bsd/nfs/nfsnode.h

   1 /*
   2  * Copyright (c) 2000-2009 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  * This code is derived from software contributed to Berkeley by
  34  * Rick Macklem at The University of Guelph.
  35  *
  36  * Redistribution and use in source and binary forms, with or without
  37  * modification, are permitted provided that the following conditions
  38  * are met:
  39  * 1. Redistributions of source code must retain the above copyright
  40  *    notice, this list of conditions and the following disclaimer.
  41  * 2. Redistributions in binary form must reproduce the above copyright
  42  *    notice, this list of conditions and the following disclaimer in the
  43  *    documentation and/or other materials provided with the distribution.
  44  * 3. All advertising materials mentioning features or use of this software
  45  *    must display the following acknowledgement:
  46  *      This product includes software developed by the University of
  47  *      California, Berkeley and its contributors.
  48  * 4. Neither the name of the University nor the names of its contributors
  49  *    may be used to endorse or promote products derived from this software
  50  *    without specific prior written permission.
  51  *
  52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  62  * SUCH DAMAGE.
  63  *
  64  *      @(#)nfsnode.h   8.9 (Berkeley) 5/14/95
  65  * FreeBSD-Id: nfsnode.h,v 1.24 1997/10/28 14:06:25 bde Exp $
  66  */
  67
  68
  69 #ifndef _NFS_NFSNODE_H_
  70 #define _NFS_NFSNODE_H_
  71
  72 #include <sys/appleapiopts.h>
  73
  74 #ifdef __APPLE_API_PRIVATE
  75 #ifndef _NFS_NFS_H_
  76 #include <nfs/nfs.h>
  77 #endif
  78
  79 /*
  80  * Silly rename structure that hangs off the nfsnode until the name
  81  * can be removed by nfs_vnop_inactive()
  82  */
  83 struct nfs_sillyrename {
  84         kauth_cred_t    nsr_cred;
  85         struct nfsnode  *nsr_dnp;
  86         int             nsr_namlen;
  87         char            nsr_name[20];
  88 };
  89
  90 /*
  91  * The nfsbuf is the nfs equivalent to a struct buf.
  92  */
  93 struct nfsbuf {
  94         LIST_ENTRY(nfsbuf)      nb_hash;        /* hash chain */
  95         LIST_ENTRY(nfsbuf)      nb_vnbufs;      /* nfsnode's nfsbuf chain */
  96         TAILQ_ENTRY(nfsbuf)     nb_free;        /* free list position if not active. */
  97         volatile uint32_t       nb_flags;       /* NB_* flags. */
  98         volatile uint32_t       nb_lflags;      /* NBL_* flags. */
  99         volatile uint32_t       nb_refs;        /* outstanding references. */
 100         uint32_t                nb_bufsize;     /* buffer size */
 101         daddr64_t               nb_lblkno;      /* logical block number. */
 102         uint64_t                nb_verf;        /* V3 write verifier */
 103         int                     nb_commitlevel; /* lowest write commit level */
 104         time_t                  nb_timestamp;   /* buffer timestamp */
 105         int                     nb_error;       /* errno value. */
 106         u_int32_t               nb_valid;       /* valid pages in buf */
 107         u_int32_t               nb_dirty;       /* dirty pages in buf */
 108         int                     nb_validoff;    /* offset in buffer of valid region. */
 109         int                     nb_validend;    /* offset of end of valid region. */
 110         int                     nb_dirtyoff;    /* offset in buffer of dirty region. */
 111         int                     nb_dirtyend;    /* offset of end of dirty region. */
 112         int                     nb_offio;       /* offset in buffer of I/O region. */
 113         int                     nb_endio;       /* offset of end of I/O region. */
 114         int                     nb_rpcs;        /* Count of RPCs remaining for this buffer. */
 115         caddr_t                 nb_data;        /* mapped buffer */
 116         nfsnode_t               nb_np;          /* nfsnode buffer belongs to */
 117         kauth_cred_t            nb_rcred;       /* read credentials reference */
 118         kauth_cred_t            nb_wcred;       /* write credentials reference */
 119         void *                  nb_pagelist;    /* upl */
 120 };
 121
 122 #define NFS_MAXBSIZE    (32 * PAGE_SIZE)        /* valid/dirty page masks limit buffer size */
 123
 124 #define NFS_A_LOT_OF_NEEDCOMMITS        256                     /* max# uncommitted buffers for a node */
 125 #define NFS_A_LOT_OF_DELAYED_WRITES     MAX(nfsbufcnt/8,512)    /* max# "delwri" buffers in system */
 126
 127 /*
 128  * These flags are kept in b_lflags...
 129  * nfs_buf_mutex must be held before examining/updating
 130  */
 131 #define NBL_BUSY        0x00000001      /* I/O in progress. */
 132 #define NBL_WANTED      0x00000002      /* Process wants this buffer. */
 133
 134 /*
 135  * These flags are kept in nb_flags and they're (purposefully)
 136  * very similar to the B_* flags for struct buf.
 137  * nfs_buf_mutex is not needed to examine/update these.
 138  */
 139 #define NB_STALEWVERF   0x00000001      /* write verifier changed on us */
 140 #define NB_NEEDCOMMIT   0x00000002      /* buffer needs to be committed */
 141 #define NB_ASYNC        0x00000004      /* Start I/O, do not wait. */
 142 #define NB_CACHE        0x00000020      /* buffer data found in the cache */
 143 #define NB_STABLE       0x00000040      /* write FILESYNC not UNSTABLE */
 144 #define NB_DELWRI       0x00000080      /* delayed write: dirty range needs to be written */
 145 #define NB_DONE         0x00000200      /* I/O completed. */
 146 #define NB_EINTR        0x00000400      /* I/O was interrupted */
 147 #define NB_ERROR        0x00000800      /* I/O error occurred. */
 148 #define NB_INVAL        0x00002000      /* Does not contain valid info. */
 149 #define NB_NCRDAHEAD    0x00004000      /* "nocache readahead" data */
 150 #define NB_NOCACHE      0x00008000      /* Do not cache block after use. */
 151 #define NB_WRITE        0x00000000      /* Write buffer (pseudo flag). */
 152 #define NB_READ         0x00100000      /* Read buffer. */
 153 #define NB_MULTASYNCRPC 0x00200000      /* multiple async RPCs issued for buffer */
 154 #define NB_PAGELIST     0x00400000      /* Buffer describes pagelist I/O. */
 155 #define NB_WRITEINPROG  0x01000000      /* Write in progress. */
 156 #define NB_META         0x40000000      /* buffer contains meta-data. */
 157
 158 /* Flags for operation type in nfs_buf_get() */
 159 #define NBLK_READ       0x00000001      /* buffer for read */
 160 #define NBLK_WRITE      0x00000002      /* buffer for write */
 161 #define NBLK_META       0x00000004      /* buffer for metadata */
 162 #define NBLK_OPMASK     0x00000007      /* operation mask */
 163 /* modifiers for above flags... */
 164 #define NBLK_NOWAIT     0x40000000      /* don't wait on busy buffer */
 165 #define NBLK_ONLYVALID  0x80000000      /* only return cached buffer */
 166
 167 /* These flags are used for nfsbuf iterating */
 168 #define NBI_ITER                0x01    /* iteration in progress */
 169 #define NBI_ITERWANT            0x02    /* waiting to iterate */
 170 #define NBI_CLEAN               0x04    /* requesting clean buffers */
 171 #define NBI_DIRTY               0x08    /* requesting dirty buffers */
 172 #define NBI_NOWAIT              0x10    /* don't block on NBI_ITER */
 173
 174 /* Flags for nfs_buf_acquire */
 175 #define NBAC_NOWAIT             0x01    /* Don't wait if buffer is busy */
 176 #define NBAC_REMOVE             0x02    /* Remove from free list once buffer is acquired */
 177
 178 /* some convenience macros...  */
 179 #define NBOFF(BP)                       ((off_t)(BP)->nb_lblkno * (off_t)(BP)->nb_bufsize)
 180 #define NBPGVALID(BP,P)                 (((BP)->nb_valid >> (P)) & 0x1)
 181 #define NBPGDIRTY(BP,P)                 (((BP)->nb_dirty >> (P)) & 0x1)
 182 #define NBPGVALID_SET(BP,P)             ((BP)->nb_valid |= (1 << (P)))
 183 #define NBPGDIRTY_SET(BP,P)             ((BP)->nb_dirty |= (1 << (P)))
 184
 185 #define NBUFSTAMPVALID(BP)              ((BP)->nb_timestamp != ~0)
 186 #define NBUFSTAMPINVALIDATE(BP)         ((BP)->nb_timestamp = ~0)
 187
 188 #define NFS_BUF_MAP(BP) \
 189         do { \
 190                 if (!(BP)->nb_data && nfs_buf_map(BP)) \
 191                         panic("nfs_buf_map failed"); \
 192         } while (0)
 193
 194 LIST_HEAD(nfsbuflists, nfsbuf);
 195 TAILQ_HEAD(nfsbuffreehead, nfsbuf);
 196
 197 #define NFSNOLIST ((void*)0xdeadbeef)
 198
 199 __private_extern__ lck_mtx_t *nfs_buf_mutex;
 200 __private_extern__ int nfsbufcnt, nfsbufmin, nfsbufmax, nfsbufmetacnt, nfsbufmetamax;
 201 __private_extern__ int nfsbuffreecnt, nfsbuffreemetacnt, nfsbufdelwricnt, nfsneedbuffer;
 202 __private_extern__ int nfs_nbdwrite;
 203 __private_extern__ struct nfsbuffreehead nfsbuffree, nfsbufdelwri;
 204
 205 #ifdef NFSBUFDEBUG
 206 #define NFSBUFCNTCHK() \
 207         do { \
 208         if (    (nfsbufcnt < 0) || \
 209                 (nfsbufcnt > nfsbufmax) || \
 210                 (nfsbufmetacnt < 0) || \
 211                 (nfsbufmetacnt > nfsbufmetamax) || \
 212                 (nfsbufmetacnt > nfsbufcnt) || \
 213                 (nfsbuffreecnt < 0) || \
 214                 (nfsbuffreecnt > nfsbufmax) || \
 215                 (nfsbuffreecnt > nfsbufcnt) || \
 216                 (nfsbuffreemetacnt < 0) || \
 217                 (nfsbuffreemetacnt > nfsbufmax) || \
 218                 (nfsbuffreemetacnt > nfsbufcnt) || \
 219                 (nfsbuffreemetacnt > nfsbufmetamax) || \
 220                 (nfsbuffreemetacnt > nfsbufmetacnt) || \
 221                 (nfsbufdelwricnt < 0) || \
 222                 (nfsbufdelwricnt > nfsbufmax) || \
 223                 (nfsbufdelwricnt > nfsbufcnt) || \
 224                 (nfs_nbdwrite < 0) || \
 225                 (nfs_nbdwrite > nfsbufcnt) || \
 226                 0) \
 227                 panic("nfsbuf count error: max %d meta %d cnt %d meta %d free %d meta %d delwr %d bdw %d\n", \
 228                         nfsbufmax, nfsbufmetamax, nfsbufcnt, nfsbufmetacnt, nfsbuffreecnt, nfsbuffreemetacnt, \
 229                         nfsbufdelwricnt, nfs_nbdwrite); \
 230         } while (0)
 231 #else
 232 #define NFSBUFCNTCHK()
 233 #endif
 234
 235 /*
 236  * NFS directory buffer
 237  *
 238  * Each buffer for a directory consists of:
 239  *
 240  * - a small header
 241  * - a packed list of direntry structures
 242  *   (if RDIRPLUS is enabled, a file handle and attrstamp are
 243  *   packed after the direntry name.)
 244  * - free/unused space
 245  * - if RDIRPLUS is enabled, an array of attributes
 246  *   that is indexed backwards from the end of the buffer.
 247  */
 248 struct nfs_dir_buf_header {
 249         uint16_t        ndbh_flags;     /* flags (see below) */
 250         uint16_t        ndbh_count;     /* # of entries */
 251         uint32_t        ndbh_entry_end; /* end offset of direntry data */
 252         uint32_t        ndbh_ncgen;     /* name cache generation# */
 253         uint32_t        ndbh_pad;       /* reserved */
 254 };
 255 /* ndbh_flags */
 256 #define NDB_FULL        0x0001  /* buffer has been filled */
 257 #define NDB_EOF         0x0002  /* buffer contains EOF */
 258
 259 #define NFS_DIR_BUF_FIRST_DIRENTRY(BP) \
 260         ((struct direntry*)((char*)((BP)->nb_data) + sizeof(*ndbhp)))
 261 #define NFS_DIR_BUF_NVATTR(BP, IDX) \
 262         (&((struct nfs_vattr*)((char*)((BP)->nb_data) + (BP)->nb_bufsize))[-((IDX)+1)])
 263 #define NFS_DIRENTRY_LEN(namlen) \
 264         ((sizeof(struct direntry) + (namlen) - (MAXPATHLEN-1) + 7) & ~7)
 265 #define NFS_DIRENT_LEN(namlen) \
 266         ((sizeof(struct dirent) - (NAME_MAX+1)) + (((namlen) + 1 + 3) &~ 3))
 267 #define NFS_DIRENTRY_NEXT(DP) \
 268         ((struct direntry*)((char*)(DP) + (DP)->d_reclen))
 269 #define NFS_DIR_COOKIE_POTENTIALLY_TRUNCATED(C) \
 270         ((C) && ((((C) >> 32) == 0) || (((C) & 0x80000000ULL) && (((C) >> 32) == 0xffffffff))))
 271 #define NFS_DIR_COOKIE_SAME32(C1, C2) \
 272         (((C1) & 0xffffffffULL) == ((C2) & 0xffffffffULL))
 273
 274 /*
 275  * NFS directory cookie cache
 276  *
 277  * This structure is used to cache cookie-to-buffer mappings for
 278  * cookies recently returned from READDIR.  The entries are kept in an
 279  * array.  The most-recently-used (MRU) list is headed by the entry at
 280  * index "mru".  The index of the next entry in the list is kept in the
 281  * "next" array.  (An index value of -1 marks an invalid entry.)
 282  */
 283 #define NFSNUMCOOKIES           14
 284 struct nfsdmap {
 285         int8_t          free;                   /* next unused slot */
 286         int8_t          mru;                    /* head of MRU list */
 287         int8_t          next[NFSNUMCOOKIES];    /* MRU list links */
 288         struct {
 289             uint64_t    key;                    /* cookie */
 290             uint64_t    lbn;                    /* lbn of buffer */
 291         } cookies[NFSNUMCOOKIES];               /* MRU list entries */
 292 };
 293
 294 /*
 295  * NFS vnode attribute structure
 296  */
 297 #define NFSTIME_ACCESS  0       /* time of last access */
 298 #define NFSTIME_MODIFY  1       /* time of last modification */
 299 #define NFSTIME_CHANGE  2       /* time file changed */
 300 #define NFSTIME_CREATE  3       /* time file created */
 301 #define NFSTIME_BACKUP  4       /* time of last backup */
 302 #define NFSTIME_COUNT   5
 303
 304 #define NFS_COMPARE_MTIME(TVP, NVAP, CMP) \
 305         (((TVP)->tv_sec == (NVAP)->nva_timesec[NFSTIME_MODIFY]) ?       \
 306          ((TVP)->tv_nsec CMP (NVAP)->nva_timensec[NFSTIME_MODIFY]) :    \
 307          ((TVP)->tv_sec CMP (NVAP)->nva_timesec[NFSTIME_MODIFY]))
 308 #define NFS_COPY_TIME(TVP, NVAP, WHICH) \
 309         do { \
 310         (TVP)->tv_sec = (NVAP)->nva_timesec[NFSTIME_##WHICH]; \
 311         (TVP)->tv_nsec = (NVAP)->nva_timensec[NFSTIME_##WHICH]; \
 312         } while (0)
 313
 314 struct nfs_vattr {
 315         enum vtype      nva_type;       /* vnode type (for create) */
 316         uint32_t        nva_mode;       /* files access mode (and type) */
 317         uid_t           nva_uid;        /* owner user id */
 318         gid_t           nva_gid;        /* owner group id */
 319         nfs_specdata    nva_rawdev;     /* device the special file represents */
 320         uint32_t        nva_flags;      /* file flags */
 321         uint32_t        nva_maxlink;    /* maximum # of links (v4) */
 322         uint64_t        nva_nlink;      /* number of references to file */
 323         uint64_t        nva_fileid;     /* file id */
 324         nfs_fsid        nva_fsid;       /* file system id */
 325         uint64_t        nva_size;       /* file size in bytes */
 326         uint64_t        nva_bytes;      /* bytes of disk space held by file */
 327         uint64_t        nva_change;     /* change attribute */
 328         int64_t         nva_timesec[NFSTIME_COUNT];
 329         int32_t         nva_timensec[NFSTIME_COUNT];
 330         uint32_t        nva_bitmap[NFS_ATTR_BITMAP_LEN]; /* attributes that are valid */
 331 };
 332
 333 #define NFS_FFLAG_ARCHIVED      0x0001
 334 #define NFS_FFLAG_HIDDEN        0x0002
 335 #define NFS_FFLAG_NAMED_ATTR    0x0004  /* file has named attributes */
 336
 337 /* flags for nfs_getattr() */
 338 #define NGA_CACHED      0
 339 #define NGA_UNCACHED    1
 340
 341 /*
 342  * macros for detecting node changes
 343  *
 344  * These macros help us determine if a file has been changed on the server and
 345  * thus whether or not we need to invalidate any cached data.
 346  *
 347  * For NFSv2/v3, the modification time is used.
 348  * For NFSv4, the change attribute is used.
 349  */
 350 #define NFS_CHANGED(VERS, NP, NVAP) \
 351                 (((VERS) >= NFS_VER4) ? \
 352                         ((NP)->n_change != (NVAP)->nva_change) : \
 353                         NFS_COMPARE_MTIME(&(NP)->n_mtime, (NVAP), !=))
 354 #define NFS_CHANGED_NC(VERS, NP, NVAP) \
 355                 (((VERS) >= NFS_VER4) ? \
 356                         ((NP)->n_ncchange != (NVAP)->nva_change) : \
 357                         NFS_COMPARE_MTIME(&(NP)->n_ncmtime, (NVAP), !=))
 358 #define NFS_CHANGED_UPDATE(VERS, NP, NVAP) \
 359         do { \
 360                 if ((VERS) >= NFS_VER4) \
 361                         (NP)->n_change = (NVAP)->nva_change; \
 362                 else \
 363                         NFS_COPY_TIME(&(NP)->n_mtime, (NVAP), MODIFY); \
 364         } while (0)
 365 #define NFS_CHANGED_UPDATE_NC(VERS, NP, NVAP) \
 366         do { \
 367                 if ((VERS) >= NFS_VER4) \
 368                         (NP)->n_ncchange = (NVAP)->nva_change; \
 369                 else \
 370                         NFS_COPY_TIME(&(NP)->n_ncmtime, (NVAP), MODIFY); \
 371         } while (0)
 372
 373
 374 __private_extern__ lck_grp_t *nfs_open_grp;
 375 __private_extern__ uint32_t nfs_open_owner_seqnum, nfs_lock_owner_seqnum;
 376
 377 /*
 378  * NFSv4 open owner structure - one per cred per mount
 379  */
 380 struct nfs_open_owner {
 381         TAILQ_ENTRY(nfs_open_owner)     noo_link;       /* List of open owners (on mount) */
 382         lck_mtx_t                       noo_lock;       /* owner mutex */
 383         struct nfsmount *               noo_mount;      /* NFS mount */
 384         uint32_t                        noo_refcnt;     /* # outstanding references */
 385         uint32_t                        noo_flags;      /* see below */
 386         kauth_cred_t                    noo_cred;       /* credentials of open owner */
 387         uint32_t                        noo_name;       /* unique name used otw */
 388         uint32_t                        noo_seqid;      /* client-side sequence ID */
 389         TAILQ_HEAD(,nfs_open_file)      noo_opens;      /* list of open files */
 390 };
 391 /* noo_flags */
 392 #define NFS_OPEN_OWNER_LINK     0x1     /* linked into mount's open owner list */
 393 #define NFS_OPEN_OWNER_BUSY     0x2     /* open state-modifying operation in progress */
 394 #define NFS_OPEN_OWNER_WANT     0x4     /* someone else wants to mark busy */
 395
 396 /*
 397  * NFS open file structure - one per open owner per nfsnode
 398  */
 399 struct nfs_open_file {
 400         lck_mtx_t                       nof_lock;               /* open file mutex */
 401         TAILQ_ENTRY(nfs_open_file)      nof_link;               /* list of open files */
 402         TAILQ_ENTRY(nfs_open_file)      nof_oolink;             /* list of open owner's open files */
 403         struct nfs_open_owner *         nof_owner;              /* open owner */
 404         nfsnode_t                       nof_np;                 /* nfsnode this open is for */
 405         nfs_stateid                     nof_stateid;            /* open stateid */
 406         thread_t                        nof_creator;            /* thread that created file */
 407         uint32_t                        nof_opencnt;            /* open file count */
 408         uint16_t                        nof_flags;              /* see below */
 409         uint8_t                         nof_access:4;           /* access mode for this open */
 410         uint8_t                         nof_deny:4;             /* deny mode for this open */
 411         uint8_t                         nof_mmap_access:4;      /* mmap open access mode */
 412         uint8_t                         nof_mmap_deny:4;        /* mmap open deny mode */
 413         /* counts of access/deny mode open combinations */
 414         uint32_t                        nof_r;                  /* read opens (deny none) */
 415         uint32_t                        nof_w;                  /* write opens (deny none) */
 416         uint32_t                        nof_rw;                 /* read/write opens (deny none) */
 417         uint32_t                        nof_r_dw;               /* read deny-write opens */
 418         /* the rest of the counts have a max of 2 (1 for open + 1 for mmap) */
 419         uint32_t                        nof_w_dw:4;             /* write deny-write opens (max 2) */
 420         uint32_t                        nof_rw_dw:4;            /* read/write deny-write opens (max 2) */
 421         uint32_t                        nof_r_drw:4;            /* read deny-read/write opens (max 2) */
 422         uint32_t                        nof_w_drw:4;            /* write deny-read/write opens (max 2) */
 423         uint32_t                        nof_rw_drw:4;           /* read/write deny-read/write opens (max 2) */
 424 };
 425 /* nof_flags */
 426 #define NFS_OPEN_FILE_BUSY      0x0001  /* open state-modifying operation in progress */
 427 #define NFS_OPEN_FILE_WANT      0x0002  /* someone else wants to mark busy */
 428 #define NFS_OPEN_FILE_CREATE    0x0004  /* has an open(RW) from a VNOP_CREATE call */
 429 #define NFS_OPEN_FILE_NEEDCLOSE 0x0008  /* has an open(R) from an (unopen) VNOP_READ call */
 430 #define NFS_OPEN_FILE_SETATTR   0x0020  /* has an open(W) to perform a SETATTR(size) */
 431 #define NFS_OPEN_FILE_POSIXLOCK 0x0040  /* server supports POSIX locking semantics */
 432 #define NFS_OPEN_FILE_LOST      0x0080  /* open state has been lost */
 433 #define NFS_OPEN_FILE_REOPEN    0x0100  /* file needs to be reopened */
 434 #define NFS_OPEN_FILE_REOPENING 0x0200  /* file is being reopened */
 435
 436 struct nfs_lock_owner;
 437 /*
 438  * NFS file lock
 439  *
 440  * Each lock request (pending or granted) has an
 441  * nfs_file_lock structure representing its state.
 442  */
 443 struct nfs_file_lock {
 444         TAILQ_ENTRY(nfs_file_lock)      nfl_link;       /* List of locks on nfsnode */
 445         TAILQ_ENTRY(nfs_file_lock)      nfl_lolink;     /* List of locks held by locker */
 446         struct nfs_lock_owner *         nfl_owner;      /* lock owner that holds this lock */
 447         uint64_t                        nfl_start;      /* starting offset */
 448         uint64_t                        nfl_end;        /* ending offset (inclusive) */
 449         uint32_t                        nfl_blockcnt;   /* # locks blocked on this lock */
 450         uint16_t                        nfl_flags;      /* see below */
 451         uint8_t                         nfl_type;       /* lock type: read/write */
 452 };
 453 /* nfl_flags */
 454 #define NFS_FILE_LOCK_ALLOC             0x01    /* lock was allocated */
 455 #define NFS_FILE_LOCK_STYLE_POSIX       0x02    /* POSIX-style fcntl() lock */
 456 #define NFS_FILE_LOCK_STYLE_FLOCK       0x04    /* flock(2)-style lock */
 457 #define NFS_FILE_LOCK_STYLE_MASK        0x06    /* lock style mask */
 458 #define NFS_FILE_LOCK_WAIT              0x08    /* may block on conflicting locks */
 459 #define NFS_FILE_LOCK_BLOCKED           0x10    /* request is blocked */
 460 #define NFS_FILE_LOCK_DEAD              0x20    /* lock (request) no longer exists */
 461
 462 TAILQ_HEAD(nfs_file_lock_queue, nfs_file_lock);
 463
 464 /*
 465  * Calculate length of lock range given the endpoints.
 466  * Note that struct flock has "to EOF" reported as 0 but
 467  * the NFSv4 protocol has "to EOF" reported as UINT64_MAX.
 468  */
 469 #define NFS_FLOCK_LENGTH(S, E)  (((E) == UINT64_MAX) ? 0 : ((E) - (S) + 1))
 470 #define NFS_LOCK_LENGTH(S, E)   (((E) == UINT64_MAX) ? UINT64_MAX : ((E) - (S) + 1))
 471
 472 /*
 473  * NFSv4 lock owner structure - per open owner per process per nfsnode
 474  *
 475  * A lock owner is a process + an nfsnode.
 476  *
 477  * Note that flock(2) locks technically should have the lock owner be
 478  * an fglob pointer instead of a process.  However, implementing that
 479  * correctly would not be trivial.  So, for now, flock(2) locks are
 480  * essentially treated like whole-file POSIX locks.
 481  */
 482 struct nfs_lock_owner {
 483         lck_mtx_t                       nlo_lock;       /* owner mutex */
 484         TAILQ_ENTRY(nfs_lock_owner)     nlo_link;       /* List of lock owners (on nfsnode) */
 485         struct nfs_open_owner *         nlo_open_owner; /* corresponding open owner */
 486         struct nfs_file_lock_queue      nlo_locks;      /* list of locks held */
 487         struct nfs_file_lock            nlo_alock;      /* most lockers will only ever have one */
 488         struct timeval                  nlo_pid_start;  /* Start time of process id */
 489         pid_t                           nlo_pid;        /* lock-owning process ID */
 490         uint32_t                        nlo_refcnt;     /* # outstanding references */
 491         uint32_t                        nlo_flags;      /* see below */
 492         uint32_t                        nlo_name;       /* unique name used otw */
 493         uint32_t                        nlo_seqid;      /* client-side sequence ID */
 494         uint32_t                        nlo_stategenid; /* mount state generation ID */
 495         nfs_stateid                     nlo_stateid;    /* lock stateid */
 496 };
 497 /* nlo_flags */
 498 #define NFS_LOCK_OWNER_LINK     0x1     /* linked into mount's lock owner list */
 499 #define NFS_LOCK_OWNER_BUSY     0x2     /* lock state-modifying operation in progress */
 500 #define NFS_LOCK_OWNER_WANT     0x4     /* someone else wants to mark busy */
 501
 502 /*
 503  * The nfsnode is the NFS equivalent of an inode.
 504  * There is a unique nfsnode for each NFS vnode.
 505  * An nfsnode is 'named' by its file handle. (nget/nfs_node.c)
 506  * NB: Hopefully the current order of the fields is such that everything will
 507  *     be well aligned and, therefore, tightly packed.
 508  */
 509
 510 #define NFS_ACCESS_CACHE_SIZE   3
 511
 512 struct nfsnode {
 513         lck_mtx_t               n_lock;         /* nfs node lock */
 514         lck_rw_t                n_datalock;     /* nfs node data lock */
 515         void                    *n_datalockowner;/* nfs node data lock owner (exclusive) */
 516         LIST_ENTRY(nfsnode)     n_hash;         /* Hash chain */
 517         u_quad_t                n_size;         /* Current size of file */
 518         u_quad_t                n_newsize;      /* new size of file (pending update) */
 519         u_int64_t               n_xid;          /* last xid to loadattr */
 520         struct nfs_vattr        n_vattr;        /* Vnode attribute cache */
 521         time_t                  n_attrstamp;    /* Attr. cache timestamp */
 522         u_int8_t                n_mode[NFS_ACCESS_CACHE_SIZE+1];        /* ACCESS mode cache */
 523         uid_t                   n_modeuid[NFS_ACCESS_CACHE_SIZE];       /* credentials having mode */
 524         time_t                  n_modestamp[NFS_ACCESS_CACHE_SIZE];     /* mode cache timestamp */
 525         union {
 526             struct {
 527                 struct timespec n3_mtime;       /* Prev modify time. */
 528                 struct timespec n3_ncmtime;     /* namecache modify time. */
 529             } v3;
 530             struct {
 531                 uint64_t        n4_change;      /* prev change attribute */
 532                 uint64_t        n4_ncchange;    /* namecache change attribute */
 533             } v4;
 534         } n_un4;
 535         vnode_t                 n_parent;       /* this node's parent */
 536         u_char                  *n_fhp;         /* NFS File Handle */
 537         vnode_t                 n_vnode;        /* associated vnode */
 538         mount_t                 n_mount;        /* associated mount (NHINIT) */
 539         int                     n_error;        /* Save write error value */
 540         union {
 541                 struct timespec ns_atim;        /* Special file times */
 542                 daddr64_t       nf_lastread;    /* last block# read from (for readahead) */
 543                 uint64_t        nd_cookieverf;  /* Cookie verifier (dir only) */
 544         } n_un1;
 545         union {
 546                 struct timespec ns_mtim;        /* Special file times */
 547                 daddr64_t       nf_lastrahead;  /* last block# read ahead */
 548                 uint64_t        nd_eofcookie;   /* Dir. EOF cookie cache */
 549         } n_un2;
 550         union {
 551                 struct nfs_sillyrename *nf_silly;/* Ptr to silly rename struct */
 552                 struct nfsdmap *nd_cookiecache; /* dir cookie cache */
 553         } n_un3;
 554         u_short                 n_fhsize;       /* size in bytes, of fh */
 555         u_short                 n_flag;         /* node flags */
 556         u_short                 n_hflag;        /* node hash flags */
 557         u_short                 n_bflag;        /* node buffer flags */
 558         u_char                  n_fh[NFS_SMALLFH];/* Small File Handle */
 559         struct nfsbuflists      n_cleanblkhd;   /* clean blocklist head */
 560         struct nfsbuflists      n_dirtyblkhd;   /* dirty blocklist head */
 561         union {
 562                 int             nf_wrbusy;      /* # threads in write/fsync */
 563                 uint32_t        nd_ncgen;       /* dir name cache generation# */
 564         } n_un5;
 565         union {
 566                 int             nf_needcommitcnt;/* # bufs that need committing */
 567                 daddr64_t       nd_lastdbl;     /* last dir buf lookup block# */
 568         } n_un6;
 569         int                     n_bufiterflags; /* buf iterator flags */
 570         int                     n_numoutput;    /* I/O in progress */
 571         /* open state */
 572         lck_mtx_t               n_openlock;     /* nfs node open lock */
 573         uint32_t                n_openflags;    /* open state flags */
 574         uint32_t                n_openrefcnt;   /* # non-file opens */
 575         TAILQ_HEAD(,nfs_open_file) n_opens;     /* list of open files */
 576         /* lock state */
 577         TAILQ_HEAD(, nfs_lock_owner) n_lock_owners; /* list of lock owners */
 578         struct nfs_file_lock_queue n_locks;     /* list of locks */
 579         /* delegation state */
 580         nfs_stateid             n_dstateid;     /* delegation stateid */
 581         TAILQ_ENTRY(nfsnode)    n_dlink;        /* delegation recall list link */
 582 };
 583
 584 #define NFS_DATA_LOCK_SHARED    1
 585 #define NFS_DATA_LOCK_EXCLUSIVE 2
 586
 587 #define nfstimespeccmp(tvp, uvp, cmp)           \
 588         (((tvp)->tv_sec == (uvp)->tv_sec) ?     \
 589          ((tvp)->tv_nsec cmp (uvp)->tv_nsec) :  \
 590          ((tvp)->tv_sec cmp (uvp)->tv_sec))
 591
 592 #define CHECK_NEEDCOMMITCNT(np) \
 593         do { \
 594                 if ((np)->n_needcommitcnt < 0) { \
 595                         printf("nfs: n_needcommitcnt negative\n"); \
 596                         (np)->n_needcommitcnt = 0; \
 597                 } \
 598         } while (0)
 599
 600 #define n_atim                  n_un1.ns_atim
 601 #define n_mtim                  n_un2.ns_mtim
 602 #define n_lastread              n_un1.nf_lastread
 603 #define n_lastrahead            n_un2.nf_lastrahead
 604 #define n_sillyrename           n_un3.nf_silly
 605 #define n_wrbusy                n_un5.nf_wrbusy
 606 #define n_needcommitcnt         n_un6.nf_needcommitcnt
 607 #define n_cookieverf            n_un1.nd_cookieverf
 608 #define n_eofcookie             n_un2.nd_eofcookie
 609 #define n_cookiecache           n_un3.nd_cookiecache
 610 #define n_ncgen                 n_un5.nd_ncgen
 611 #define n_lastdbl               n_un6.nd_lastdbl
 612 #define n_mtime                 n_un4.v3.n3_mtime
 613 #define n_ncmtime               n_un4.v3.n3_ncmtime
 614 #define n_change                n_un4.v4.n4_change
 615 #define n_ncchange              n_un4.v4.n4_ncchange
 616
 617 /*
 618  * Flags for n_flag
 619  */
 620 #define NUPDATESIZE     0x0001  /* size of file needs updating */
 621 #define NMODIFIED       0x0004  /* Might have a modified buffer in bio */
 622 #define NWRITEERR       0x0008  /* Flag write errors so close will know */
 623 #define NNEEDINVALIDATE 0x0010  /* need to call vinvalbuf() */
 624 #define NGETATTRINPROG  0x0020  /* GETATTR RPC in progress */
 625 #define NGETATTRWANT    0x0040  /* waiting for GETATTR RPC */
 626 #define NACC            0x0100  /* Special file accessed */
 627 #define NUPD            0x0200  /* Special file updated */
 628 #define NCHG            0x0400  /* Special file times changed */
 629 #define NNEGNCENTRIES   0x0800  /* directory has negative name cache entries */
 630 #define NBUSY           0x1000  /* node is busy */
 631 #define NBUSYWANT       0x2000  /* waiting on busy node */
 632
 633 /*
 634  * Flags for n_hflag
 635  * Note: protected by nfs_node_hash_mutex
 636  */
 637 #define NHHASHED        0x0001  /* node is in hash table */
 638 #define NHINIT          0x0002  /* node is being initialized */
 639 #define NHLOCKED        0x0004  /* node is locked (initting or deleting) */
 640 #define NHLOCKWANT      0x0008  /* someone wants to lock */
 641
 642 /*
 643  * Flags for n_bflag
 644  * Note: protected by nfs_buf_mutex
 645  */
 646 #define NBFLUSHINPROG   0x0001  /* Avoid multiple calls to nfs_flush() */
 647 #define NBFLUSHWANT     0x0002  /* waiting for nfs_flush() to complete */
 648 #define NBINVALINPROG   0x0004  /* Avoid multiple calls to nfs_vinvalbuf() */
 649 #define NBINVALWANT     0x0008  /* waiting for nfs_vinvalbuf() to complete */
 650
 651 /*
 652  * n_openflags
 653  * Note: protected by n_openlock
 654  */
 655 #define N_OPENBUSY              0x0001  /* open state is busy - being updated */
 656 #define N_OPENWANT              0x0002  /* someone wants to mark busy */
 657 #define N_DELEG_READ            0x0004  /* we have a read delegation */
 658 #define N_DELEG_WRITE           0x0008  /* we have a write delegation */
 659 #define N_DELEG_MASK            0x000c  /* delegation mask */
 660
 661 /* attr/mode timestamp macros */
 662 #define NATTRVALID(np)          ((np)->n_attrstamp != ~0)
 663 #define NATTRINVALIDATE(np)     ((np)->n_attrstamp = ~0)
 664 #define NMODEVALID(np, slot)    (((slot) >= 0) && ((slot) < 3) && ((np)->n_modestamp[(slot)] != ~0))
 665 #define NMODEINVALIDATE(np) \
 666         do { \
 667                 (np)->n_modestamp[0] = ~0; \
 668                 (np)->n_modestamp[1] = ~0; \
 669                 (np)->n_modestamp[2] = ~0; \
 670                 (np)->n_mode[3] = 0; \
 671         } while (0)
 672
 673 /*
 674  * NFS-specific flags for nfs_vinvalbuf/nfs_flush
 675  */
 676 #define V_IGNORE_WRITEERR       0x8000
 677
 678 /*
 679  * Flags for nfs_nget()
 680  */
 681 #define NG_MARKROOT     0x0001  /* mark vnode as root of FS */
 682 #define NG_MAKEENTRY    0x0002  /* add name cache entry for vnode */
 683 #define NG_NOCREATE     0x0004  /* don't create a new node, return existing one */
 684
 685 /*
 686  * Convert between nfsnode pointers and vnode pointers
 687  */
 688 #define VTONFS(vp)      ((nfsnode_t)vnode_fsnode(vp))
 689 #define NFSTOV(np)      ((np)->n_vnode)
 690
 691 /* nfsnode hash table mutex */
 692 __private_extern__ lck_mtx_t *nfs_node_hash_mutex;
 693
 694 /*
 695  * nfsiod structures
 696  */
 697 struct nfsiod {
 698         TAILQ_ENTRY(nfsiod)     niod_link;      /* List of nfsiods */
 699         struct nfsmount *       niod_nmp;       /* mount point for this nfsiod */
 700 };
 701 TAILQ_HEAD(nfsiodlist, nfsiod);
 702 TAILQ_HEAD(nfsiodmountlist, nfsmount);
 703 __private_extern__ struct nfsiodlist nfsiodfree, nfsiodwork;
 704 __private_extern__ struct nfsiodmountlist nfsiodmounts;
 705 __private_extern__ lck_mtx_t *nfsiod_mutex;
 706
 707 #if defined(KERNEL)
 708
 709 typedef int     vnop_t(void *);
 710 extern  vnop_t  **fifo_nfsv2nodeop_p;
 711 extern  vnop_t  **nfsv2_vnodeop_p;
 712 extern  vnop_t  **spec_nfsv2nodeop_p;
 713 extern  vnop_t  **fifo_nfsv4nodeop_p;
 714 extern  vnop_t  **nfsv4_vnodeop_p;
 715 extern  vnop_t  **spec_nfsv4nodeop_p;
 716
 717 /*
 718  * Prototypes for NFS vnode operations
 719  */
 720 #define nfs_vnop_revoke nop_revoke
 721 int     nfs_vnop_inactive(struct vnop_inactive_args *);
 722 int     nfs_vnop_reclaim(struct vnop_reclaim_args *);
 723
 724 int nfs_node_lock(nfsnode_t);
 725 int nfs_node_lock_internal(nfsnode_t, int);
 726 void nfs_node_lock_force(nfsnode_t);
 727 void nfs_node_unlock(nfsnode_t);
 728 int nfs_node_lock2(nfsnode_t, nfsnode_t);
 729 void nfs_node_unlock2(nfsnode_t, nfsnode_t);
 730 int nfs_node_set_busy(nfsnode_t, thread_t);
 731 int nfs_node_set_busy2(nfsnode_t, nfsnode_t, thread_t);
 732 int nfs_node_set_busy4(nfsnode_t, nfsnode_t, nfsnode_t, nfsnode_t, thread_t);
 733 void nfs_node_clear_busy(nfsnode_t);
 734 void nfs_node_clear_busy2(nfsnode_t, nfsnode_t);
 735 void nfs_node_clear_busy4(nfsnode_t, nfsnode_t, nfsnode_t, nfsnode_t);
 736 void nfs_data_lock(nfsnode_t, int);
 737 void nfs_data_lock_noupdate(nfsnode_t, int);
 738 void nfs_data_lock_internal(nfsnode_t, int, int);
 739 void nfs_data_unlock(nfsnode_t);
 740 void nfs_data_unlock_noupdate(nfsnode_t);
 741 void nfs_data_unlock_internal(nfsnode_t, int);
 742 void nfs_data_update_size(nfsnode_t, int);
 743
 744 /* other stuff */
 745 int nfs_removeit(struct nfs_sillyrename *);
 746 int nfs_nget(mount_t,nfsnode_t,struct componentname *,u_char *,int,struct nfs_vattr *,u_int64_t *,int,nfsnode_t*);
 747 void nfs_dir_cookie_cache(nfsnode_t, uint64_t, uint64_t);
 748 int nfs_dir_cookie_to_lbn(nfsnode_t, uint64_t, int *, uint64_t *);
 749 void nfs_invaldir(nfsnode_t);
 750 uint32_t nfs_dir_buf_freespace(struct nfsbuf *, int);
 751
 752 /* nfsbuf functions */
 753 void nfs_nbinit(void);
 754 void nfs_buf_timer(void *, void *);
 755 void nfs_buf_remfree(struct nfsbuf *);
 756 boolean_t nfs_buf_is_incore(nfsnode_t, daddr64_t);
 757 struct nfsbuf * nfs_buf_incore(nfsnode_t, daddr64_t);
 758 int nfs_buf_get(nfsnode_t, daddr64_t, uint32_t, thread_t, int, struct nfsbuf **);
 759 int nfs_buf_upl_setup(struct nfsbuf *bp);
 760 void nfs_buf_upl_check(struct nfsbuf *bp);
 761 void nfs_buf_normalize_valid_range(nfsnode_t, struct nfsbuf *);
 762 int nfs_buf_map(struct nfsbuf *);
 763 void nfs_buf_release(struct nfsbuf *, int);
 764 int nfs_buf_iowait(struct nfsbuf *);
 765 void nfs_buf_iodone(struct nfsbuf *);
 766 void nfs_buf_write_delayed(struct nfsbuf *);
 767 void nfs_buf_check_write_verifier(nfsnode_t, struct nfsbuf *);
 768 void nfs_buf_freeup(int);
 769 void nfs_buf_refget(struct nfsbuf *bp);
 770 void nfs_buf_refrele(struct nfsbuf *bp);
 771 void nfs_buf_drop(struct nfsbuf *);
 772 errno_t nfs_buf_acquire(struct nfsbuf *, int, int, int);
 773 int nfs_buf_iterprepare(nfsnode_t, struct nfsbuflists *, int);
 774 void nfs_buf_itercomplete(nfsnode_t, struct nfsbuflists *, int);
 775
 776 int nfs_bioread(nfsnode_t, uio_t, int, vfs_context_t);
 777 int nfs_buf_readahead(nfsnode_t, int, daddr64_t *, daddr64_t, thread_t, kauth_cred_t);
 778 int nfs_buf_readdir(struct nfsbuf *, vfs_context_t);
 779 int nfs_buf_read(struct nfsbuf *);
 780 void nfs_buf_read_finish(struct nfsbuf *);
 781 int nfs_buf_read_rpc(struct nfsbuf *, thread_t, kauth_cred_t);
 782 void nfs_buf_read_rpc_finish(struct nfsreq *);
 783 int nfs_buf_write(struct nfsbuf *);
 784 void nfs_buf_write_finish(struct nfsbuf *, thread_t, kauth_cred_t);
 785 int nfs_buf_write_rpc(struct nfsbuf *, int, thread_t, kauth_cred_t);
 786 void nfs_buf_write_rpc_finish(struct nfsreq *);
 787 int nfs_buf_write_dirty_pages(struct nfsbuf *, thread_t, kauth_cred_t);
 788
 789 int nfs_flushcommits(nfsnode_t, int);
 790 int nfs_flush(nfsnode_t, int, thread_t, int);
 791 void nfs_buf_delwri_push(int);
 792 void nfs_buf_delwri_service(void);
 793 void nfs_buf_delwri_thread(void *, wait_result_t);;
 794
 795 int nfsiod_start(void);
 796 void nfsiod_terminate(struct nfsiod *);
 797 void nfsiod_thread(void);
 798 int nfsiod_continue(int);
 799 void nfs_asyncio_finish(struct nfsreq *);
 800 void nfs_asyncio_resend(struct nfsreq *);
 801 int nfs_async_write_start(struct nfsmount *);
 802 void nfs_async_write_done(struct nfsmount *);
 803
 804 #endif /* KERNEL */
 805
 806 #endif /* __APPLE_API_PRIVATE */
 807 #endif /* _NFS_NFSNODE_H_ */