bsd/sys/vnode_internal.h

   1 /*
   2  * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
  29 /*
  30  * Copyright (c) 1989, 1993
  31  *      The Regents of the University of California.  All rights reserved.
  32  *
  33  * Redistribution and use in source and binary forms, with or without
  34  * modification, are permitted provided that the following conditions
  35  * are met:
  36  * 1. Redistributions of source code must retain the above copyright
  37  *    notice, this list of conditions and the following disclaimer.
  38  * 2. Redistributions in binary form must reproduce the above copyright
  39  *    notice, this list of conditions and the following disclaimer in the
  40  *    documentation and/or other materials provided with the distribution.
  41  * 3. All advertising materials mentioning features or use of this software
  42  *    must display the following acknowledgement:
  43  *      This product includes software developed by the University of
  44  *      California, Berkeley and its contributors.
  45  * 4. Neither the name of the University nor the names of its contributors
  46  *    may be used to endorse or promote products derived from this software
  47  *    without specific prior written permission.
  48  *
  49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  59  * SUCH DAMAGE.
  60  *
  61  *      @(#)vnode.h     8.17 (Berkeley) 5/20/95
  62  */
  63 /*
  64  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
  65  * support for mandatory and extensible security protections.  This notice
  66  * is included in support of clause 2.2 (b) of the Apple Public License,
  67  * Version 2.0.
  68  */
  69
  70 #ifndef _SYS_VNODE_INTERNAL_H_
  71 #define _SYS_VNODE_INTERNAL_H_
  72
  73 #define INTERIM_FSNODE_LOCK     1
  74
  75 #include <sys/appleapiopts.h>
  76 #include <sys/cdefs.h>
  77 #include <sys/queue.h>
  78 #include <sys/lock.h>
  79
  80 #include <sys/time.h>
  81 #include <sys/uio.h>
  82
  83 #include <sys/vm.h>
  84 #include <sys/systm.h>
  85 #include <kern/locks.h>
  86 #include <vm/vm_kern.h>
  87 #include <sys/vnode.h>
  88 #include <sys/namei.h>
  89 #include <sys/vfs_context.h>
  90 #include <sys/sysctl.h>
  91
  92
  93 struct lockf;
  94 struct label;
  95
  96 LIST_HEAD(buflists, buf);
  97
  98
  99 struct unsafe_fsnode {
 100         lck_mtx_t fsnodelock;
 101         int32_t   fsnode_count;
 102         void *    fsnodeowner;
 103 };
 104
 105 /*
 106  * Reading or writing any of these items requires holding the appropriate lock.
 107  * v_freelist is locked by the global vnode_list_lock
 108  * v_mntvnodes is locked by the mount_lock
 109  * v_nclinks and v_ncchildren are protected by the global name_cache_lock
 110  * v_cleanblkhd and v_dirtyblkhd and v_iterblkflags are locked via the global buf_mtxp
 111  * the rest of the structure is protected by the vnode_lock
 112  */
 113 struct vnode {
 114         lck_mtx_t v_lock;                       /* vnode mutex */
 115         TAILQ_ENTRY(vnode) v_freelist;          /* vnode freelist */
 116         TAILQ_ENTRY(vnode) v_mntvnodes;         /* vnodes for mount point */
 117         LIST_HEAD(, namecache) v_nclinks;       /* name cache entries that name this vnode */
 118         LIST_HEAD(, namecache) v_ncchildren;    /* name cache entries that regard us as there parent */
 119         vnode_t  v_defer_reclaimlist;           /* in case we have to defer the reclaim to avoid recursion */
 120         u_long   v_listflag;                    /* flags protected by the vnode_list_lock (see below) */
 121         u_long   v_flag;                        /* vnode flags (see below) */
 122         u_short  v_lflag;                       /* vnode local and named ref flags */
 123         u_char   v_iterblkflags;                /* buf iterator flags */
 124         u_char   v_references;                  /* number of times io_count has been granted */
 125         int32_t  v_kusecount;                   /* count of in-kernel refs */
 126         int32_t  v_usecount;                    /* reference count of users */
 127         int32_t  v_iocount;                     /* iocounters */
 128         void *   v_owner;                       /* act that owns the vnode */
 129         u_short  v_type;                        /* vnode type */
 130         u_short  v_tag;                         /* type of underlying data */
 131         int      v_id;                          /* identity of vnode contents */
 132         union {
 133                 struct mount    *vu_mountedhere;/* ptr to mounted vfs (VDIR) */
 134                 struct socket   *vu_socket;     /* unix ipc (VSOCK) */
 135                 struct specinfo *vu_specinfo;   /* device (VCHR, VBLK) */
 136                 struct fifoinfo *vu_fifoinfo;   /* fifo (VFIFO) */
 137                 struct ubc_info *vu_ubcinfo;    /* valid for (VREG) */
 138         } v_un;
 139         struct  buflists v_cleanblkhd;          /* clean blocklist head */
 140         struct  buflists v_dirtyblkhd;          /* dirty blocklist head */
 141         /*
 142          * the following 4 fields are protected
 143          * by the name_cache_lock held in
 144          * excluive mode
 145          */
 146         kauth_cred_t    v_cred;                 /* last authorized credential */
 147         kauth_action_t  v_authorized_actions;   /* current authorized actions for v_cred */
 148         int             v_cred_timestamp;       /* determine if entry is stale for MNTK_AUTH_OPAQUE */
 149         int             v_nc_generation;        /* changes when nodes are removed from the name cache */
 150         /*
 151          * back to the vnode lock for protection
 152          */
 153         long    v_numoutput;                    /* num of writes in progress */
 154         long    v_writecount;                   /* reference count of writers */
 155         const char *v_name;                     /* name component of the vnode */
 156         vnode_t v_parent;                       /* pointer to parent vnode */
 157 #ifdef INTERIM_FSNODE_LOCK
 158         struct lockf    *v_lockf;               /* advisory lock list head */
 159         struct unsafe_fsnode *v_unsafefs;       /* pointer to struct used to lock */
 160 #endif                                          /* vnodes on unsafe filesystems */
 161         int     (**v_op)(void *);               /* vnode operations vector */
 162         mount_t v_mount;                        /* ptr to vfs we are in */
 163         void *  v_data;                         /* private data for fs */
 164 #if CONFIG_MACF
 165         struct label *v_label;                  /* MAC security label */
 166 #endif
 167 };
 168
 169 #define v_mountedhere   v_un.vu_mountedhere
 170 #define v_socket        v_un.vu_socket
 171 #define v_specinfo      v_un.vu_specinfo
 172 #define v_fifoinfo      v_un.vu_fifoinfo
 173 #define v_ubcinfo       v_un.vu_ubcinfo
 174
 175
 176 /*
 177  * v_iterblkflags
 178  */
 179 #define VBI_ITER                0x1
 180 #define VBI_ITERWANT            0x2
 181 #define VBI_CLEAN               0x4
 182 #define VBI_DIRTY               0x8
 183 #define VBI_NEWBUF              0x10
 184
 185 /*
 186  * v_listflag
 187  */
 188 #define VLIST_RAGE    0x01            /* vnode is currently in the rapid age list */
 189 #define VLIST_DEAD    0x02            /* vnode is currently in the dead list */
 190
 191 /*
 192  * v_lflags
 193  */
 194 #define VL_SUSPENDED    0x0001          /* vnode is suspended */
 195 #define VL_DRAIN        0x0002          /* vnode is being drained */
 196 #define VL_TERMINATE    0x0004          /* vnode is marked for termination */
 197 #define VL_TERMWANT     0x0008          /* vnode is marked for termination */
 198 #define VL_DEAD         0x0010          /* vnode is dead and completed recycle */
 199 #define VL_MARKTERM     0x0020          /* vnode is dead and completed recycle */
 200 #define VL_MOUNTDEAD    0x0040          /* v_moutnedhere is dead   */
 201 #define VL_NEEDINACTIVE 0x0080          /* delay VNOP_INACTIVE until iocount goes to 0 */
 202
 203 #define VL_LABEL        0x0100          /* vnode is marked for labeling */
 204 #define VL_LABELWAIT    0x0200          /* vnode is marked for labeling */
 205 #define VL_LABELED      0x0400          /* vnode is labeled */
 206 #define VL_LWARNED      0x0800
 207
 208 #define VNAMED_UBC      0x2000          /* ubc named reference */
 209 #define VNAMED_MOUNT    0x4000          /* mount point named reference */
 210 #define VNAMED_FSHASH   0x8000          /* FS hash named reference */
 211
 212 /*
 213  * v_flags
 214  */
 215 #define VROOT           0x000001        /* root of its file system */
 216 #define VTEXT           0x000002        /* vnode is a pure text prototype */
 217 #define VSYSTEM         0x000004        /* vnode being used by kernel */
 218 #define VISTTY          0x000008        /* vnode represents a tty */
 219 #define VRAGE           0x000010        /* vnode is in rapid age state */
 220 #define VBDEVVP         0x000020        /* vnode created by bdevvp */
 221 #define VDEVFLUSH       0x000040        /* device vnode after vflush */
 222 #define VMOUNT          0x000080        /* mount operation in progress */
 223 #define VBWAIT          0x000100        /* waiting for output to complete */
 224 #define VALIASED        0x000200        /* vnode has an alias */
 225 #define VNOCACHE_DATA   0x000400        /* don't keep data cached once it's been consumed */
 226 #define VSTANDARD       0x000800        /* vnode obtained from common pool */
 227 #define VAGE            0x001000        /* Insert vnode at head of free list */
 228 #define VRAOFF          0x002000        /* read ahead disabled */
 229 #define VNCACHEABLE     0x004000        /* vnode is allowed to be put back in name cache */
 230 #define VUINACTIVE      0x008000        /* UBC vnode is on inactive list */
 231 #define VSWAP           0x010000        /* vnode is being used as swapfile */
 232 #define VTHROTTLED      0x020000        /* writes or pageouts have been throttled */
 233                 /* wakeup tasks waiting when count falls below threshold */
 234 #define VNOFLUSH        0x040000        /* don't vflush() if SKIPSYSTEM */
 235 #define VLOCKLOCAL      0x080000        /* this vnode does adv locking in vfs */
 236 #define VISHARDLINK     0x100000        /* hard link needs special processing on lookup and in volfs */
 237 #define VISUNION        0x200000        /* union special processing */
 238 #if NAMEDSTREAMS
 239 #define VISNAMEDSTREAM  0x400000        /* vnode is a named stream (eg HFS resource fork) */
 240 #endif
 241 #define VOPENEVT        0x800000        /* if process is P_CHECKOPENEVT, then or in the O_EVTONLY flag on open */
 242
 243 /*
 244  * Global vnode data.
 245  */
 246 extern  struct vnode *rootvnode;        /* root (i.e. "/") vnode */
 247
 248
 249 /*
 250  * Mods for exensibility.
 251  */
 252
 253 /*
 254  * Flags for vdesc_flags:
 255  */
 256 #define VDESC_MAX_VPS           16
 257 /* Low order 16 flag bits are reserved for willrele flags for vp arguments. */
 258 #define VDESC_VP0_WILLRELE      0x0001
 259 #define VDESC_VP1_WILLRELE      0x0002
 260 #define VDESC_VP2_WILLRELE      0x0004
 261 #define VDESC_VP3_WILLRELE      0x0008
 262 #define VDESC_NOMAP_VPP         0x0100
 263 #define VDESC_VPP_WILLRELE      0x0200
 264
 265 /*
 266  * VDESC_NO_OFFSET is used to identify the end of the offset list
 267  * and in places where no such field exists.
 268  */
 269 #define VDESC_NO_OFFSET -1
 270
 271 /*
 272  * This structure describes the vnode operation taking place.
 273  */
 274 struct vnodeop_desc {
 275         int     vdesc_offset;           /* offset in vector--first for speed */
 276         const char *vdesc_name;         /* a readable name for debugging */
 277         int     vdesc_flags;            /* VDESC_* flags */
 278
 279         /*
 280          * These ops are used by bypass routines to map and locate arguments.
 281          * Creds and procs are not needed in bypass routines, but sometimes
 282          * they are useful to (for example) transport layers.
 283          * Nameidata is useful because it has a cred in it.
 284          */
 285         int     *vdesc_vp_offsets;      /* list ended by VDESC_NO_OFFSET */
 286         int     vdesc_vpp_offset;       /* return vpp location */
 287         int     vdesc_cred_offset;      /* cred location, if any */
 288         int     vdesc_proc_offset;      /* proc location, if any */
 289         int     vdesc_componentname_offset; /* if any */
 290         int     vdesc_context_offset;   /* context location, if any */
 291         /*
 292          * Finally, we've got a list of private data (about each operation)
 293          * for each transport layer.  (Support to manage this list is not
 294          * yet part of BSD.)
 295          */
 296         caddr_t *vdesc_transports;
 297 };
 298
 299 /*
 300  * A list of all the operation descs.
 301  */
 302 extern struct vnodeop_desc *vnodeop_descs[];
 303
 304
 305 /*
 306  * This macro is very helpful in defining those offsets in the vdesc struct.
 307  *
 308  * This is stolen from X11R4.  I ingored all the fancy stuff for
 309  * Crays, so if you decide to port this to such a serious machine,
 310  * you might want to consult Intrisics.h's XtOffset{,Of,To}.
 311  */
 312 #define VOPARG_OFFSET(p_type,field) \
 313         ((int) (((char *) (&(((p_type)NULL)->field))) - ((char *) NULL)))
 314 #define VOPARG_OFFSETOF(s_type,field) \
 315         VOPARG_OFFSET(s_type*,field)
 316 #define VOPARG_OFFSETTO(S_TYPE,S_OFFSET,STRUCT_P) \
 317         ((S_TYPE)(((char*)(STRUCT_P))+(S_OFFSET)))
 318
 319
 320
 321 /*
 322  * VOCALL calls an op given an ops vector.  We break it out because BSD's
 323  * vclean changes the ops vector and then wants to call ops with the old
 324  * vector.
 325  */
 326 #define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP))
 327
 328 /*
 329  * This call works for vnodes in the kernel.
 330  */
 331 #define VCALL(VP,OFF,AP) VOCALL((VP)->v_op,(OFF),(AP))
 332 #define VDESC(OP) (& __CONCAT(OP,_desc))
 333 #define VOFFSET(OP) (VDESC(OP)->vdesc_offset)
 334
 335 struct ostat;
 336
 337 #define BUILDPATH_NO_FS_ENTER 0x1 /* Use cache values, do not enter file system */
 338 int     build_path(vnode_t first_vp, char *buff, int buflen, int *outlen, int flags, vfs_context_t ctx);
 339
 340 int     bdevvp(dev_t dev, struct vnode **vpp);
 341 void    cvtstat(struct stat *st, struct ostat *ost);
 342 void    vprint(const char *label, struct vnode *vp);
 343
 344
 345 __private_extern__ int is_package_name(const char *name, int len);
 346 __private_extern__ int set_package_extensions_table(void *data, int nentries, int maxwidth);
 347 int     vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base,
 348                         int len, off_t offset, enum uio_seg segflg, int ioflg,
 349                         kauth_cred_t cred, int *aresid, struct proc *p);
 350 int     vn_rdwr_64(enum uio_rw rw, struct vnode *vp, uint64_t base,
 351                         int64_t len, off_t offset, enum uio_seg segflg,
 352                         int ioflg, kauth_cred_t cred, int *aresid,
 353                         struct proc *p);
 354 #if CONFIG_MACF
 355 int     vn_setlabel (struct vnode *vp, struct label *intlabel,
 356             vfs_context_t context);
 357 #endif
 358 void    fifo_printinfo(struct vnode *vp);
 359 int     vn_lock(struct vnode *vp, int flags, struct proc *p);
 360 int     vn_open(struct nameidata *ndp, int fmode, int cmode);
 361 int     vn_open_modflags(struct nameidata *ndp, int *fmode, int cmode);
 362 int     vn_open_auth(struct nameidata *ndp, int *fmode, struct vnode_attr *);
 363 int     vn_close(vnode_t, int flags, vfs_context_t ctx);
 364
 365 #define VN_CREATE_NOAUTH                (1<<0)
 366 #define VN_CREATE_NOINHERIT             (1<<1)
 367 #define VN_CREATE_UNION                 (1<<2)
 368 #define VN_CREATE_NOLABEL               (1<<3)
 369 errno_t vn_create(vnode_t, vnode_t *, struct componentname *, struct vnode_attr *, int flags, vfs_context_t);
 370
 371
 372 int     vn_getxattr(vnode_t, const char *, uio_t, size_t *, int, vfs_context_t);
 373 int     vn_setxattr(vnode_t, const char *, uio_t, int, vfs_context_t);
 374 int     vn_removexattr(vnode_t, const char *, int, vfs_context_t);
 375 int     vn_listxattr(vnode_t, uio_t, size_t *, int, vfs_context_t);
 376
 377 int     default_getxattr(vnode_t, const char *, uio_t, size_t *, int, vfs_context_t);
 378 int     default_setxattr(vnode_t, const char *, uio_t, int, vfs_context_t);
 379 int     default_removexattr(vnode_t, const char *, int, vfs_context_t);
 380
 381 #if NAMEDSTREAMS
 382 errno_t  vnode_getnamedstream(vnode_t, vnode_t *, const char *, enum nsoperation, int, vfs_context_t);
 383 errno_t  vnode_makenamedstream(vnode_t, vnode_t *, const char *, int, vfs_context_t);
 384 errno_t  vnode_removenamedstream(vnode_t, vnode_t, const char *, int, vfs_context_t);
 385 errno_t  vnode_flushnamedstream(vnode_t vp, vnode_t svp, vfs_context_t context);
 386 errno_t  vnode_relenamedstream(vnode_t vp, vnode_t svp, vfs_context_t context);
 387 #endif
 388
 389 int vn_path_package_check(vnode_t vp, char *path, int pathlen, int *component);
 390
 391 void    nchinit(void) __attribute__((section("__TEXT, initcode")));
 392 int     resize_namecache(u_int newsize);
 393 void    name_cache_lock_shared(void);
 394 void    name_cache_lock(void);
 395 void    name_cache_unlock(void);
 396 void    cache_enter_with_gen(vnode_t dvp, vnode_t vp, struct componentname *cnp, int gen);
 397
 398 const char      *vnode_getname(vnode_t vp);
 399 void    vnode_putname(const char *name);
 400
 401 vnode_t vnode_getparent(vnode_t vp);
 402
 403 int vn_pathconf(vnode_t, int, register_t *, vfs_context_t);
 404
 405 #define vnode_lock_convert(v)   lck_mtx_convert_spin(&(v)->v_lock)
 406
 407 void    vnode_lock_spin(vnode_t);
 408
 409
 410 void    vnode_list_lock(void);
 411 void    vnode_list_unlock(void);
 412 int     vnode_ref_ext(vnode_t, int);
 413 void    vnode_rele_ext(vnode_t, int, int);
 414 void    vnode_rele_internal(vnode_t, int, int, int);
 415 int     vnode_getwithref(vnode_t);
 416 int     vnode_get_locked(vnode_t);
 417 int     vnode_put_locked(vnode_t);
 418
 419 int     vnode_issock(vnode_t);
 420
 421 void    unlock_fsnode(vnode_t, int *);
 422 int     lock_fsnode(vnode_t, int *);
 423
 424 errno_t vnode_resume(vnode_t);
 425 errno_t vnode_suspend(vnode_t);
 426
 427
 428 errno_t vnode_size(vnode_t, off_t *, vfs_context_t);
 429 errno_t vnode_setsize(vnode_t, off_t, int ioflag, vfs_context_t);
 430 int     vnode_setattr_fallback(vnode_t vp, struct vnode_attr *vap, vfs_context_t ctx);
 431
 432 void vn_setunionwait(vnode_t);
 433 void vn_checkunionwait(vnode_t);
 434 void vn_clearunionwait(vnode_t, int);
 435
 436 void SPECHASH_LOCK(void);
 437 void SPECHASH_UNLOCK(void);
 438
 439 int check_cdevmounted(dev_t, enum vtype, int *);
 440
 441 void    vnode_authorize_init(void) __attribute__((section("__TEXT, initcode")));
 442
 443 void    vfsinit(void);
 444
 445 /*
 446  * XXX exported symbols; should be static
 447  */
 448 void    vfs_op_init(void) __attribute__((section("__TEXT, initcode")));
 449 void    vfs_opv_init(void) __attribute__((section("__TEXT, initcode")));
 450 int     vfs_sysctl(int *name, u_int namelen, user_addr_t oldp, size_t *oldlenp,
 451                 user_addr_t newp, size_t newlen, struct proc *p);
 452 int     sysctl_vnode(struct sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req);
 453
 454 #endif /* !_SYS_VNODE_INTERNAL_H_ */