vnode.h revision 24623 
1/*
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by the University of
16 *	California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 *	@(#)vnode.h	8.7 (Berkeley) 2/4/94
34 * $Id: vnode.h,v 1.42 1997/02/22 09:46:29 peter Exp $
35 */
36
37#ifndef _SYS_VNODE_H_
38#define	_SYS_VNODE_H_
39
40#include <sys/lock.h>
41#include <sys/queue.h>
42
43/*
44 * The vnode is the focus of all file activity in UNIX.  There is a
45 * unique vnode allocated for each active file, each current directory,
46 * each mounted-on file, text file, and the root.
47 */
48
49/*
50 * Vnode types.  VNON means no type.
51 */
52enum vtype	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD };
53
54/*
55 * Vnode tag types.
56 * These are for the benefit of external programs only (e.g., pstat)
57 * and should NEVER be inspected by the kernel.
58 */
59enum vtagtype	{
60	VT_NON, VT_UFS, VT_NFS, VT_MFS, VT_PC, VT_LFS, VT_LOFS, VT_FDESC,
61	VT_PORTAL, VT_NULL, VT_UMAP, VT_KERNFS, VT_PROCFS, VT_AFS, VT_ISOFS,
62	VT_UNION, VT_MSDOSFS, VT_DEVFS, VT_TFS
63};
64
65/*
66 * Each underlying filesystem allocates its own private area and hangs
67 * it from v_data.  If non-null, this area is freed in getnewvnode().
68 */
69LIST_HEAD(buflists, buf);
70
71typedef	int 	vop_t __P((void *));
72struct vm_object;
73
74/*
75 * Reading or writing any of these items requires holding the appropriate lock.
76 * v_freelist is locked by the global vnode_free_list simple lock.
77 * v_mntvnodes is locked by the global mntvnodes simple lock.
78 * v_flag, v_usecount, v_holdcount and v_writecount are
79 *    locked by the v_interlock simple lock.
80 */
81struct vnode {
82	u_long	v_flag;				/* vnode flags (see below) */
83	int	v_usecount;			/* reference count of users */
84	int	v_writecount;			/* reference count of writers */
85	int	v_holdcnt;			/* page & buffer references */
86	daddr_t	v_lastr;			/* last read (read-ahead) */
87	u_long	v_id;				/* capability identifier */
88	struct	mount *v_mount;			/* ptr to vfs we are in */
89	vop_t	**v_op;				/* vnode operations vector */
90	TAILQ_ENTRY(vnode) v_freelist;		/* vnode freelist */
91	LIST_ENTRY(vnode) v_mntvnodes;		/* vnodes for mount point */
92	struct	buflists v_cleanblkhd;		/* clean blocklist head */
93	struct	buflists v_dirtyblkhd;		/* dirty blocklist head */
94	long	v_numoutput;			/* num of writes in progress */
95	enum	vtype v_type;			/* vnode type */
96	union {
97		struct mount	*vu_mountedhere;/* ptr to mounted vfs (VDIR) */
98		struct socket	*vu_socket;	/* unix ipc (VSOCK) */
99		struct specinfo	*vu_specinfo;	/* device (VCHR, VBLK) */
100		struct fifoinfo	*vu_fifoinfo;	/* fifo (VFIFO) */
101	} v_un;
102	struct	nqlease *v_lease;		/* Soft reference to lease */
103	daddr_t	v_lastw;			/* last write (write cluster) */
104	daddr_t	v_cstart;			/* start block of cluster */
105	daddr_t	v_lasta;			/* last allocation */
106	int	v_clen;				/* length of current cluster */
107	int	v_usage;			/* Vnode usage counter */
108	struct vm_object *v_object;		/* Place to store VM object */
109	struct	simplelock v_interlock;		/* lock on usecount and flag */
110	struct	lock *v_vnlock;			/* used for non-locking fs's */
111	enum	vtagtype v_tag;			/* type of underlying data */
112	void 	*v_data;			/* private data for fs */
113};
114#define	v_mountedhere	v_un.vu_mountedhere
115#define	v_socket	v_un.vu_socket
116#define	v_specinfo	v_un.vu_specinfo
117#define	v_fifoinfo	v_un.vu_fifoinfo
118
119/*
120 * Vnode flags.
121 */
122#define	VROOT		0x00001	/* root of its file system */
123#define	VTEXT		0x00002	/* vnode is a pure text prototype */
124#define	VSYSTEM		0x00004	/* vnode being used by kernel */
125#define	VISTTY		0x00008	/* vnode represents a tty */
126#define	VXLOCK		0x00100	/* vnode is locked to change underlying type */
127#define	VXWANT		0x00200	/* process is waiting for vnode */
128#define	VBWAIT		0x00400	/* waiting for output to complete */
129#define	VALIASED	0x00800	/* vnode has an alias */
130#define	VDIROP		0x01000	/* LFS: vnode is involved in a directory op */
131#define	VVMIO		0x02000	/* VMIO flag */
132#define	VNINACT		0x04000	/* LFS: skip ufs_inactive() in lfs_vunref */
133#define	VAGE		0x08000	/* Insert vnode at head of free list */
134#define	VOLOCK		0x10000	/* vnode is locked waiting for an object */
135#define	VOWANT		0x20000	/* a process is waiting for VOLOCK */
136
137/*
138 * Vnode attributes.  A field value of VNOVAL represents a field whose value
139 * is unavailable (getattr) or which is not to be changed (setattr).
140 */
141struct vattr {
142	enum vtype	va_type;	/* vnode type (for create) */
143	u_short		va_mode;	/* files access mode and type */
144	short		va_nlink;	/* number of references to file */
145	uid_t		va_uid;		/* owner user id */
146	gid_t		va_gid;		/* owner group id */
147	long		va_fsid;	/* file system id (dev for now) */
148	long		va_fileid;	/* file id */
149	u_quad_t	va_size;	/* file size in bytes */
150	long		va_blocksize;	/* blocksize preferred for i/o */
151	struct timespec	va_atime;	/* time of last access */
152	struct timespec	va_mtime;	/* time of last modification */
153	struct timespec	va_ctime;	/* time file changed */
154	u_long		va_gen;		/* generation number of file */
155	u_long		va_flags;	/* flags defined for file */
156	dev_t		va_rdev;	/* device the special file represents */
157	u_quad_t	va_bytes;	/* bytes of disk space held by file */
158	u_quad_t	va_filerev;	/* file modification number */
159	u_int		va_vaflags;	/* operations flags, see below */
160	long		va_spare;	/* remain quad aligned */
161};
162
163/*
164 * Flags for va_vaflags.
165 */
166#define	VA_UTIMES_NULL	0x01		/* utimes argument was NULL */
167#define VA_EXCLUSIVE	0x02		/* exclusive create request */
168
169/*
170 * Flags for ioflag.
171 */
172#define	IO_UNIT		0x01		/* do I/O as atomic unit */
173#define	IO_APPEND	0x02		/* append write to end */
174#define	IO_SYNC		0x04		/* do I/O synchronously */
175#define	IO_NODELOCKED	0x08		/* underlying node already locked */
176#define	IO_NDELAY	0x10		/* FNDELAY flag set in file table */
177#define	IO_VMIO		0x20		/* data already in VMIO space */
178
179/*
180 *  Modes.  Some values same as Ixxx entries from inode.h for now.
181 */
182#define	VSUID	04000		/* set user id on execution */
183#define	VSGID	02000		/* set group id on execution */
184#define	VSVTX	01000		/* save swapped text even after use */
185#define	VREAD	00400		/* read, write, execute permissions */
186#define	VWRITE	00200
187#define	VEXEC	00100
188
189/*
190 * Token indicating no attribute value yet assigned.
191 */
192#define	VNOVAL	(-1)
193
194#ifdef KERNEL
195/*
196 * Convert between vnode types and inode formats (since POSIX.1
197 * defines mode word of stat structure in terms of inode formats).
198 */
199extern enum vtype	iftovt_tab[];
200extern int		vttoif_tab[];
201#define IFTOVT(mode)	(iftovt_tab[((mode) & S_IFMT) >> 12])
202#define VTTOIF(indx)	(vttoif_tab[(int)(indx)])
203#define MAKEIMODE(indx, mode)	(int)(VTTOIF(indx) | (mode))
204
205/*
206 * Flags to various vnode functions.
207 */
208#define	SKIPSYSTEM	0x0001		/* vflush: skip vnodes marked VSYSTEM */
209#define	FORCECLOSE	0x0002		/* vflush: force file closure */
210#define	WRITECLOSE	0x0004		/* vflush: only close writable files */
211#define	DOCLOSE		0x0008		/* vclean: close active files */
212#define	V_SAVE		0x0001		/* vinvalbuf: sync file first */
213#define	V_SAVEMETA	0x0002		/* vinvalbuf: leave indirect blocks */
214#define	REVOKEALL	0x0001		/* vop_revoke: revoke all aliases */
215
216#ifdef DIAGNOSTIC
217#define	HOLDRELE(vp)	holdrele(vp)
218#define	VATTR_NULL(vap)	vattr_null(vap)
219#define	VHOLD(vp)	vhold(vp)
220#define	VREF(vp)	vref(vp)
221
222void	holdrele __P((struct vnode *));
223void	vhold __P((struct vnode *));
224void 	vref __P((struct vnode *vp));
225#else
226#define	VATTR_NULL(vap)	(*(vap) = va_null)	/* initialize a vattr */
227#define	HOLDRELE(vp)	holdrele(vp)		/* decrease buf or page ref */
228static __inline void
229holdrele(struct vnode *vp)
230{
231	simple_lock(&vp->v_interlock);
232	vp->v_holdcnt--;
233	simple_unlock(&vp->v_interlock);
234}
235#define	VHOLD(vp)	vhold(vp)		/* increase buf or page ref */
236static __inline void
237vhold(struct vnode *vp)
238{
239	simple_lock(&vp->v_interlock);
240	vp->v_holdcnt++;
241	simple_unlock(&vp->v_interlock);
242}
243#define	VREF(vp)	vref(vp)		/* increase reference */
244void 	vref __P((struct vnode *vp));
245#endif /* DIAGNOSTIC */
246
247#define	NULLVP	((struct vnode *)NULL)
248
249#ifdef VFS_LKM
250#define	VNODEOP_SET(f) DATA_SET(MODVNOPS,f)
251#else
252#define	VNODEOP_SET(f) DATA_SET(vfs_opv_descs_,f)
253#endif
254
255/*
256 * Global vnode data.
257 */
258extern	struct vnode *rootvnode;	/* root (i.e. "/") vnode */
259extern	int desiredvnodes;		/* number of vnodes desired */
260extern	int prtactive;			/* nonzero to call vprint() */
261extern	struct vattr va_null;		/* predefined null vattr structure */
262
263/*
264 * Macro/function to check for client cache inconsistency w.r.t. leasing.
265 */
266#define	LEASE_READ	0x1		/* Check lease for readers */
267#define	LEASE_WRITE	0x2		/* Check lease for modifiers */
268
269
270extern void	(*lease_updatetime) __P((int deltat));
271
272#ifdef NFS
273#ifdef NQNFS
274#define	LEASE_UPDATETIME(dt)		lease_updatetime(dt)
275#else
276#define	LEASE_UPDATETIME(dt)
277#endif /* NQNFS */
278#else
279#define	LEASE_UPDATETIME(dt) \
280	do { if(lease_updatetime) lease_updatetime(dt); } while(0)
281#endif /* NFS */
282
283#endif /* KERNEL */
284
285
286/*
287 * Mods for extensibility.
288 */
289
290/*
291 * Flags for vdesc_flags:
292 */
293#define VDESC_MAX_VPS		16
294/* Low order 16 flag bits are reserved for willrele flags for vp arguments. */
295#define VDESC_VP0_WILLRELE	0x0001
296#define VDESC_VP1_WILLRELE	0x0002
297#define VDESC_VP2_WILLRELE	0x0004
298#define VDESC_VP3_WILLRELE	0x0008
299#define VDESC_NOMAP_VPP		0x0100
300#define VDESC_VPP_WILLRELE	0x0200
301
302/*
303 * VDESC_NO_OFFSET is used to identify the end of the offset list
304 * and in places where no such field exists.
305 */
306#define VDESC_NO_OFFSET -1
307
308/*
309 * This structure describes the vnode operation taking place.
310 */
311struct vnodeop_desc {
312	int	vdesc_offset;		/* offset in vector--first for speed */
313	char    *vdesc_name;		/* a readable name for debugging */
314	int	vdesc_flags;		/* VDESC_* flags */
315
316	/*
317	 * These ops are used by bypass routines to map and locate arguments.
318	 * Creds and procs are not needed in bypass routines, but sometimes
319	 * they are useful to (for example) transport layers.
320	 * Nameidata is useful because it has a cred in it.
321	 */
322	int	*vdesc_vp_offsets;	/* list ended by VDESC_NO_OFFSET */
323	int	vdesc_vpp_offset;	/* return vpp location */
324	int	vdesc_cred_offset;	/* cred location, if any */
325	int	vdesc_proc_offset;	/* proc location, if any */
326	int	vdesc_componentname_offset; /* if any */
327	/*
328	 * Finally, we've got a list of private data (about each operation)
329	 * for each transport layer.  (Support to manage this list is not
330	 * yet part of BSD.)
331	 */
332	caddr_t	*vdesc_transports;
333};
334
335#ifdef KERNEL
336/*
337 * A list of all the operation descs.
338 */
339extern struct vnodeop_desc *vnodeop_descs[];
340
341/*
342 * Interlock for scanning list of vnodes attached to a mountpoint
343 */
344extern struct simplelock mntvnode_slock;
345
346/*
347 * This macro is very helpful in defining those offsets in the vdesc struct.
348 *
349 * This is stolen from X11R4.  I ignored all the fancy stuff for
350 * Crays, so if you decide to port this to such a serious machine,
351 * you might want to consult Intrinsic.h's XtOffset{,Of,To}.
352 */
353#define VOPARG_OFFSET(p_type,field) \
354        ((int) (((char *) (&(((p_type)NULL)->field))) - ((char *) NULL)))
355#define VOPARG_OFFSETOF(s_type,field) \
356	VOPARG_OFFSET(s_type*,field)
357#define VOPARG_OFFSETTO(S_TYPE,S_OFFSET,STRUCT_P) \
358	((S_TYPE)(((char*)(STRUCT_P))+(S_OFFSET)))
359
360
361/*
362 * This structure is used to configure the new vnodeops vector.
363 */
364struct vnodeopv_entry_desc {
365	struct vnodeop_desc *opve_op;   /* which operation this is */
366	vop_t *opve_impl;		/* code implementing this operation */
367};
368struct vnodeopv_desc {
369			/* ptr to the ptr to the vector where op should go */
370	vop_t ***opv_desc_vector_p;
371	struct vnodeopv_entry_desc *opv_desc_ops;   /* null terminated list */
372};
373
374/*
375 * A default routine which just returns an error.
376 */
377int vn_default_error __P((void));
378
379/*
380 * A generic structure.
381 * This can be used by bypass routines to identify generic arguments.
382 */
383struct vop_generic_args {
384	struct vnodeop_desc *a_desc;
385	/* other random data follows, presumably */
386};
387
388#ifdef DEBUG_VFS_LOCKS
389/*
390 * Macros to aid in tracing VFS locking problems.  Not totally
391 * reliable since if the process sleeps between changing the lock
392 * state and checking it with the assert, some other process could
393 * change the state.  They are good enough for debugging a single
394 * filesystem using a single-threaded test.  I find that 'cvs co src'
395 * is a pretty good test.
396 */
397
398/*
399 * [dfr] Kludge until I get around to fixing all the vfs locking.
400 */
401#define IS_LOCKING_VFS(vp)	((vp)->v_tag == VT_UFS		\
402				 || (vp)->v_tag == VT_MFS	\
403				 || (vp)->v_tag == VT_NFS	\
404				 || (vp)->v_tag == VT_LFS	\
405				 || (vp)->v_tag == VT_ISOFS	\
406				 || (vp)->v_tag == VT_MSDOSFS	\
407				 || (vp)->v_tag == VT_DEVFS)
408
409#define ASSERT_VOP_LOCKED(vp, str)				\
410    if ((vp) && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp)) {	\
411	panic("%s: %x is not locked but should be", str, vp);	\
412    }
413
414#define ASSERT_VOP_UNLOCKED(vp, str)				\
415    if ((vp) && IS_LOCKING_VFS(vp) && VOP_ISLOCKED(vp)) {	\
416	panic("%s: %x is locked but shouldn't be", str, vp);	\
417    }
418
419#else
420
421#define ASSERT_VOP_LOCKED(vp, str)
422#define ASSERT_VOP_UNLOCKED(vp, str)
423
424#endif
425
426/*
427 * VOCALL calls an op given an ops vector.  We break it out because BSD's
428 * vclean changes the ops vector and then wants to call ops with the old
429 * vector.
430 */
431#define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP))
432
433/*
434 * This call works for vnodes in the kernel.
435 */
436#define VCALL(VP,OFF,AP) VOCALL((VP)->v_op,(OFF),(AP))
437#define VDESC(OP) (& __CONCAT(OP,_desc))
438#define VOFFSET(OP) (VDESC(OP)->vdesc_offset)
439
440/*
441 * Finally, include the default set of vnode operations.
442 */
443#include "vnode_if.h"
444
445/*
446 * Public vnode manipulation functions.
447 */
448struct componentname;
449struct file;
450struct mount;
451struct nameidata;
452struct ostat;
453struct proc;
454struct stat;
455struct ucred;
456struct uio;
457struct vattr;
458struct vnode;
459struct vop_bwrite_args;
460
461extern int	(*lease_check_hook) __P((struct vop_lease_args *));
462
463int 	bdevvp __P((dev_t dev, struct vnode **vpp));
464/* cache_* may belong in namei.h. */
465void	cache_enter __P((struct vnode *dvp, struct vnode *vp,
466	    struct componentname *cnp));
467int	cache_lookup __P((struct vnode *dvp, struct vnode **vpp,
468	    struct componentname *cnp));
469void	cache_purge __P((struct vnode *vp));
470void	cache_purgevfs __P((struct mount *mp));
471void	cvtstat __P((struct stat *st, struct ostat *ost));
472int 	getnewvnode __P((enum vtagtype tag,
473	    struct mount *mp, vop_t **vops, struct vnode **vpp));
474void	insmntque __P((struct vnode *vp, struct mount *mp));
475int	lease_check __P((struct vop_lease_args *ap));
476void 	vattr_null __P((struct vattr *vap));
477int 	vcount __P((struct vnode *vp));
478int	vfinddev __P((dev_t dev, enum vtype type, struct vnode **vpp));
479void	vfs_opv_init __P((struct vnodeopv_desc **them));
480int	vflush __P((struct mount *mp, struct vnode *skipvp, int flags));
481int 	vget __P((struct vnode *vp, int lockflag, struct proc *p));
482void 	vgone __P((struct vnode *vp));
483int	vinvalbuf __P((struct vnode *vp, int save, struct ucred *cred,
484	    struct proc *p, int slpflag, int slptimeo));
485void	vprint __P((char *label, struct vnode *vp));
486int	vrecycle __P((struct vnode *vp, struct simplelock *inter_lkp,
487	    struct proc *p));
488int	vn_bwrite __P((struct vop_bwrite_args *ap));
489int 	vn_close __P((struct vnode *vp,
490	    int flags, struct ucred *cred, struct proc *p));
491int	vn_lock __P((struct vnode *vp, int flags, struct proc *p));
492int 	vn_open __P((struct nameidata *ndp, int fmode, int cmode));
493int 	vn_rdwr __P((enum uio_rw rw, struct vnode *vp, caddr_t base,
494	    int len, off_t offset, enum uio_seg segflg, int ioflg,
495	    struct ucred *cred, int *aresid, struct proc *p));
496int	vn_stat __P((struct vnode *vp, struct stat *sb, struct proc *p));
497int	vfs_object_create __P((struct vnode *vp, struct proc *p,
498                struct ucred *cred, int waslocked));
499int 	vn_writechk __P((struct vnode *vp));
500int	vop_noislocked __P((struct vop_islocked_args *));
501int	vop_nolock __P((struct vop_lock_args *));
502int	vop_nounlock __P((struct vop_unlock_args *));
503int	vop_sharedlock __P((struct vop_lock_args *));
504int	vop_revoke __P((struct vop_revoke_args *));
505struct vnode *
506	checkalias __P((struct vnode *vp, dev_t nvp_rdev, struct mount *mp));
507void 	vput __P((struct vnode *vp));
508void 	vrele __P((struct vnode *vp));
509#endif /* KERNEL */
510
511#endif /* !_SYS_VNODE_H_ */
512