojun
6978064Sumetypedef	int	vop_t(void *);
7057120Sshinstruct namecache;
7157120Sshin
7262587Sitojunstruct vpollinfo {
7362587Sitojun	struct	mtx vpi_lock;		/* lock to protect below */
7462587Sitojun	struct	selinfo vpi_selinfo;	/* identity of poller(s) */
7552904Sshin	short	vpi_events;		/* what they are looking for */
7652904Sshin	short	vpi_revents;		/* what has happened */
7778064Sume};
7878064Sume
7952904Sshin/*
8062587Sitojun * Reading or writing any of these items requires holding the appropriate lock.
8178064Sume *
8252904Sshin * Lock reference:
8352904Sshin *	c - namecache mutex
8452904Sshin *	f - freelist mutex
8552904Sshin *	G - Giant
8652904Sshin *	i - interlock
8752904Sshin *	m - mntvnodes mutex
8852904Sshin *	p - pollinfo lock
8952904Sshin *	s - spechash mutex
9052904Sshin *	S - syncer mutex
9152904Sshin *	u - Only a reference to the vnode is needed to read.
9262587Sitojun *	v - vnode lock
9352904Sshin *
9452904Sshin * Vnodes may be found on many lists.  The general way to deal with operating
9552904Sshin * on a vnode that is on a list is:
9652904Sshin *	1) Lock the list and find the vnode.
9752904Sshin *	2) Lock interlock so that the vnode does not go away.
9852904Sshin *	3) Unlock the list to avoid lock order reversals.
9952904Sshin *	4) vget with LK_INTERLOCK and check for ENOENT, or
10052904Sshin *	5) Check for XLOCK if the vnode lock is not required.
10152904Sshin *	6) Perform your operation, then vput().
10252904Sshin *
10352904Sshin * XXX Not all fields are locked yet and some fields that are marked are not
10452904Sshin * locked consistently.  This is a work in progress.  Requires Giant!
10552904Sshin */
10652904Sshin
10752904Sshinstruct vnode {
10852904Sshin	struct	mtx v_interlock;		/* lock for "i" things */
10952904Sshin	u_long	v_iflag;			/* i vnode flags (see below) */
11097181Smike	int	v_usecount;			/* i ref count of users */
11152904Sshin	long	v_numoutput;			/* i writes in progress */
11252904Sshin	struct thread *v_vxthread;		/* i thread owning VXLOCK */
11352904Sshin	int	v_holdcnt;			/* i page & buffer references */
11452904Sshin	struct	buflists v_cleanblkhd;		/* i SORTED clean blocklist */
11552904Sshin	struct buf	*v_cleanblkroot;	/* i clean buf splay tree  */
11697181Smike	int	v_cleanbufcnt;			/* i number of clean buffers */
11752904Sshin	struct	buflists v_dirtyblkhd;		/* i SORTED dirty blocklist */
11852904Sshin	struct buf	*v_dirtyblkroot;	/* i dirty buf splay tree */
11952904Sshin	int	v_dirtybufcnt;			/* i number of dirty buffers */
12052904Sshin	u_long	v_vflag;			/* v vnode flags */
12152904Sshin	int	v_writecount;			/* v ref count of writers */
12252904Sshin	struct vm_object *v_object;		/* v Place to store VM object */
12397181Smike	daddr_t	v_lastw;			/* v last write (write cluster) */
12497181Smike	daddr_t	v_cstart;			/* v start block of cluster */
12597181Smike	daddr_t	v_lasta;			/* v last allocation (cluster) */
12652904Sshin	int	v_clen;				/* v length of current cluster */
12752904Sshin	union {
12852904Sshin		struct mount	*vu_mountedhere;/* v ptr to mounted vfs (VDIR) */
12962587Sitojun		struct socket	*vu_socket;	/* v unix ipc (VSOCK) */
13095023Ssuz		struct {
13162587Sitojun			struct cdev	*vu_cdev; /* v device (VCHR, VBLK) */
13262587Sitojun			SLIST_ENTRY(vnode) vu_specnext;	/* s device aliases */
13362587Sitojun		} vu_spec;
13452904Sshin		struct fifoinfo	*vu_fifoinfo;	/* v fifo (VFIFO) */
13552904Sshin	} v_un;
13662587Sitojun	TAILQ_ENTRY(vnode) v_freelist;		/* f vnode freelist */
13752904Sshin	TAILQ_ENTRY(vnode) v_nmntvnodes;	/* m vnodes for mount point */
13852904Sshin	LIST_ENTRY(vnode) v_synclist;		/* S dirty vnode list */
13997181Smike	enum	vtype v_type;			/* u vnode type */
14097181Smike	const char *v_tag;			/* u type of underlying data */
14197181Smike	void	*v_data;			/* u private data for fs */
14297181Smike	struct	lock v_lock;			/* u used if fs don't have one */
14352904Sshin	struct	lock *v_vnlock;			/* u pointer to vnode lock */
14452904Sshin	vop_t	**v_op;				/* u vnode operations vector */
14597181Smike	struct	mount *v_mount;			/* u ptr to vfs we are in */
14662587Sitojun	LIST_HEAD(, namecache) v_cache_src;	/* c Cache entries from us */
14752904Sshin	TAILQ_HEAD(, namecache) v_cache_dst;	/* c Cache entries to us */
14897181Smike	u_long	v_id;				/* c capability identifier */
14952904Sshin	struct	vnode *v_dd;			/* c .. vnode */
150100503Sume	u_long	v_ddid;				/* c .. capability identifier */
15197181Smike	struct vpollinfo *v_pollinfo;		/* G Poll events, p for *v_pi */
15297181Smike	struct label *v_label;			/* MAC label for vnode */
15397181Smike#ifdef	DEBUG_LOCKS
15462587Sitojun	const char *filename;			/* Source file doing locking */
15597181Smike	int line;				/* Line number doing locking */
15652904Sshin#endif
15752904Sshin	dev_t	v_cachedfs;			/* cached fs id */
15852904Sshin	ino_t	v_cachedid;			/* cached file id */
15952904Sshin	int	v_bsize;			/* block size for I/O */
16052904Sshin};
16195023Ssuz#define	v_mountedhere	v_un.vu_mountedhere
16262587Sitojun#define	v_socket	v_un.vu_socket
16362587Sitojun#define	v_rdev		v_un.vu_spec.vu_cdev
16452904Sshin#define	v_specnext	v_un.vu_spec.vu_specnext
16562587Sitojun#define	v_fifoinfo	v_un.vu_fifoinfo
16652904Sshin
16762587Sitojun/*
16852904Sshin * Userland version of struct vnode, for sysctl.
16962587Sitojun */
17052904Sshinstruct xvnode {
17152904Sshin	size_t	xv_size;			/* sizeof(struct xvnode) */
17252904Sshin	void	*xv_vnode;			/* address of real vnode */
17352904Sshin	u_long	xv_flag;			/* vnode vflags */
17478064Sume	int	xv_usecount;			/* reference count of users */
17578064Sume	int	xv_writecount;			/* reference count of writers */
17652904Sshin	int	xv_holdcnt;			/* page & buffer references */
17752904Sshin	u_long	xv_id;				/* capability identifier */
17852904Sshin	void	*xv_mount;			/* address of parent mount */
17952904Sshin	long	xv_numoutput;			/* num of writes in progress */
18052904Sshin	enum	vtype xv_type;			/* vnode type */
18152904Sshin	union {
18252904Sshin		void	*xvu_socket;		/* socket, if VSOCK */
18352904Sshin		void	*xvu_fifo;		/* fifo, if VFIFO */
18452904Sshin		dev_t	xvu_rdev;		/* maj/min, if VBLK/VCHR */
18552904Sshin		struct {
18695023Ssuz			dev_t	xvu_dev;	/* device, if VDIR/VREG/VLNK */
18797181Smike			ino_t	xvu_ino;	/* id, if VDIR/VREG/VLNK */
18862587Sitojun		} xv_uns;
18962587Sitojun	} xv_un;
19062587Sitojun};
19162587Sitojun#define xv_socket	xv_un.xvu_socket
19262587Sitojun#define xv_fifo		xv_un.xvu_fifo
19362587Sitojun#define xv_rdev		xv_un.xvu_rdev
19462587Sitojun#define xv_dev		xv_un.xv_uns.xvu_dev
19562587Sitojun#define xv_ino		xv_un.xv_uns.xvu_ino
19697181Smike
19762587Sitojun#define	VN_POLLEVENT(vp, events)				\
19862587Sitojun	do {							\
19962587Sitojun		if ((vp)->v_pollinfo != NULL && 		\
20062587Sitojun		    (vp)->v_pollinfo->vpi_events & (events))	\
20162587Sitojun			vn_pollevent((vp), (events));		\
20262587Sitojun	} while (0)
20362587Sitojun
20462587Sitojun#define VN_KNOTE(vp, b, a)						\
20552904Sshin	do {							\
20652904Sshin		if ((vp)->v_pollinfo != NULL)			\
20752904Sshin			KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b), (a)); \
20852904Sshin	} while (0)
20952904Sshin#define VN_KNOTE_LOCKED(vp, b)		VN_KNOTE(vp, b, 1)
21052904Sshin#define VN_KNOTE_UNLOCKED(vp, b)	VN_KNOTE(vp, b, 0)
21197181Smike
21262587Sitojun/*
21352904Sshin * Vnode flags.
21452904Sshin *	VI flags are protected by interlock and live in v_iflag
21562587Sitojun *	VV flags are protected by the vnode lock and live in v_vflag
21652904Sshin */
21752904Sshin#define	VI_XLOCK	0x0001	/* vnode is locked to change vtype */
21862587Sitojun#define	VI_XWANT	0x0002	/* thread is waiting for vnode */
21952904Sshin#define	VI_BWAIT	0x0004	/* waiting for output to complete */
22052904Sshin#define	VI_OLOCK	0x0008	/* vnode is locked waiting for an object */
22162587Sitojun#define	VI_OWANT	0x0010	/* a thread is waiting for VOLOCK */
22252904Sshin#define	VI_MOUNT	0x0020	/* Mount in progress */
22352904Sshin#define	VI_AGE		0x0040	/* Insert vnode at head of free list */
22462587Sitojun#define	VI_DOOMED	0x0080	/* This vnode is being recycled */
22552904Sshin#define	VI_FREE		0x0100	/* This vnode is on the freelist */
22652904Sshin#define	VI_OBJDIRTY	0x0400	/* object might be dirty */
22797181Smike#define	VI_DOINGINACT	0x0800	/* VOP_INACTIVE is in progress */
22852904Sshin/*
22952904Sshin * XXX VI_ONWORKLST could be replaced with a check for NULL list elements
23052904Sshin * in v_synclist.
23197181Smike */
23252904Sshin#define	VI_ONWORKLST	0x0200	/* On syncer work-list */
23352904Sshin
23452904Sshin#define	VV_ROOT		0x0001	/* root of its filesystem */
23597181Smike#define	VV_ISTTY	0x0002	/* vnode represents a tty */
23652904Sshin#define	VV_NOSYNC	0x0004	/* unlinked, stop syncing */
23752904Sshin#define	VV_OBJBUF	0x0008	/* Allocate buffers in VM object */
23852904Sshin#define	VV_CACHEDLABEL	0x0010	/* Vnode has valid cached MAC label */
23953877Sitojun#define	VV_TEXT		0x0020	/* vnode is a pure text prototype */
24053877Sitojun#define	VV_COPYONWRITE	0x0040	/* vnode is doing copy-on-write */
24153877Sitojun#define	VV_SYSTEM	0x0080	/* vnode being used by kernel */
24252904Sshin#define	VV_PROCDEP	0x0100	/* vnode is process dependent */
24353877Sitojun
24462587Sitojun/*
24562587Sitojun * Vnode attributes.  A field value of VNOVAL represents a field whose value
24653877Sitojun * is unavailable (getattr) or which is not to be changed (setattr).
24797181Smike */
24862587Sitojunstruct vattr {
24962587Sitojun	enum vtype	va_type;	/* vnode type (for create) */
25053877Sitojun	u_short		va_mode;	/* files access mode and type */
25197181Smike	short		va_nlink;	/* number of references to file */
25252904Sshin	uid_t		va_uid;		/* owner user id */
25378064Sume	gid_t		va_gid;		/* owner group id */
25478064Sume	dev_t		va_fsid;	/* filesystem id */
25578064Sume	long		va_fileid;	/* file id */
25678064Sume	u_quad_t	va_size;	/* file size in bytes */
25778064Sume	long		va_blocksize;	/* blocksize preferred for i/o */
25878064Sume	struct timespec	va_atime;	/* time of last access */
25978064Sume	struct timespec	va_mtime;	/* time of last modification */
26078064Sume	struct timespec	va_ctime;	/* time file changed */
26152904Sshin	struct timespec	va_birthtime;	/* time file created */
26252904Sshin	u_long		va_gen;		/* generation number of file */
26352904Sshin	u_long		va_flags;	/* flags defined for file */
26462587Sitojun	dev_t		va_rdev;	/* device the special file represents */
26578064Sume	u_quad_t	va_bytes;	/* bytes of disk space held by file */
26678064Sume	u_quad_t	va_filerev;	/* file modification number */
26778064Sume	u_int		va_vaflags;	/* operations flags, see below */
26878064Sume	long		va_spare;	/* remain quad aligned */
26952904Sshin};
27052904Sshin
27152904Sshin/*
27252904Sshin * Flags for va_vaflags.
27362587Sitojun */
27478064Sume#define	VA_UTIMES_NULL	0x01		/* utimes argument was NULL */
27578064Sume#define	VA_EXCLUSIVE	0x02		/* exclusive create request */
27678064Sume
27778064Sume/*
27852904Sshin * Flags for ioflag. (high 16 bits used to ask for read-ahead and
27952904Sshin * help with write clustering)
28052904Sshin */
28152904Sshin#define	IO_UNIT		0x0001		/* do I/O as atomic unit */
28262587Sitojun#define	IO_APPEND	0x0002		/* append write to end */
28378064Sume#define	IO_SYNC		0x0004		/* do I/O synchronously */
28478064Sume#define	IO_NODELOCKED	0x0008		/* underlying node already locked */
28578064Sume#define	IO_NDELAY	0x0010		/* FNDELAY flag set in file table */
28678064Sume#define	IO_VMIO		0x0020		/* data already in VMIO space */
28778064Sume#define	IO_INVAL	0x0040		/* invalidate after I/O */
28852904Sshin#define	IO_ASYNC	0x0080		/* bawrite rather then bdwrite */
28952904Sshin#define	IO_DIRECT	0x0100		/* attempt to bypass buffer cache */
29052904Sshin#define	IO_EXT		0x0400		/* operate on external attributes */
29152904Sshin#define	IO_NORMAL	0x0800		/* operate on regular data */
29262587Sitojun#define	IO_NOMACCHECK	0x1000		/* MAC checks unnecessary */
29378064Sume
29478064Sume#define IO_SEQMAX	0x7F		/* seq heuristic max value */
29578064Sume#define IO_SEQSHIFT	16		/* seq heuristic in upper 16 bits */
29652904Sshin
29752904Sshin/*
29852904Sshin *  Modes.  Some values same as Ixxx entries from inode.h for now.
29952904Sshin */
30052904Sshin#define	VEXEC	000100		/* execute/search permission */
30195023Ssuz#define	VWRITE	000200		/* write permission */
30262587Sitojun#define	VREAD	000400		/* read permission */
30362587Sitojun#define	VSVTX	001000		/* save swapped text even after use */
30462587Sitojun#define	VSGID	002000		/* set group id on execution */
30562587Sitojun#define	VSUID	004000		/* set user id on execution */
30662587Sitojun#define	VADMIN	010000		/* permission to administer */
30752904Sshin#define	VSTAT	020000		/* permission to retrieve attrs */
30862587Sitojun#define	VAPPEND	040000		/* permission to write/append */
30962587Sitojun#define	VALLPERM	(VEXEC | VWRITE | VREAD | VADMIN | VSTAT | VAPPEND)
31062587Sitojun
31162587Sitojun/*
31262587Sitojun * Token indicating no attribute value yet assigned.
31352904Sshin */
31452904Sshin#define	VNOVAL	(-1)
31552904Sshin
31652904Sshin/*
31752904Sshin * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon)
31852904Sshin */
31962587Sitojun#define VLKTIMEOUT	(hz / 20 + 1)
32052904Sshin
32162587Sitojun#ifdef _KERNEL
32252904Sshin
32352904Sshin#ifdef MALLOC_DECLARE
32452904SshinMALLOC_DECLARE(M_VNODE);
32552904Sshin#endif
32652904Sshin
32752904Sshin/*
32852904Sshin * Convert between vnode types and inode formats (since POSIX.1
32995023Ssuz * defines mode word of stat structure in terms of inode formats).
33062587Sitojun */
33152904Sshinextern enum vtype	iftovt_tab[];
33262587Sitojunextern int		vttoif_tab[];
33352904Sshin#define	IFTOVT(mode)	(iftovt_tab[((mode) & S_IFMT) >> 12])
33452904Sshin#define	VTTOIF(indx)	(vttoif_tab[(int)(indx)])
33552904Sshin#define	MAKEIMODE(indx, mode)	(int)(VTTOIF(indx) | (mode))
33652904Sshin
33752904Sshin/*
33895023Ssuz * Flags to various vnode functions.
33962587Sitojun */
34052904Sshin#define	SKIPSYSTEM	0x0001	/* vflush: skip vnodes marked VSYSTEM */
34152904Sshin#define	FORCECLOSE	0x0002	/* vflush: force file closure */
34262587Sitojun#define	WRITECLOSE	0x0004	/* vflush: only close writable files */
34352904Sshin#define	DOCLOSE		0x0008	/* vclean: close active files */
34452904Sshin#define	V_SAVE		0x0001	/* vinvalbuf: sync file first */
34562587Sitojun#define	V_ALT		0x0002	/* vinvalbuf: invalidate only alternate bufs */
34652904Sshin#define	V_NORMAL	0x0004	/* vinvalbuf: invalidate only regular bufs */
34752904Sshin#define	REVOKEALL	0x0001	/* vop_revoke: revoke all aliases */
34862587Sitojun#define	V_WAIT		0x0001	/* vn_start_write: sleep for suspend */
34952904Sshin#define	V_NOWAIT	0x0002	/* vn_start_write: don't sleep for suspend */
35052904Sshin#define	V_XSLEEP	0x0004	/* vn_start_write: just return after sleep */
35162587Sitojun
35252904Sshin#define	VREF(vp)	vref(vp)
35352904Sshin
35452904Sshin#ifdef DIAGNOSTIC
35562587Sitojun#define	VATTR_NULL(vap)	vattr_null(vap)
35652904Sshin#else
35752904Sshin#define	VATTR_NULL(vap)	(*(vap) = va_null)	/* initialize a vattr */
35862587Sitojun#endif /* DIAGNOSTIC */
35952904Sshin
36052904Sshin#define	NULLVP	((struct vnode *)NULL)
36162587Sitojun
36252904Sshin#define	VNODEOP_SET(f) \
36352904Sshin	C_SYSINIT(f##init, SI_SUB_VFS, SI_ORDER_SECOND, vfs_add_vnodeops, &f); \
36462587Sitojun	C_SYSUNINIT(f##uninit, SI_SUB_VFS, SI_ORDER_SECOND, vfs_rm_vnodeops, &f);
36552904Sshin
36652904Sshin/*
36762587Sitojun * Global vnode data.
36852904Sshin */
36952904Sshinextern	struct vnode *rootvnode;	/* root (i.e. "/") vnode */
37052904Sshinextern	int async_io_version;		/* 0 or POSIX version of AIO i'face */
37152904Sshinextern	int desiredvnodes;		/* number of vnodes desired */
37295023Ssuzextern	struct uma_zone *namei_zone;
37378064Sumeextern	int prtactive;			/* nonzero to call vprint() */
37452904Sshinextern	struct vattr va_null;		/* predefined null vattr structure */
37552904Sshin
37662587Sitojun/*
37752904Sshin * Macro/function to check for client cache inconsistency w.r.t. leasing.
37852904Sshin */
37952904Sshin#define	LEASE_READ	0x1		/* Check lease for readers */
38078064Sume#define	LEASE_WRITE	0x2		/* Check lease for modifiers */
38178064Sume
38278064Sumeextern void	(*lease_updatetime)(int deltat);
38378064Sume
38478064Sume/* Requires interlock. */
38578064Sume#define	VSHOULDFREE(vp)	\
38678064Sume	(!((vp)->v_iflag & (VI_FREE|VI_DOOMED|VI_DOINGINACT)) && \
38778064Sume	 !(vp)->v_holdcnt && !(vp)->v_usecount && \
38852904Sshin	 (!(vp)->v_object || \
38952904Sshin	  !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
39052904Sshin
39197181Smike/* Requires interlock. */
39252904Sshin#define	VMIGHTFREE(vp) \
39352904Sshin	(!((vp)->v_iflag & (VI_FREE|VI_DOOMED|VI_XLOCK|VI_DOINGINACT)) && \
39452904Sshin	 LIST_EMPTY(&(vp)->v_cache_src) && !(vp)->v_usecount)
39552904Sshin
39652904Sshin/* Requires interlock. */
39752904Sshin#define	VSHOULDBUSY(vp)	\
39852904Sshin	(((vp)->v_iflag & VI_FREE) && \
39952904Sshin	 ((vp)->v_holdcnt || (vp)->v_usecount))
40052904Sshin
40152904Sshin#define	VI_LOCK(vp)	mtx_lock(&(vp)->v_interlock)
40297181Smike#define	VI_TRYLOCK(vp)	mtx_trylock(&(vp)->v_interlock)
40397181Smike#define	VI_UNLOCK(vp)	mtx_unlock(&(vp)->v_interlock)
40497181Smike#define	VI_MTX(vp)	(&(vp)->v_interlock)
40597181Smike
40697181Smike#endif /* _KERNEL */
40797181Smike
40897181Smike/*
40997181Smike * Mods for extensibility.
41097181Smike */
41152904Sshin
41262587Sitojun/*
41362587Sitojun * Flags for vdesc_flags:
41462587Sitojun */
41562587Sitojun#define	VDESC_MAX_VPS		16
41662587Sitojun/* Low order 16 flag bits are reserved for willrele flags for vp arguments. */
41762587Sitojun#define	VDESC_VP0_WILLRELE	0x0001
41862587Sitojun#define	VDESC_VP1_WILLRELE	0x0002
41962587Sitojun#define	VDESC_VP2_WILLRELE	0x0004
42062587Sitojun#define	VDESC_VP3_WILLRELE	0x0008
42162587Sitojun#define	VDESC_NOMAP_VPP		0x0100
42278064Sume#define	VDESC_VPP_WILLRELE	0x0200
42378064Sume
42462587Sitojun/*
42562587Sitojun * VDESC_NO_OFFSET is used to identify the end of the offset list
42662587Sitojun * and in places where no such field exists.
42762587Sitojun */
42862587Sitojun#define VDESC_NO_OFFSET -1
42962587Sitojun
43078064Sume/*
43162587Sitojun * This structure describes the vnode operation taking place.
43278064Sume */
43378064Sumestruct vnodeop_desc {
43478064Sume	int	 vdesc_offset;		/* offset in vector,first for speed */
43552904Sshin	char	*vdesc_name;		/* a readable name for debugging */
43695023Ssuz	int	 vdesc_flags;		/* VDESC_* flags */
43762587Sitojun
43878064Sume	/*
43962587Sitojun	 * These ops are used by bypass routines to map and locate arguments.
44052904Sshin	 * Creds and procs are not needed in bypass routines, but sometimes
44195023Ssuz	 * they are useful to (for example) transport layers.
44262587Sitojun	 * Nameidata is useful because it has a cred in it.
44362587Sitojun	 */
44462587Sitojun	int	*vdesc_vp_offsets;	/* list ended by VDESC_NO_OFFSET */
44562587Sitojun	int	vdesc_vpp_offset;	/* return vpp location */
44662587Sitojun	int	vdesc_cred_offset;	/* cred location, if any */
44778064Sume	int	vdesc_thread_offset;	/* thread location, if any */
44878064Sume	int	vdesc_componentname_offset; /* if any */
44965124Sitojun	/*
45052904Sshin	 * Finally, we've got a list of private data (about each operation)
45162587Sitojun	 * for each transport layer.  (Support to manage this list is not
45262587Sitojun	 * yet part of BSD.)
45362587Sitojun	 */
45452904Sshin	caddr_t	*vdesc_transports;
45552904Sshin};
45652904Sshin
45752904Sshin#ifdef _KERNEL
45895023Ssuz/*
45995023Ssuz * A list of all the operation descs.
46052904Sshin */
46152904Sshinextern struct vnodeop_desc *vnodeop_descs[];
46252904Sshin
46352904Sshin#define	VOPARG_OFFSETOF(s_type, field)	__offsetof(s_type, field)
46452904Sshin#define	VOPARG_OFFSETTO(s_type, s_offset, struct_p) \
46562587Sitojun    ((s_type)(((char*)(struct_p)) + (s_offset)))
46662587Sitojun
46752904Sshin/*
46852904Sshin * This structure is used to configure the new vnodeops vector.
46952904Sshin */
47052904Sshinstruct vnodeopv_entry_desc {
47152904Sshin	struct vnodeop_desc *opve_op;   /* which operation this is */
47252904Sshin	vop_t *opve_impl;		/* code implementing this operation */
47362587Sitojun};
47462587Sitojunstruct vnodeopv_desc {
47552904Sshin			/* ptr to the ptr to the vector where op should go */
47652904Sshin	vop_t ***opv_desc_vector_p;
47752904Sshin	struct vnodeopv_entry_desc *opv_desc_ops;   /* null terminated list */
47852904Sshin};
47952904Sshin
48052904Sshin/*
48152904Sshin * A generic structure.
48252904Sshin * This can be used by bypass routines to identify generic arguments.
48352904Sshin */
48452904Sshinstruct vop_generic_args {
48552904Sshin	struct vnodeop_desc *a_desc;
48652904Sshin	/* other random data follows, presumably */
48752904Sshin};
48852904Sshin
48952904Sshin#ifdef DEBUG_VFS_LOCKS
49052904Sshin/*
49162587Sitojun * Support code to aid in debugging VFS locking problems.  Not totally
49252904Sshin * reliable since if the thread sleeps between changing the lock
49362587Sitojun * state and checking it with the assert, some other thread could
49452904Sshin * change the state.  They are good enough for debugging a single
49552904Sshin * filesystem using a single-threaded test.
49652904Sshin */
49752904Sshinvoid	assert_vi_locked(struct vnode *vp, const char *str);
49852904Sshinvoid	assert_vi_unlocked(struct vnode *vp, const char *str);
49952904Sshin#if 0
50052904Sshinvoid	assert_vop_elocked(struct vnode *vp, const char *str);
50152904Sshinvoid	assert_vop_elocked_other(struct vnode *vp, const char *str);
50252904Sshin#endif
50352904Sshinvoid	assert_vop_locked(struct vnode *vp, const char *str);
50452904Sshin#if 0
50552904Sshinvoi0	assert_vop_slocked(struct vnode *vp, const char *str);
50652904Sshin#endif
50752904Sshinvoid	assert_vop_unlocked(struct vnode *vp, const char *str);
50852904Sshin
50952904Sshin/* These are called from within the actual VOPS. */
51052904Sshinvoid	vop_lock_pre(void *a);
51152904Sshinvoid	vop_lock_post(void *a, int rc);
51252904Sshinvoid	vop_lookup_post(void *a, int rc);
51352904Sshinvoid	vop_lookup_pre(void *a);
51452904Sshinvoid	vop_rename_pre(void *a);
51552904Sshinvoid	vop_strategy_pre(void *a);
51652904Sshinvoid	vop_unlock_post(void *a, int rc);
51752904Sshinvoid	vop_unlock_pre(void *a);
51852904Sshin
51952904Sshin#define	ASSERT_VI_LOCKED(vp, str)	assert_vi_locked((vp), (str))
52052904Sshin#define	ASSERT_VI_UNLOCKED(vp, str)	assert_vi_unlocked((vp), (str))
52152904Sshin#if 0
52252904Sshin#define	ASSERT_VOP_ELOCKED(vp, str)	assert_vop_elocked((vp), (str))
52352904Sshin#define	ASSERT_VOP_ELOCKED_OTHER(vp, str) assert_vop_locked_other((vp), (str))
52452904Sshin#endif
52552904Sshin#define	ASSERT_VOP_LOCKED(vp, str)	assert_vop_locked((vp), (str))
52652904Sshin#if 0
52752904Sshin#define	ASSERT_VOP_SLOCKED(vp, str)	assert_vop_slocked((vp), (str))
52852904Sshin#endif
52952904Sshin#define	ASSERT_VOP_UNLOCKED(vp, str)	assert_vop_unlocked((vp), (str))
53052904Sshin
53152904Sshin#else /* !DEBUG_VFS_LOCKS */
53252904Sshin
53352904Sshin#define	ASSERT_VI_LOCKED(vp, str)
53452904Sshin#define	ASSERT_VI_UNLOCKED(vp, str)
53552904Sshin#if 0
53652904Sshin#define	ASSERT_VOP_ELOCKED(vp, str)
53752904Sshin#define	ASSERT_VOP_ELOCKED_OTHER(vp, str)
53852904Sshin#endif
53952904Sshin#define	ASSERT_VOP_LOCKED(vp, str)
54052904Sshin#if 0
54152904Sshin#define	ASSERT_VOP_SLOCKED(vp, str)
54252904Sshin#endif
54362587Sitojun#define	ASSERT_VOP_UNLOCKED(vp, str)
54462587Sitojun#endif /* DEBUG_VFS_LOCKS */
54562587Sitojun
54652904Sshin/*
54762587Sitojun * VOCALL calls an op given an ops vector.  We break it out because BSD's
54852904Sshin * vclean changes the ops vector and then wants to call ops with the old
54962587Sitojun * vector.
55062587Sitojun */
55162587Sitojun#define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP))
55262587Sitojun
55362587Sitojun/*
55462587Sitojun * This call works for vnodes in the kernel.
55562587Sitojun */
55662587Sitojun#define VCALL(VP,OFF,AP) VOCALL((VP)->v_op,(OFF),(AP))
55762587Sitojun#define VDESC(OP) (& __CONCAT(OP,_desc))
55862587Sitojun#define VOFFSET(OP) (VDESC(OP)->vdesc_offset)
55962587Sitojun
56062587Sitojun/*
56162587Sitojun * VMIO support inline
56262587Sitojun */
56362587Sitojun
56462587Sitojunextern int vmiodirenable;
56562587Sitojun
56662587Sitojunstatic __inline int
56795023Ssuzvn_canvmio(struct vnode *vp)
56862587Sitojun{
56962604Sitojun      if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR)))
57078064Sume		return(TRUE);
57162604Sitojun	return(FALSE);
57262604Sitojun}
57362604Sitojun
57478064Sume/*
57578064Sume * Finally, include the default set of vnode operations.
57678064Sume */
57778064Sume#include "vnode_if.h"
57878064Sume
57978064Sume/*
58078064Sume * Public vnode manipulation functions.
58152904Sshin */
58265124Sitojunstruct componentname;
58378064Sumestruct file;
58478064Sumestruct mount;
58553541Sshinstruct nameidata;
58653541Sshinstruct ostat;
58753541Sshinstruct thread;
58878064Sumestruct proc;
58978064Sumestruct stat;
59078064Sumestruct nstat;
59178064Sumestruct ucred;
59253541Sshinstruct uio;
59352904Sshinstruct vattr;
59478064Sumestruct vnode;
59552904Sshin
59662587Sitojunextern int	(*lease_check_hook)(struct vop_lease_args *);
59752904Sshinextern int	(*softdep_fsync_hook)(struct vnode *);
59852904Sshinextern int	(*softdep_process_worklist_hook)(struct mount *);
59952904Sshin
60052904Sshinstruct	vnode *addaliasu(struct vnode *vp, dev_t nvp_rdev);
60178064Sumeint	bdevvp(struct cdev *dev, struct vnode **vpp);
60278064Sume/* cache_* may belong in namei.h. */
60392700Sdarrenrvoid	cache_enter(struct vnode *dvp, struct vnode *vp,
60452904Sshin	    struct componentname *cnp);
60552904Sshinint	cache_lookup(struct vnode *dvp, struct vnode **vpp,
60652904Sshin	    struct componentname *cnp);
60752904Sshinvoid	cache_purge(struct vnode *vp);
60852904Sshinvoid	cache_purgevfs(struct mount *mp);
60952904Sshinint	cache_leaf_test(struct vnode *vp);
61052904Sshinint	change_dir(struct vnode *vp, struct thread *td);
61152904Sshinint	change_root(struct vnode *vp, struct thread *td);
61278064Sumevoid	cvtstat(struct stat *st, struct ostat *ost);
61378064Sumevoid	cvtnstat(struct stat *sb, struct nstat *nsb);
61478064Sumeint	getnewvnode(const char *tag, struct mount *mp, vop_t **vops,
61583934Sbrooks	    struct vnode **vpp);
61683934Sbrooksint	lease_check(struct vop_lease_args *ap);
61752904Sshinint	spec_vnoperate(struct vop_generic_args *);
61852904Sshinint	speedup_syncer(void);
61997181Smike#define textvp_fullpath(p, rb, rfb) \
62097181Smike	vn_fullpath(FIRST_THREAD_IN_PROC(p), (p)->p_textvp, rb, rfb)
62197181Smikeint	vn_fullpath(struct thread *td, struct vnode *vn,
62297181Smike	    char **retbuf, char **freebuf);
62397181Smikeint	vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
62452904Sshin	    mode_t acc_mode, struct ucred *cred, int *privused);
62552904Sshinint	vaccess_acl_posix1e(enum vtype type, uid_t file_uid,
62652904Sshin	    gid_t file_gid, struct acl *acl, mode_t acc_mode,
62778064Sume	    struct ucred *cred, int *privused);
62878064Sumevoid	vattr_null(struct vattr *vap);
62997181Smikeint	vcount(struct vnode *vp);
63078064Sumevoid	vdrop(struct vnode *);
63197181Smikevoid	vdropl(struct vnode *);
63297181Smikeint	vfinddev(struct cdev *dev, struct vnode **vpp);
63397181Smikevoid	vfs_add_vnodeops(const void *);
63452904Sshinvoid	vfs_rm_vnodeops(const void *);
63578064Sumeint	vflush(struct mount *mp, int rootrefs, int flags, struct thread *td);
63678064Sumeint	vget(struct vnode *vp, int lockflag, struct thread *td);
63778064Sumevoid	vgone(struct vnode *vp);
63878064Sumevoid	vgonel(struct vnode *vp, struct thread *td);
63978064Sumevoid	vhold(struct vnode *);
64078064Sumevoid	vholdl(struct vnode *);
64178064Sumeint	vinvalbuf(struct vnode *vp, int save, struct ucred *cred,
64278064Sume	    struct thread *td, int slpflag, int slptimeo);
64378064Sumeint	vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
64478064Sume	    off_t length, int blksize);
64578064Sumevoid	vprint(char *label, struct vnode *vp);
64678064Sumeint	vrecycle(struct vnode *vp, struct mtx *inter_lkp,
64778064Sume	    struct thread *td);
64897181Smikeint	vn_close(struct vnode *vp,
64997181Smike	    int flags, struct ucred *file_cred, struct thread *td);
65078064Sumevoid	vn_finished_write(struct mount *mp);
65178064Sumeint	vn_isdisk(struct vnode *vp, int *errp);
65278064Sumeint	vn_lock(struct vnode *vp, int flags, struct thread *td);
65397181Smike#ifdef	DEBUG_LOCKS
65478064Sumeint	debug_vn_lock(struct vnode *vp, int flags, struct thread *p,
65597181Smike	    const char *filename, int line);
65678064Sume#define vn_lock(vp,flags,p) debug_vn_lock(vp,flags,p,__FILE__,__LINE__)
65778064Sume#endif
65878064Sumeint	vn_open(struct nameidata *ndp, int *flagp, int cmode, int fdidx);
65978064Sumeint	vn_open_cred(struct nameidata *ndp, int *flagp, int cmode,
66078064Sume	    struct ucred *cred, int fdidx);
66178064Sumevoid	vn_pollevent(struct vnode *vp, int events);
66278064Sumevoid	vn_pollgone(struct vnode *vp);
66378064Sumeint	vn_pollrecord(struct vnode *vp, struct thread *p, int events);
66452904Sshinint	vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base,
66552904Sshin	    int len, off_t offset, enum uio_seg segflg, int ioflg,
66697181Smike	    struct ucred *active_cred, struct ucred *file_cred, int *aresid,
66797181Smike	    struct thread *td);
66852904Sshinint	vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base,
669	    size_t len, off_t offset, enum uio_seg segflg, int ioflg,
670	    struct ucred *active_cred, struct ucred *file_cred, size_t *aresid,
671	    struct thread *td);
672int	vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
673	    struct ucred *file_cred, struct thread *td);
674int	vn_start_write(struct vnode *vp, struct mount **mpp, int flags);
675struct cdev *vn_todev(struct vnode *vp);
676int	vn_write_suspend_wait(struct vnode *vp, struct mount *mp,
677	    int flags);
678int	vn_writechk(struct vnode *vp);
679int	vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
680	    const char *attrname, int *buflen, char *buf, struct thread *td);
681int	vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
682	    const char *attrname, int buflen, char *buf, struct thread *td);
683int	vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
684	    const char *attrname, struct thread *td);
685int	vfs_cache_lookup(struct vop_lookup_args *ap);
686int	vfs_object_create(struct vnode *vp, struct thread *td,
687	    struct ucred *cred);
688void	vfs_timestamp(struct timespec *);
689void	vfs_write_resume(struct mount *mp);
690int	vfs_write_suspend(struct mount *mp);
691int	vop_stdbmap(struct vop_bmap_args *);
692int	vop_stdfsync(struct vop_fsync_args *);
693int	vop_stdgetwritemount(struct vop_getwritemount_args *);
694int	vop_stdgetpages(struct vop_getpages_args *);
695int	vop_stdinactive(struct vop_inactive_args *);
696int	vop_stdislocked(struct vop_islocked_args *);
697int	vop_stdlock(struct vop_lock_args *);
698int	vop_stdputpages(struct vop_putpages_args *);
699int	vop_stdunlock(struct vop_unlock_args *);
700int	vop_nopoll(struct vop_poll_args *);
701int	vop_stdpathconf(struct vop_pathconf_args *);
702int	vop_stdpoll(struct vop_poll_args *);
703int	vop_revoke(struct vop_revoke_args *);
704int	vop_eopnotsupp(struct vop_generic_args *ap);
705int	vop_ebadf(struct vop_generic_args *ap);
706int	vop_einval(struct vop_generic_args *ap);
707int	vop_enotty(struct vop_generic_args *ap);
708int	vop_defaultop(struct vop_generic_args *ap);
709int	vop_null(struct vop_generic_args *ap);
710int	vop_panic(struct vop_generic_args *ap);
711int	vop_stdcreatevobject(struct vop_createvobject_args *ap);
712int	vop_stddestroyvobject(struct vop_destroyvobject_args *ap);
713int	vop_stdgetvobject(struct vop_getvobject_args *ap);
714
715void	vfree(struct vnode *);
716void	vput(struct vnode *vp);
717void	vrele(struct vnode *vp);
718void	vref(struct vnode *vp);
719int	vrefcnt(struct vnode *vp);
720void	vbusy(struct vnode *vp);
721void 	v_addpollinfo(struct vnode *vp);
722
723extern	vop_t **default_vnodeop_p;
724extern	vop_t **spec_vnodeop_p;
725extern	vop_t **dead_vnodeop_p;
726
727#endif /* _KERNEL */
728
729#endif /* !_SYS_VNODE_H_ */
730