vnode.h revision 21002
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 1.36 1996/10/17 17:12:04 jkh Exp $ 35 */ 36 37#ifndef _SYS_VNODE_H_ 38#define _SYS_VNODE_H_ 39 40#include <sys/queue.h> 41 42/* 43 * The vnode is the focus of all file activity in UNIX. There is a 44 * unique vnode allocated for each active file, each current directory, 45 * each mounted-on file, text file, and the root. 46 */ 47 48/* 49 * Vnode types. VNON means no type. 50 */ 51enum vtype { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD }; 52 53/* 54 * Vnode tag types. 55 * These are for the benefit of external programs only (e.g., pstat) 56 * and should NEVER be inspected by the kernel. 57 */ 58enum vtagtype { 59 VT_NON, VT_UFS, VT_NFS, VT_MFS, VT_PC, VT_LFS, VT_LOFS, VT_FDESC, 60 VT_PORTAL, VT_NULL, VT_UMAP, VT_KERNFS, VT_PROCFS, VT_AFS, VT_ISOFS, 61 VT_UNION, VT_MSDOSFS, VT_DEVFS, VT_TFS 62}; 63 64/* 65 * Each underlying filesystem allocates its own private area and hangs 66 * it from v_data. If non-null, this area is freed in getnewvnode(). 67 */ 68LIST_HEAD(buflists, buf); 69 70typedef int vop_t __P((void *)); 71struct vm_object; 72 73struct vnode { 74 u_long v_flag; /* vnode flags (see below) */ 75 int v_usecount; /* reference count of users */ 76 int v_writecount; /* reference count of writers */ 77 int v_holdcnt; /* page & buffer references */ 78 daddr_t v_lastr; /* last read (read-ahead) */ 79 u_long v_id; /* capability identifier */ 80 struct mount *v_mount; /* ptr to vfs we are in */ 81 vop_t **v_op; /* vnode operations vector */ 82 TAILQ_ENTRY(vnode) v_freelist; /* vnode freelist */ 83 LIST_ENTRY(vnode) v_mntvnodes; /* vnodes for mount point */ 84 struct buflists v_cleanblkhd; /* clean blocklist head */ 85 struct buflists v_dirtyblkhd; /* dirty blocklist head */ 86 long v_numoutput; /* num of writes in progress */ 87 enum vtype v_type; /* vnode type */ 88 union { 89 struct mount *vu_mountedhere;/* ptr to mounted vfs (VDIR) */ 90 struct socket *vu_socket; /* unix ipc (VSOCK) */ 91 struct specinfo *vu_specinfo; /* device (VCHR, VBLK) */ 92 struct fifoinfo *vu_fifoinfo; /* fifo (VFIFO) */ 93 } v_un; 94 struct nqlease *v_lease; /* Soft reference to lease */ 95 daddr_t v_lastw; /* last write (write cluster) */ 96 daddr_t v_cstart; /* start block of cluster */ 97 daddr_t v_lasta; /* last allocation */ 98 int v_clen; /* length of current cluster */ 99 int v_usage; /* Vnode usage counter */ 100 struct vm_object *v_object; /* Place to store VM object */ 101 enum vtagtype v_tag; /* type of underlying data */ 102 void *v_data; /* private data for fs */ 103}; 104#define v_mountedhere v_un.vu_mountedhere 105#define v_socket v_un.vu_socket 106#define v_specinfo v_un.vu_specinfo 107#define v_fifoinfo v_un.vu_fifoinfo 108 109/* 110 * Vnode flags. 111 */ 112#define VROOT 0x0001 /* root of its file system */ 113#define VTEXT 0x0002 /* vnode is a pure text prototype */ 114#define VSYSTEM 0x0004 /* vnode being used by kernel */ 115#define VOLOCK 0x0008 /* vnode is locked waiting for an object */ 116#define VOWANT 0x0010 /* a process is waiting for VOLOCK */ 117#define VXLOCK 0x0100 /* vnode is locked to change underlying type */ 118#define VXWANT 0x0200 /* process is waiting for vnode */ 119#define VBWAIT 0x0400 /* waiting for output to complete */ 120#define VALIASED 0x0800 /* vnode has an alias */ 121#define VDIROP 0x1000 /* LFS: vnode is involved in a directory op */ 122#define VVMIO 0x2000 /* VMIO flag */ 123#define VNINACT 0x4000 /* LFS: skip ufs_inactive() in lfs_vunref */ 124#define VAGE 0x8000 /* Insert vnode at head of free list */ 125 126/* 127 * Vnode attributes. A field value of VNOVAL represents a field whose value 128 * is unavailable (getattr) or which is not to be changed (setattr). 129 */ 130struct vattr { 131 enum vtype va_type; /* vnode type (for create) */ 132 u_short va_mode; /* files access mode and type */ 133 short va_nlink; /* number of references to file */ 134 uid_t va_uid; /* owner user id */ 135 gid_t va_gid; /* owner group id */ 136 long va_fsid; /* file system id (dev for now) */ 137 long va_fileid; /* file id */ 138 u_quad_t va_size; /* file size in bytes */ 139 long va_blocksize; /* blocksize preferred for i/o */ 140 struct timespec va_atime; /* time of last access */ 141 struct timespec va_mtime; /* time of last modification */ 142 struct timespec va_ctime; /* time file changed */ 143 u_long va_gen; /* generation number of file */ 144 u_long va_flags; /* flags defined for file */ 145 dev_t va_rdev; /* device the special file represents */ 146 u_quad_t va_bytes; /* bytes of disk space held by file */ 147 u_quad_t va_filerev; /* file modification number */ 148 u_int va_vaflags; /* operations flags, see below */ 149 long va_spare; /* remain quad aligned */ 150}; 151 152/* 153 * Flags for va_cflags. 154 */ 155#define VA_UTIMES_NULL 0x01 /* utimes argument was NULL */ 156 157/* 158 * Flags for ioflag. 159 */ 160#define IO_UNIT 0x01 /* do I/O as atomic unit */ 161#define IO_APPEND 0x02 /* append write to end */ 162#define IO_SYNC 0x04 /* do I/O synchronously */ 163#define IO_NODELOCKED 0x08 /* underlying node already locked */ 164#define IO_NDELAY 0x10 /* FNDELAY flag set in file table */ 165#define IO_VMIO 0x20 /* data already in VMIO space */ 166 167/* 168 * Modes. Some values same as Ixxx entries from inode.h for now. 169 */ 170#define VSUID 04000 /* set user id on execution */ 171#define VSGID 02000 /* set group id on execution */ 172#define VSVTX 01000 /* save swapped text even after use */ 173#define VREAD 00400 /* read, write, execute permissions */ 174#define VWRITE 00200 175#define VEXEC 00100 176 177/* 178 * Token indicating no attribute value yet assigned. 179 */ 180#define VNOVAL (-1) 181 182#ifdef KERNEL 183/* 184 * Convert between vnode types and inode formats (since POSIX.1 185 * defines mode word of stat structure in terms of inode formats). 186 */ 187extern enum vtype iftovt_tab[]; 188extern int vttoif_tab[]; 189#define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12]) 190#define VTTOIF(indx) (vttoif_tab[(int)(indx)]) 191#define MAKEIMODE(indx, mode) (int)(VTTOIF(indx) | (mode)) 192 193/* 194 * Flags to various vnode functions. 195 */ 196#define SKIPSYSTEM 0x0001 /* vflush: skip vnodes marked VSYSTEM */ 197#define FORCECLOSE 0x0002 /* vflush: force file closure */ 198#define WRITECLOSE 0x0004 /* vflush: only close writable files */ 199#define DOCLOSE 0x0008 /* vclean: close active files */ 200#define V_SAVE 0x0001 /* vinvalbuf: sync file first */ 201#define V_SAVEMETA 0x0002 /* vinvalbuf: leave indirect blocks */ 202 203#ifdef DIAGNOSTIC 204#define HOLDRELE(vp) holdrele(vp) 205#define VATTR_NULL(vap) vattr_null(vap) 206#define VHOLD(vp) vhold(vp) 207#define VREF(vp) vref(vp) 208 209void holdrele __P((struct vnode *)); 210void vhold __P((struct vnode *)); 211#else 212#define HOLDRELE(vp) (vp)->v_holdcnt-- /* decrease buf or page ref */ 213#define VATTR_NULL(vap) (*(vap) = va_null) /* initialize a vattr */ 214#define VHOLD(vp) (vp)->v_holdcnt++ /* increase buf or page ref */ 215#define VREF(vp) vref(vp) /* increase reference */ 216#endif 217 218#define NULLVP ((struct vnode *)NULL) 219 220#ifdef VFS_LKM 221#define VNODEOP_SET(f) DATA_SET(MODVNOPS,f) 222#else 223#define VNODEOP_SET(f) DATA_SET(vfs_opv_descs_,f) 224#endif 225 226/* 227 * Global vnode data. 228 */ 229extern struct vnode *rootvnode; /* root (i.e. "/") vnode */ 230extern int desiredvnodes; /* number of vnodes desired */ 231extern int extravnodes; /* extra vnodes to allocate at boot */ 232extern int prtactive; /* nonzero to call vprint() */ 233extern struct vattr va_null; /* predefined null vattr structure */ 234 235/* 236 * Macro/function to check for client cache inconsistency w.r.t. leasing. 237 */ 238#define LEASE_READ 0x1 /* Check lease for readers */ 239#define LEASE_WRITE 0x2 /* Check lease for modifiers */ 240 241extern void (*lease_check) __P((struct vnode *vp, struct proc *p, 242 struct ucred *ucred, int flag)); 243extern void (*lease_updatetime) __P((int deltat)); 244 245#ifdef NFS 246#ifdef NQNFS 247#define LEASE_CHECK(vp, p, cred, flag) lease_check((vp), (p), (cred), (flag)) 248#define LEASE_UPDATETIME(dt) lease_updatetime(dt) 249#else 250#define LEASE_CHECK(vp, p, cred, flag) 251#define LEASE_UPDATETIME(dt) 252#endif /* NQNFS */ 253#else 254#define LEASE_CHECK(vp, p, cred, flag) \ 255 do { if(lease_check) lease_check((vp), (p), (cred), (flag)); } while(0) 256#define LEASE_UPDATETIME(dt) \ 257 do { if(lease_updatetime) lease_updatetime(dt); } while(0) 258#endif /* NFS */ 259#endif /* KERNEL */ 260 261 262/* 263 * Mods for extensibility. 264 */ 265 266/* 267 * Flags for vdesc_flags: 268 */ 269#define VDESC_MAX_VPS 16 270/* Low order 16 flag bits are reserved for willrele flags for vp arguments. */ 271#define VDESC_VP0_WILLRELE 0x0001 272#define VDESC_VP1_WILLRELE 0x0002 273#define VDESC_VP2_WILLRELE 0x0004 274#define VDESC_VP3_WILLRELE 0x0008 275#define VDESC_NOMAP_VPP 0x0100 276#define VDESC_VPP_WILLRELE 0x0200 277 278/* 279 * VDESC_NO_OFFSET is used to identify the end of the offset list 280 * and in places where no such field exists. 281 */ 282#define VDESC_NO_OFFSET -1 283 284/* 285 * This structure describes the vnode operation taking place. 286 */ 287struct vnodeop_desc { 288 int vdesc_offset; /* offset in vector--first for speed */ 289 char *vdesc_name; /* a readable name for debugging */ 290 int vdesc_flags; /* VDESC_* flags */ 291 292 /* 293 * These ops are used by bypass routines to map and locate arguments. 294 * Creds and procs are not needed in bypass routines, but sometimes 295 * they are useful to (for example) transport layers. 296 * Nameidata is useful because it has a cred in it. 297 */ 298 int *vdesc_vp_offsets; /* list ended by VDESC_NO_OFFSET */ 299 int vdesc_vpp_offset; /* return vpp location */ 300 int vdesc_cred_offset; /* cred location, if any */ 301 int vdesc_proc_offset; /* proc location, if any */ 302 int vdesc_componentname_offset; /* if any */ 303 /* 304 * Finally, we've got a list of private data (about each operation) 305 * for each transport layer. (Support to manage this list is not 306 * yet part of BSD.) 307 */ 308 caddr_t *vdesc_transports; 309}; 310 311#ifdef KERNEL 312/* 313 * A list of all the operation descs. 314 */ 315extern struct vnodeop_desc *vnodeop_descs[]; 316 317 318/* 319 * This macro is very helpful in defining those offsets in the vdesc struct. 320 * 321 * This is stolen from X11R4. I ignored all the fancy stuff for 322 * Crays, so if you decide to port this to such a serious machine, 323 * you might want to consult Intrinsic.h's XtOffset{,Of,To}. 324 */ 325#define VOPARG_OFFSET(p_type,field) \ 326 ((int) (((char *) (&(((p_type)NULL)->field))) - ((char *) NULL))) 327#define VOPARG_OFFSETOF(s_type,field) \ 328 VOPARG_OFFSET(s_type*,field) 329#define VOPARG_OFFSETTO(S_TYPE,S_OFFSET,STRUCT_P) \ 330 ((S_TYPE)(((char*)(STRUCT_P))+(S_OFFSET))) 331 332 333/* 334 * This structure is used to configure the new vnodeops vector. 335 */ 336struct vnodeopv_entry_desc { 337 struct vnodeop_desc *opve_op; /* which operation this is */ 338 vop_t *opve_impl; /* code implementing this operation */ 339}; 340struct vnodeopv_desc { 341 /* ptr to the ptr to the vector where op should go */ 342 vop_t ***opv_desc_vector_p; 343 struct vnodeopv_entry_desc *opv_desc_ops; /* null terminated list */ 344}; 345 346/* 347 * A default routine which just returns an error. 348 */ 349int vn_default_error __P((void)); 350 351/* 352 * A generic structure. 353 * This can be used by bypass routines to identify generic arguments. 354 */ 355struct vop_generic_args { 356 struct vnodeop_desc *a_desc; 357 /* other random data follows, presumably */ 358}; 359 360/* 361 * VOCALL calls an op given an ops vector. We break it out because BSD's 362 * vclean changes the ops vector and then wants to call ops with the old 363 * vector. 364 */ 365#define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP)) 366 367/* 368 * This call works for vnodes in the kernel. 369 */ 370#define VCALL(VP,OFF,AP) VOCALL((VP)->v_op,(OFF),(AP)) 371#define VDESC(OP) (& __CONCAT(OP,_desc)) 372#define VOFFSET(OP) (VDESC(OP)->vdesc_offset) 373 374/* 375 * Finally, include the default set of vnode operations. 376 */ 377#include "vnode_if.h" 378 379/* 380 * Public vnode manipulation functions. 381 */ 382struct componentname; 383struct file; 384struct mount; 385struct nameidata; 386struct proc; 387struct stat; 388struct ucred; 389struct uio; 390struct vattr; 391struct vnode; 392struct vop_bwrite_args; 393 394int bdevvp __P((dev_t dev, struct vnode **vpp)); 395/* cache_* may belong in namei.h. */ 396void cache_enter __P((struct vnode *dvp, struct vnode *vp, 397 struct componentname *cnp)); 398int cache_lookup __P((struct vnode *dvp, struct vnode **vpp, 399 struct componentname *cnp)); 400void cache_purge __P((struct vnode *vp)); 401void cache_purgevfs __P((struct mount *mp)); 402struct vnode * 403 checkalias __P((struct vnode *vp, dev_t nvp_rdev, struct mount *mp)); 404int getnewvnode __P((enum vtagtype tag, 405 struct mount *mp, vop_t **vops, struct vnode **vpp)); 406void insmntque __P((struct vnode *vp, struct mount *mp)); 407void vattr_null __P((struct vattr *vap)); 408int vcount __P((struct vnode *vp)); 409int vfinddev __P((dev_t dev, enum vtype type, struct vnode **vpp)); 410void vfs_opv_init __P((struct vnodeopv_desc **them)); 411int vget __P((struct vnode *vp, int lockflag)); 412void vgone __P((struct vnode *vp)); 413void vgoneall __P((struct vnode *vp)); 414int vinvalbuf __P((struct vnode *vp, int save, struct ucred *cred, 415 struct proc *p, int slpflag, int slptimeo)); 416int vn_bwrite __P((struct vop_bwrite_args *ap)); 417int vn_close __P((struct vnode *vp, 418 int flags, struct ucred *cred, struct proc *p)); 419int vn_open __P((struct nameidata *ndp, int fmode, int cmode)); 420int vn_rdwr __P((enum uio_rw rw, struct vnode *vp, caddr_t base, 421 int len, off_t offset, enum uio_seg segflg, int ioflg, 422 struct ucred *cred, int *aresid, struct proc *p)); 423int vn_stat __P((struct vnode *vp, struct stat *sb, struct proc *p)); 424int vfs_object_create __P((struct vnode *vp, struct proc *p, 425 struct ucred *cred, int waslocked)); 426int vn_writechk __P((struct vnode *vp)); 427void vprint __P((char *label, struct vnode *vp)); 428void vput __P((struct vnode *vp)); 429void vref __P((struct vnode *vp)); 430void vrele __P((struct vnode *vp)); 431#endif /* KERNEL */ 432 433#endif /* !_SYS_VNODE_H_ */ 434