1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25 
  26 /*      Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
  27 /*        All Rights Reserved   */
  28 
  29 /*
  30  * University Copyright- Copyright (c) 1982, 1986, 1988
  31  * The Regents of the University of California
  32  * All Rights Reserved
  33  *
  34  * University Acknowledgment- Portions of this document are derived from
  35  * software developed by the University of California, Berkeley, and its
  36  * contributors.
  37  */
  38 
  39 #include <sys/types.h>
  40 #include <sys/param.h>
  41 #include <sys/t_lock.h>
  42 #include <sys/errno.h>
  43 #include <sys/cred.h>
  44 #include <sys/user.h>
  45 #include <sys/uio.h>
  46 #include <sys/file.h>
  47 #include <sys/pathname.h>
  48 #include <sys/vfs.h>
  49 #include <sys/vfs_opreg.h>
  50 #include <sys/vnode.h>
  51 #include <sys/rwstlock.h>
  52 #include <sys/fem.h>
  53 #include <sys/stat.h>
  54 #include <sys/mode.h>
  55 #include <sys/conf.h>
  56 #include <sys/sysmacros.h>
  57 #include <sys/cmn_err.h>
  58 #include <sys/systm.h>
  59 #include <sys/kmem.h>
  60 #include <sys/debug.h>
  61 #include <c2/audit.h>
  62 #include <sys/acl.h>
  63 #include <sys/nbmlock.h>
  64 #include <sys/fcntl.h>
  65 #include <fs/fs_subr.h>
  66 #include <sys/taskq.h>
  67 #include <fs/fs_reparse.h>
  68 
  69 /* Determine if this vnode is a file that is read-only */
  70 #define ISROFILE(vp)    \
  71         ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \
  72             (vp)->v_type != VFIFO && vn_is_readonly(vp))
  73 
  74 /* Tunable via /etc/system; used only by admin/install */
  75 int nfs_global_client_only;
  76 
  77 /*
  78  * Array of vopstats_t for per-FS-type vopstats.  This array has the same
  79  * number of entries as and parallel to the vfssw table.  (Arguably, it could
  80  * be part of the vfssw table.)  Once it's initialized, it's accessed using
  81  * the same fstype index that is used to index into the vfssw table.
  82  */
  83 vopstats_t **vopstats_fstype;
  84 
  85 /* vopstats initialization template used for fast initialization via bcopy() */
  86 static vopstats_t *vs_templatep;
  87 
  88 /* Kmem cache handle for vsk_anchor_t allocations */
  89 kmem_cache_t *vsk_anchor_cache;
  90 
  91 /* file events cleanup routine */
  92 extern void free_fopdata(vnode_t *);
  93 
  94 /*
  95  * Root of AVL tree for the kstats associated with vopstats.  Lock protects
  96  * updates to vsktat_tree.
  97  */
  98 avl_tree_t      vskstat_tree;
  99 kmutex_t        vskstat_tree_lock;
 100 
 101 /* Global variable which enables/disables the vopstats collection */
 102 int vopstats_enabled = 1;
 103 
 104 /*
 105  * forward declarations for internal vnode specific data (vsd)
 106  */
 107 static void *vsd_realloc(void *, size_t, size_t);
 108 
 109 /*
 110  * forward declarations for reparse point functions
 111  */
 112 static int fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr);
 113 
 114 /*
 115  * VSD -- VNODE SPECIFIC DATA
 116  * The v_data pointer is typically used by a file system to store a
 117  * pointer to the file system's private node (e.g. ufs inode, nfs rnode).
 118  * However, there are times when additional project private data needs
 119  * to be stored separately from the data (node) pointed to by v_data.
 120  * This additional data could be stored by the file system itself or
 121  * by a completely different kernel entity.  VSD provides a way for
 122  * callers to obtain a key and store a pointer to private data associated
 123  * with a vnode.
 124  *
 125  * Callers are responsible for protecting the vsd by holding v_vsd_lock
 126  * for calls to vsd_set() and vsd_get().
 127  */
 128 
 129 /*
 130  * vsd_lock protects:
 131  *   vsd_nkeys - creation and deletion of vsd keys
 132  *   vsd_list - insertion and deletion of vsd_node in the vsd_list
 133  *   vsd_destructor - adding and removing destructors to the list
 134  */
 135 static kmutex_t         vsd_lock;
 136 static uint_t           vsd_nkeys;       /* size of destructor array */
 137 /* list of vsd_node's */
 138 static list_t *vsd_list = NULL;
 139 /* per-key destructor funcs */
 140 static void             (**vsd_destructor)(void *);
 141 
 142 /*
 143  * The following is the common set of actions needed to update the
 144  * vopstats structure from a vnode op.  Both VOPSTATS_UPDATE() and
 145  * VOPSTATS_UPDATE_IO() do almost the same thing, except for the
 146  * recording of the bytes transferred.  Since the code is similar
 147  * but small, it is nearly a duplicate.  Consequently any changes
 148  * to one may need to be reflected in the other.
 149  * Rundown of the variables:
 150  * vp - Pointer to the vnode
 151  * counter - Partial name structure member to update in vopstats for counts
 152  * bytecounter - Partial name structure member to update in vopstats for bytes
 153  * bytesval - Value to update in vopstats for bytes
 154  * fstype - Index into vsanchor_fstype[], same as index into vfssw[]
 155  * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i])
 156  */
 157 
 158 #define VOPSTATS_UPDATE(vp, counter) {                                  \
 159         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 160         if (vfsp && vfsp->vfs_implp &&                                       \
 161             (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {   \
 162                 vopstats_t *vsp = &vfsp->vfs_vopstats;                   \
 163                 uint64_t *stataddr = &(vsp->n##counter.value.ui64);      \
 164                 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
 165                     size_t, uint64_t *);                                \
 166                 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr);     \
 167                 (*stataddr)++;                                          \
 168                 if ((vsp = vfsp->vfs_fstypevsp) != NULL) {           \
 169                         vsp->n##counter.value.ui64++;                        \
 170                 }                                                       \
 171         }                                                               \
 172 }
 173 
 174 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) {        \
 175         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 176         if (vfsp && vfsp->vfs_implp &&                                       \
 177             (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {   \
 178                 vopstats_t *vsp = &vfsp->vfs_vopstats;                   \
 179                 uint64_t *stataddr = &(vsp->n##counter.value.ui64);      \
 180                 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
 181                     size_t, uint64_t *);                                \
 182                 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \
 183                 (*stataddr)++;                                          \
 184                 vsp->bytecounter.value.ui64 += bytesval;             \
 185                 if ((vsp = vfsp->vfs_fstypevsp) != NULL) {           \
 186                         vsp->n##counter.value.ui64++;                        \
 187                         vsp->bytecounter.value.ui64 += bytesval;     \
 188                 }                                                       \
 189         }                                                               \
 190 }
 191 
 192 /*
 193  * If the filesystem does not support XIDs map credential
 194  * If the vfsp is NULL, perhaps we should also map?
 195  */
 196 #define VOPXID_MAP_CR(vp, cr)   {                                       \
 197         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 198         if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0)             \
 199                 cr = crgetmapped(cr);                                   \
 200         }
 201 
 202 /*
 203  * Convert stat(2) formats to vnode types and vice versa.  (Knows about
 204  * numerical order of S_IFMT and vnode types.)
 205  */
 206 enum vtype iftovt_tab[] = {
 207         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
 208         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
 209 };
 210 
 211 ushort_t vttoif_tab[] = {
 212         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO,
 213         S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0
 214 };
 215 
 216 /*
 217  * The system vnode cache.
 218  */
 219 
 220 kmem_cache_t *vn_cache;
 221 
 222 
 223 /*
 224  * Vnode operations vector.
 225  */
 226 
 227 static const fs_operation_trans_def_t vn_ops_table[] = {
 228         VOPNAME_OPEN, offsetof(struct vnodeops, vop_open),
 229             fs_nosys, fs_nosys,
 230 
 231         VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close),
 232             fs_nosys, fs_nosys,
 233 
 234         VOPNAME_READ, offsetof(struct vnodeops, vop_read),
 235             fs_nosys, fs_nosys,
 236 
 237         VOPNAME_WRITE, offsetof(struct vnodeops, vop_write),
 238             fs_nosys, fs_nosys,
 239 
 240         VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl),
 241             fs_nosys, fs_nosys,
 242 
 243         VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl),
 244             fs_setfl, fs_nosys,
 245 
 246         VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr),
 247             fs_nosys, fs_nosys,
 248 
 249         VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr),
 250             fs_nosys, fs_nosys,
 251 
 252         VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access),
 253             fs_nosys, fs_nosys,
 254 
 255         VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup),
 256             fs_nosys, fs_nosys,
 257 
 258         VOPNAME_CREATE, offsetof(struct vnodeops, vop_create),
 259             fs_nosys, fs_nosys,
 260 
 261         VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove),
 262             fs_nosys, fs_nosys,
 263 
 264         VOPNAME_LINK, offsetof(struct vnodeops, vop_link),
 265             fs_nosys, fs_nosys,
 266 
 267         VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename),
 268             fs_nosys, fs_nosys,
 269 
 270         VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir),
 271             fs_nosys, fs_nosys,
 272 
 273         VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir),
 274             fs_nosys, fs_nosys,
 275 
 276         VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir),
 277             fs_nosys, fs_nosys,
 278 
 279         VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink),
 280             fs_nosys, fs_nosys,
 281 
 282         VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink),
 283             fs_nosys, fs_nosys,
 284 
 285         VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync),
 286             fs_nosys, fs_nosys,
 287 
 288         VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive),
 289             fs_nosys, fs_nosys,
 290 
 291         VOPNAME_FID, offsetof(struct vnodeops, vop_fid),
 292             fs_nosys, fs_nosys,
 293 
 294         VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock),
 295             fs_rwlock, fs_rwlock,
 296 
 297         VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock),
 298             (fs_generic_func_p) fs_rwunlock,
 299             (fs_generic_func_p) fs_rwunlock,    /* no errors allowed */
 300 
 301         VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek),
 302             fs_nosys, fs_nosys,
 303 
 304         VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp),
 305             fs_cmp, fs_cmp,             /* no errors allowed */
 306 
 307         VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock),
 308             fs_frlock, fs_nosys,
 309 
 310         VOPNAME_SPACE, offsetof(struct vnodeops, vop_space),
 311             fs_nosys, fs_nosys,
 312 
 313         VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp),
 314             fs_nosys, fs_nosys,
 315 
 316         VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage),
 317             fs_nosys, fs_nosys,
 318 
 319         VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage),
 320             fs_nosys, fs_nosys,
 321 
 322         VOPNAME_MAP, offsetof(struct vnodeops, vop_map),
 323             (fs_generic_func_p) fs_nosys_map,
 324             (fs_generic_func_p) fs_nosys_map,
 325 
 326         VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap),
 327             (fs_generic_func_p) fs_nosys_addmap,
 328             (fs_generic_func_p) fs_nosys_addmap,
 329 
 330         VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap),
 331             fs_nosys, fs_nosys,
 332 
 333         VOPNAME_POLL, offsetof(struct vnodeops, vop_poll),
 334             (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll,
 335 
 336         VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump),
 337             fs_nosys, fs_nosys,
 338 
 339         VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf),
 340             fs_pathconf, fs_nosys,
 341 
 342         VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio),
 343             fs_nosys, fs_nosys,
 344 
 345         VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl),
 346             fs_nosys, fs_nosys,
 347 
 348         VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose),
 349             (fs_generic_func_p) fs_dispose,
 350             (fs_generic_func_p) fs_nodispose,
 351 
 352         VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr),
 353             fs_nosys, fs_nosys,
 354 
 355         VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr),
 356             fs_fab_acl, fs_nosys,
 357 
 358         VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock),
 359             fs_shrlock, fs_nosys,
 360 
 361         VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent),
 362             (fs_generic_func_p) fs_vnevent_nosupport,
 363             (fs_generic_func_p) fs_vnevent_nosupport,
 364 
 365         VOPNAME_REQZCBUF, offsetof(struct vnodeops, vop_reqzcbuf),
 366             fs_nosys, fs_nosys,
 367 
 368         VOPNAME_RETZCBUF, offsetof(struct vnodeops, vop_retzcbuf),
 369             fs_nosys, fs_nosys,
 370 
 371         NULL, 0, NULL, NULL
 372 };
 373 
 374 /* Extensible attribute (xva) routines. */
 375 
 376 /*
 377  * Zero out the structure, set the size of the requested/returned bitmaps,
 378  * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer
 379  * to the returned attributes array.
 380  */
 381 void
 382 xva_init(xvattr_t *xvap)
 383 {
 384         bzero(xvap, sizeof (xvattr_t));
 385         xvap->xva_mapsize = XVA_MAPSIZE;
 386         xvap->xva_magic = XVA_MAGIC;
 387         xvap->xva_vattr.va_mask = AT_XVATTR;
 388         xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0];
 389 }
 390 
 391 /*
 392  * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t
 393  * structure.  Otherwise, returns NULL.
 394  */
 395 xoptattr_t *
 396 xva_getxoptattr(xvattr_t *xvap)
 397 {
 398         xoptattr_t *xoap = NULL;
 399         if (xvap->xva_vattr.va_mask & AT_XVATTR)
 400                 xoap = &xvap->xva_xoptattrs;
 401         return (xoap);
 402 }
 403 
 404 /*
 405  * Used by the AVL routines to compare two vsk_anchor_t structures in the tree.
 406  * We use the f_fsid reported by VFS_STATVFS() since we use that for the
 407  * kstat name.
 408  */
 409 static int
 410 vska_compar(const void *n1, const void *n2)
 411 {
 412         int ret;
 413         ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid;
 414         ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid;
 415 
 416         if (p1 < p2) {
 417                 ret = -1;
 418         } else if (p1 > p2) {
 419                 ret = 1;
 420         } else {
 421                 ret = 0;
 422         }
 423 
 424         return (ret);
 425 }
 426 
 427 /*
 428  * Used to create a single template which will be bcopy()ed to a newly
 429  * allocated vsanchor_combo_t structure in new_vsanchor(), below.
 430  */
 431 static vopstats_t *
 432 create_vopstats_template()
 433 {
 434         vopstats_t              *vsp;
 435 
 436         vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP);
 437         bzero(vsp, sizeof (*vsp));      /* Start fresh */
 438 
 439         /* VOP_OPEN */
 440         kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64);
 441         /* VOP_CLOSE */
 442         kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64);
 443         /* VOP_READ I/O */
 444         kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64);
 445         kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64);
 446         /* VOP_WRITE I/O */
 447         kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64);
 448         kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64);
 449         /* VOP_IOCTL */
 450         kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64);
 451         /* VOP_SETFL */
 452         kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64);
 453         /* VOP_GETATTR */
 454         kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64);
 455         /* VOP_SETATTR */
 456         kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64);
 457         /* VOP_ACCESS */
 458         kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64);
 459         /* VOP_LOOKUP */
 460         kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64);
 461         /* VOP_CREATE */
 462         kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64);
 463         /* VOP_REMOVE */
 464         kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64);
 465         /* VOP_LINK */
 466         kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64);
 467         /* VOP_RENAME */
 468         kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64);
 469         /* VOP_MKDIR */
 470         kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64);
 471         /* VOP_RMDIR */
 472         kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64);
 473         /* VOP_READDIR I/O */
 474         kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64);
 475         kstat_named_init(&vsp->readdir_bytes, "readdir_bytes",
 476             KSTAT_DATA_UINT64);
 477         /* VOP_SYMLINK */
 478         kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64);
 479         /* VOP_READLINK */
 480         kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64);
 481         /* VOP_FSYNC */
 482         kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64);
 483         /* VOP_INACTIVE */
 484         kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64);
 485         /* VOP_FID */
 486         kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64);
 487         /* VOP_RWLOCK */
 488         kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64);
 489         /* VOP_RWUNLOCK */
 490         kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64);
 491         /* VOP_SEEK */
 492         kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64);
 493         /* VOP_CMP */
 494         kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64);
 495         /* VOP_FRLOCK */
 496         kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64);
 497         /* VOP_SPACE */
 498         kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64);
 499         /* VOP_REALVP */
 500         kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64);
 501         /* VOP_GETPAGE */
 502         kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64);
 503         /* VOP_PUTPAGE */
 504         kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64);
 505         /* VOP_MAP */
 506         kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64);
 507         /* VOP_ADDMAP */
 508         kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64);
 509         /* VOP_DELMAP */
 510         kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64);
 511         /* VOP_POLL */
 512         kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64);
 513         /* VOP_DUMP */
 514         kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64);
 515         /* VOP_PATHCONF */
 516         kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64);
 517         /* VOP_PAGEIO */
 518         kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64);
 519         /* VOP_DUMPCTL */
 520         kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64);
 521         /* VOP_DISPOSE */
 522         kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64);
 523         /* VOP_SETSECATTR */
 524         kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64);
 525         /* VOP_GETSECATTR */
 526         kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64);
 527         /* VOP_SHRLOCK */
 528         kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64);
 529         /* VOP_VNEVENT */
 530         kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64);
 531         /* VOP_REQZCBUF */
 532         kstat_named_init(&vsp->nreqzcbuf, "nreqzcbuf", KSTAT_DATA_UINT64);
 533         /* VOP_RETZCBUF */
 534         kstat_named_init(&vsp->nretzcbuf, "nretzcbuf", KSTAT_DATA_UINT64);
 535 
 536         return (vsp);
 537 }
 538 
 539 /*
 540  * Creates a kstat structure associated with a vopstats structure.
 541  */
 542 kstat_t *
 543 new_vskstat(char *ksname, vopstats_t *vsp)
 544 {
 545         kstat_t         *ksp;
 546 
 547         if (!vopstats_enabled) {
 548                 return (NULL);
 549         }
 550 
 551         ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED,
 552             sizeof (vopstats_t)/sizeof (kstat_named_t),
 553             KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE);
 554         if (ksp) {
 555                 ksp->ks_data = vsp;
 556                 kstat_install(ksp);
 557         }
 558 
 559         return (ksp);
 560 }
 561 
 562 /*
 563  * Called from vfsinit() to initialize the support mechanisms for vopstats
 564  */
 565 void
 566 vopstats_startup()
 567 {
 568         if (!vopstats_enabled)
 569                 return;
 570 
 571         /*
 572          * Creates the AVL tree which holds per-vfs vopstat anchors.  This
 573          * is necessary since we need to check if a kstat exists before we
 574          * attempt to create it.  Also, initialize its lock.
 575          */
 576         avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t),
 577             offsetof(vsk_anchor_t, vsk_node));
 578         mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL);
 579 
 580         vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache",
 581             sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL,
 582             NULL, NULL, 0);
 583 
 584         /*
 585          * Set up the array of pointers for the vopstats-by-FS-type.
 586          * The entries will be allocated/initialized as each file system
 587          * goes through modload/mod_installfs.
 588          */
 589         vopstats_fstype = (vopstats_t **)kmem_zalloc(
 590             (sizeof (vopstats_t *) * nfstype), KM_SLEEP);
 591 
 592         /* Set up the global vopstats initialization template */
 593         vs_templatep = create_vopstats_template();
 594 }
 595 
 596 /*
 597  * We need to have the all of the counters zeroed.
 598  * The initialization of the vopstats_t includes on the order of
 599  * 50 calls to kstat_named_init().  Rather that do that on every call,
 600  * we do it once in a template (vs_templatep) then bcopy it over.
 601  */
 602 void
 603 initialize_vopstats(vopstats_t *vsp)
 604 {
 605         if (vsp == NULL)
 606                 return;
 607 
 608         bcopy(vs_templatep, vsp, sizeof (vopstats_t));
 609 }
 610 
 611 /*
 612  * If possible, determine which vopstats by fstype to use and
 613  * return a pointer to the caller.
 614  */
 615 vopstats_t *
 616 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp)
 617 {
 618         int             fstype = 0;     /* Index into vfssw[] */
 619         vopstats_t      *vsp = NULL;
 620 
 621         if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 ||
 622             !vopstats_enabled)
 623                 return (NULL);
 624         /*
 625          * Set up the fstype.  We go to so much trouble because all versions
 626          * of NFS use the same fstype in their vfs even though they have
 627          * distinct entries in the vfssw[] table.
 628          * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry.
 629          */
 630         if (vswp) {
 631                 fstype = vswp - vfssw;  /* Gets us the index */
 632         } else {
 633                 fstype = vfsp->vfs_fstype;
 634         }
 635 
 636         /*
 637          * Point to the per-fstype vopstats. The only valid values are
 638          * non-zero positive values less than the number of vfssw[] table
 639          * entries.
 640          */
 641         if (fstype > 0 && fstype < nfstype) {
 642                 vsp = vopstats_fstype[fstype];
 643         }
 644 
 645         return (vsp);
 646 }
 647 
 648 /*
 649  * Generate a kstat name, create the kstat structure, and allocate a
 650  * vsk_anchor_t to hold it together.  Return the pointer to the vsk_anchor_t
 651  * to the caller.  This must only be called from a mount.
 652  */
 653 vsk_anchor_t *
 654 get_vskstat_anchor(vfs_t *vfsp)
 655 {
 656         char            kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */
 657         statvfs64_t     statvfsbuf;             /* Needed to find f_fsid */
 658         vsk_anchor_t    *vskp = NULL;           /* vfs <--> kstat anchor */
 659         kstat_t         *ksp;                   /* Ptr to new kstat */
 660         avl_index_t     where;                  /* Location in the AVL tree */
 661 
 662         if (vfsp == NULL || vfsp->vfs_implp == NULL ||
 663             (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
 664                 return (NULL);
 665 
 666         /* Need to get the fsid to build a kstat name */
 667         if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) {
 668                 /* Create a name for our kstats based on fsid */
 669                 (void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx",
 670                     VOPSTATS_STR, statvfsbuf.f_fsid);
 671 
 672                 /* Allocate and initialize the vsk_anchor_t */
 673                 vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP);
 674                 bzero(vskp, sizeof (*vskp));
 675                 vskp->vsk_fsid = statvfsbuf.f_fsid;
 676 
 677                 mutex_enter(&vskstat_tree_lock);
 678                 if (avl_find(&vskstat_tree, vskp, &where) == NULL) {
 679                         avl_insert(&vskstat_tree, vskp, where);
 680                         mutex_exit(&vskstat_tree_lock);
 681 
 682                         /*
 683                          * Now that we've got the anchor in the AVL
 684                          * tree, we can create the kstat.
 685                          */
 686                         ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats);
 687                         if (ksp) {
 688                                 vskp->vsk_ksp = ksp;
 689                         }
 690                 } else {
 691                         /* Oops, found one! Release memory and lock. */
 692                         mutex_exit(&vskstat_tree_lock);
 693                         kmem_cache_free(vsk_anchor_cache, vskp);
 694                         vskp = NULL;
 695                 }
 696         }
 697         return (vskp);
 698 }
 699 
 700 /*
 701  * We're in the process of tearing down the vfs and need to cleanup
 702  * the data structures associated with the vopstats. Must only be called
 703  * from dounmount().
 704  */
 705 void
 706 teardown_vopstats(vfs_t *vfsp)
 707 {
 708         vsk_anchor_t    *vskap;
 709         avl_index_t     where;
 710 
 711         if (vfsp == NULL || vfsp->vfs_implp == NULL ||
 712             (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
 713                 return;
 714 
 715         /* This is a safe check since VFS_STATS must be set (see above) */
 716         if ((vskap = vfsp->vfs_vskap) == NULL)
 717                 return;
 718 
 719         /* Whack the pointer right away */
 720         vfsp->vfs_vskap = NULL;
 721 
 722         /* Lock the tree, remove the node, and delete the kstat */
 723         mutex_enter(&vskstat_tree_lock);
 724         if (avl_find(&vskstat_tree, vskap, &where)) {
 725                 avl_remove(&vskstat_tree, vskap);
 726         }
 727 
 728         if (vskap->vsk_ksp) {
 729                 kstat_delete(vskap->vsk_ksp);
 730         }
 731         mutex_exit(&vskstat_tree_lock);
 732 
 733         kmem_cache_free(vsk_anchor_cache, vskap);
 734 }
 735 
 736 /*
 737  * Read or write a vnode.  Called from kernel code.
 738  */
 739 int
 740 vn_rdwr(
 741         enum uio_rw rw,
 742         struct vnode *vp,
 743         caddr_t base,
 744         ssize_t len,
 745         offset_t offset,
 746         enum uio_seg seg,
 747         int ioflag,
 748         rlim64_t ulimit,        /* meaningful only if rw is UIO_WRITE */
 749         cred_t *cr,
 750         ssize_t *residp)
 751 {
 752         struct uio uio;
 753         struct iovec iov;
 754         int error;
 755         int in_crit = 0;
 756 
 757         if (rw == UIO_WRITE && ISROFILE(vp))
 758                 return (EROFS);
 759 
 760         if (len < 0)
 761                 return (EIO);
 762 
 763         VOPXID_MAP_CR(vp, cr);
 764 
 765         iov.iov_base = base;
 766         iov.iov_len = len;
 767         uio.uio_iov = &iov;
 768         uio.uio_iovcnt = 1;
 769         uio.uio_loffset = offset;
 770         uio.uio_segflg = (short)seg;
 771         uio.uio_resid = len;
 772         uio.uio_llimit = ulimit;
 773 
 774         /*
 775          * We have to enter the critical region before calling VOP_RWLOCK
 776          * to avoid a deadlock with ufs.
 777          */
 778         if (nbl_need_check(vp)) {
 779                 int svmand;
 780 
 781                 nbl_start_crit(vp, RW_READER);
 782                 in_crit = 1;
 783                 error = nbl_svmand(vp, cr, &svmand);
 784                 if (error != 0)
 785                         goto done;
 786                 if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ,
 787                     uio.uio_offset, uio.uio_resid, svmand, NULL)) {
 788                         error = EACCES;
 789                         goto done;
 790                 }
 791         }
 792 
 793         (void) VOP_RWLOCK(vp,
 794             rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
 795         if (rw == UIO_WRITE) {
 796                 uio.uio_fmode = FWRITE;
 797                 uio.uio_extflg = UIO_COPY_DEFAULT;
 798                 error = VOP_WRITE(vp, &uio, ioflag, cr, NULL);
 799         } else {
 800                 uio.uio_fmode = FREAD;
 801                 uio.uio_extflg = UIO_COPY_CACHED;
 802                 error = VOP_READ(vp, &uio, ioflag, cr, NULL);
 803         }
 804         VOP_RWUNLOCK(vp,
 805             rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
 806         if (residp)
 807                 *residp = uio.uio_resid;
 808         else if (uio.uio_resid)
 809                 error = EIO;
 810 
 811 done:
 812         if (in_crit)
 813                 nbl_end_crit(vp);
 814         return (error);
 815 }
 816 
 817 /*
 818  * Release a vnode.  Call VOP_INACTIVE on last reference or
 819  * decrement reference count.
 820  *
 821  * To avoid race conditions, the v_count is left at 1 for
 822  * the call to VOP_INACTIVE. This prevents another thread
 823  * from reclaiming and releasing the vnode *before* the
 824  * VOP_INACTIVE routine has a chance to destroy the vnode.
 825  * We can't have more than 1 thread calling VOP_INACTIVE
 826  * on a vnode.
 827  */
 828 void
 829 vn_rele(vnode_t *vp)
 830 {
 831         VERIFY(vp->v_count > 0);
 832         mutex_enter(&vp->v_lock);
 833         if (vp->v_count == 1) {
 834                 mutex_exit(&vp->v_lock);
 835                 VOP_INACTIVE(vp, CRED(), NULL);
 836                 return;
 837         }
 838         vp->v_count--;
 839         mutex_exit(&vp->v_lock);
 840 }
 841 
 842 /*
 843  * Release a vnode referenced by the DNLC. Multiple DNLC references are treated
 844  * as a single reference, so v_count is not decremented until the last DNLC hold
 845  * is released. This makes it possible to distinguish vnodes that are referenced
 846  * only by the DNLC.
 847  */
 848 void
 849 vn_rele_dnlc(vnode_t *vp)
 850 {
 851         VERIFY((vp->v_count > 0) && (vp->v_count_dnlc > 0));
 852         mutex_enter(&vp->v_lock);
 853         if (--vp->v_count_dnlc == 0) {
 854                 if (vp->v_count == 1) {
 855                         mutex_exit(&vp->v_lock);
 856                         VOP_INACTIVE(vp, CRED(), NULL);
 857                         return;
 858                 }
 859                 vp->v_count--;
 860         }
 861         mutex_exit(&vp->v_lock);
 862 }
 863 
 864 /*
 865  * Like vn_rele() except that it clears v_stream under v_lock.
 866  * This is used by sockfs when it dismantels the association between
 867  * the sockfs node and the vnode in the underlaying file system.
 868  * v_lock has to be held to prevent a thread coming through the lookupname
 869  * path from accessing a stream head that is going away.
 870  */
 871 void
 872 vn_rele_stream(vnode_t *vp)
 873 {
 874         VERIFY(vp->v_count > 0);
 875         mutex_enter(&vp->v_lock);
 876         vp->v_stream = NULL;
 877         if (vp->v_count == 1) {
 878                 mutex_exit(&vp->v_lock);
 879                 VOP_INACTIVE(vp, CRED(), NULL);
 880                 return;
 881         }
 882         vp->v_count--;
 883         mutex_exit(&vp->v_lock);
 884 }
 885 
 886 static void
 887 vn_rele_inactive(vnode_t *vp)
 888 {
 889         VOP_INACTIVE(vp, CRED(), NULL);
 890 }
 891 
 892 /*
 893  * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it
 894  * asynchronously using a taskq. This can avoid deadlocks caused by re-entering
 895  * the file system as a result of releasing the vnode. Note, file systems
 896  * already have to handle the race where the vnode is incremented before the
 897  * inactive routine is called and does its locking.
 898  *
 899  * Warning: Excessive use of this routine can lead to performance problems.
 900  * This is because taskqs throttle back allocation if too many are created.
 901  */
 902 void
 903 vn_rele_async(vnode_t *vp, taskq_t *taskq)
 904 {
 905         VERIFY(vp->v_count > 0);
 906         mutex_enter(&vp->v_lock);
 907         if (vp->v_count == 1) {
 908                 mutex_exit(&vp->v_lock);
 909                 VERIFY(taskq_dispatch(taskq, (task_func_t *)vn_rele_inactive,
 910                     vp, TQ_SLEEP) != NULL);
 911                 return;
 912         }
 913         vp->v_count--;
 914         mutex_exit(&vp->v_lock);
 915 }
 916 
 917 int
 918 vn_open(
 919         char *pnamep,
 920         enum uio_seg seg,
 921         int filemode,
 922         int createmode,
 923         struct vnode **vpp,
 924         enum create crwhy,
 925         mode_t umask)
 926 {
 927         return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy,
 928             umask, NULL, -1));
 929 }
 930 
 931 
 932 /*
 933  * Open/create a vnode.
 934  * This may be callable by the kernel, the only known use
 935  * of user context being that the current user credentials
 936  * are used for permissions.  crwhy is defined iff filemode & FCREAT.
 937  */
 938 int
 939 vn_openat(
 940         char *pnamep,
 941         enum uio_seg seg,
 942         int filemode,
 943         int createmode,
 944         struct vnode **vpp,
 945         enum create crwhy,
 946         mode_t umask,
 947         struct vnode *startvp,
 948         int fd)
 949 {
 950         struct vnode *vp;
 951         int mode;
 952         int accessflags;
 953         int error;
 954         int in_crit = 0;
 955         int open_done = 0;
 956         int shrlock_done = 0;
 957         struct vattr vattr;
 958         enum symfollow follow;
 959         int estale_retry = 0;
 960         struct shrlock shr;
 961         struct shr_locowner shr_own;
 962 
 963         mode = 0;
 964         accessflags = 0;
 965         if (filemode & FREAD)
 966                 mode |= VREAD;
 967         if (filemode & (FWRITE|FTRUNC))
 968                 mode |= VWRITE;
 969         if (filemode & (FSEARCH|FEXEC|FXATTRDIROPEN))
 970                 mode |= VEXEC;
 971 
 972         /* symlink interpretation */
 973         if (filemode & FNOFOLLOW)
 974                 follow = NO_FOLLOW;
 975         else
 976                 follow = FOLLOW;
 977 
 978         if (filemode & FAPPEND)
 979                 accessflags |= V_APPEND;
 980 
 981 top:
 982         if (filemode & FCREAT) {
 983                 enum vcexcl excl;
 984 
 985                 /*
 986                  * Wish to create a file.
 987                  */
 988                 vattr.va_type = VREG;
 989                 vattr.va_mode = createmode;
 990                 vattr.va_mask = AT_TYPE|AT_MODE;
 991                 if (filemode & FTRUNC) {
 992                         vattr.va_size = 0;
 993                         vattr.va_mask |= AT_SIZE;
 994                 }
 995                 if (filemode & FEXCL)
 996                         excl = EXCL;
 997                 else
 998                         excl = NONEXCL;
 999 
1000                 if (error =
1001                     vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy,
1002                     (filemode & ~(FTRUNC|FEXCL)), umask, startvp))
1003                         return (error);
1004         } else {
1005                 /*
1006                  * Wish to open a file.  Just look it up.
1007                  */
1008                 if (error = lookupnameat(pnamep, seg, follow,
1009                     NULLVPP, &vp, startvp)) {
1010                         if ((error == ESTALE) &&
1011                             fs_need_estale_retry(estale_retry++))
1012                                 goto top;
1013                         return (error);
1014                 }
1015 
1016                 /*
1017                  * Get the attributes to check whether file is large.
1018                  * We do this only if the FOFFMAX flag is not set and
1019                  * only for regular files.
1020                  */
1021 
1022                 if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) {
1023                         vattr.va_mask = AT_SIZE;
1024                         if ((error = VOP_GETATTR(vp, &vattr, 0,
1025                             CRED(), NULL))) {
1026                                 goto out;
1027                         }
1028                         if (vattr.va_size > (u_offset_t)MAXOFF32_T) {
1029                                 /*
1030                                  * Large File API - regular open fails
1031                                  * if FOFFMAX flag is set in file mode
1032                                  */
1033                                 error = EOVERFLOW;
1034                                 goto out;
1035                         }
1036                 }
1037                 /*
1038                  * Can't write directories, active texts, or
1039                  * read-only filesystems.  Can't truncate files
1040                  * on which mandatory locking is in effect.
1041                  */
1042                 if (filemode & (FWRITE|FTRUNC)) {
1043                         /*
1044                          * Allow writable directory if VDIROPEN flag is set.
1045                          */
1046                         if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) {
1047                                 error = EISDIR;
1048                                 goto out;
1049                         }
1050                         if (ISROFILE(vp)) {
1051                                 error = EROFS;
1052                                 goto out;
1053                         }
1054                         /*
1055                          * Can't truncate files on which
1056                          * sysv mandatory locking is in effect.
1057                          */
1058                         if (filemode & FTRUNC) {
1059                                 vnode_t *rvp;
1060 
1061                                 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1062                                         rvp = vp;
1063                                 if (rvp->v_filocks != NULL) {
1064                                         vattr.va_mask = AT_MODE;
1065                                         if ((error = VOP_GETATTR(vp,
1066                                             &vattr, 0, CRED(), NULL)) == 0 &&
1067                                             MANDLOCK(vp, vattr.va_mode))
1068                                                 error = EAGAIN;
1069                                 }
1070                         }
1071                         if (error)
1072                                 goto out;
1073                 }
1074                 /*
1075                  * Check permissions.
1076                  */
1077                 if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL))
1078                         goto out;
1079                 /*
1080                  * Require FSEARCH to return a directory.
1081                  * Require FEXEC to return a regular file.
1082                  */
1083                 if ((filemode & FSEARCH) && vp->v_type != VDIR) {
1084                         error = ENOTDIR;
1085                         goto out;
1086                 }
1087                 if ((filemode & FEXEC) && vp->v_type != VREG) {
1088                         error = ENOEXEC;        /* XXX: error code? */
1089                         goto out;
1090                 }
1091         }
1092 
1093         /*
1094          * Do remaining checks for FNOFOLLOW and FNOLINKS.
1095          */
1096         if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) {
1097                 error = ELOOP;
1098                 goto out;
1099         }
1100         if (filemode & FNOLINKS) {
1101                 vattr.va_mask = AT_NLINK;
1102                 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) {
1103                         goto out;
1104                 }
1105                 if (vattr.va_nlink != 1) {
1106                         error = EMLINK;
1107                         goto out;
1108                 }
1109         }
1110 
1111         /*
1112          * Opening a socket corresponding to the AF_UNIX pathname
1113          * in the filesystem name space is not supported.
1114          * However, VSOCK nodes in namefs are supported in order
1115          * to make fattach work for sockets.
1116          *
1117          * XXX This uses VOP_REALVP to distinguish between
1118          * an unopened namefs node (where VOP_REALVP returns a
1119          * different VSOCK vnode) and a VSOCK created by vn_create
1120          * in some file system (where VOP_REALVP would never return
1121          * a different vnode).
1122          */
1123         if (vp->v_type == VSOCK) {
1124                 struct vnode *nvp;
1125 
1126                 error = VOP_REALVP(vp, &nvp, NULL);
1127                 if (error != 0 || nvp == NULL || nvp == vp ||
1128                     nvp->v_type != VSOCK) {
1129                         error = EOPNOTSUPP;
1130                         goto out;
1131                 }
1132         }
1133 
1134         if ((vp->v_type == VREG) && nbl_need_check(vp)) {
1135                 /* get share reservation */
1136                 shr.s_access = 0;
1137                 if (filemode & FWRITE)
1138                         shr.s_access |= F_WRACC;
1139                 if (filemode & FREAD)
1140                         shr.s_access |= F_RDACC;
1141                 shr.s_deny = 0;
1142                 shr.s_sysid = 0;
1143                 shr.s_pid = ttoproc(curthread)->p_pid;
1144                 shr_own.sl_pid = shr.s_pid;
1145                 shr_own.sl_id = fd;
1146                 shr.s_own_len = sizeof (shr_own);
1147                 shr.s_owner = (caddr_t)&shr_own;
1148                 error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(),
1149                     NULL);
1150                 if (error)
1151                         goto out;
1152                 shrlock_done = 1;
1153 
1154                 /* nbmand conflict check if truncating file */
1155                 if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1156                         nbl_start_crit(vp, RW_READER);
1157                         in_crit = 1;
1158 
1159                         vattr.va_mask = AT_SIZE;
1160                         if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
1161                                 goto out;
1162                         if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0,
1163                             NULL)) {
1164                                 error = EACCES;
1165                                 goto out;
1166                         }
1167                 }
1168         }
1169 
1170         /*
1171          * Do opening protocol.
1172          */
1173         error = VOP_OPEN(&vp, filemode, CRED(), NULL);
1174         if (error)
1175                 goto out;
1176         open_done = 1;
1177 
1178         /*
1179          * Truncate if required.
1180          */
1181         if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1182                 vattr.va_size = 0;
1183                 vattr.va_mask = AT_SIZE;
1184                 if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
1185                         goto out;
1186         }
1187 out:
1188         ASSERT(vp->v_count > 0);
1189 
1190         if (in_crit) {
1191                 nbl_end_crit(vp);
1192                 in_crit = 0;
1193         }
1194         if (error) {
1195                 if (open_done) {
1196                         (void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(),
1197                             NULL);
1198                         open_done = 0;
1199                         shrlock_done = 0;
1200                 }
1201                 if (shrlock_done) {
1202                         (void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(),
1203                             NULL);
1204                         shrlock_done = 0;
1205                 }
1206 
1207                 /*
1208                  * The following clause was added to handle a problem
1209                  * with NFS consistency.  It is possible that a lookup
1210                  * of the file to be opened succeeded, but the file
1211                  * itself doesn't actually exist on the server.  This
1212                  * is chiefly due to the DNLC containing an entry for
1213                  * the file which has been removed on the server.  In
1214                  * this case, we just start over.  If there was some
1215                  * other cause for the ESTALE error, then the lookup
1216                  * of the file will fail and the error will be returned
1217                  * above instead of looping around from here.
1218                  */
1219                 VN_RELE(vp);
1220                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1221                         goto top;
1222         } else
1223                 *vpp = vp;
1224         return (error);
1225 }
1226 
1227 /*
1228  * The following two accessor functions are for the NFSv4 server.  Since there
1229  * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the
1230  * vnode open counts correct when a client "upgrades" an open or does an
1231  * open_downgrade.  In NFS, an upgrade or downgrade can not only change the
1232  * open mode (add or subtract read or write), but also change the share/deny
1233  * modes.  However, share reservations are not integrated with OPEN, yet, so
1234  * we need to handle each separately.  These functions are cleaner than having
1235  * the NFS server manipulate the counts directly, however, nobody else should
1236  * use these functions.
1237  */
1238 void
1239 vn_open_upgrade(
1240         vnode_t *vp,
1241         int filemode)
1242 {
1243         ASSERT(vp->v_type == VREG);
1244 
1245         if (filemode & FREAD)
1246                 atomic_add_32(&(vp->v_rdcnt), 1);
1247         if (filemode & FWRITE)
1248                 atomic_add_32(&(vp->v_wrcnt), 1);
1249 
1250 }
1251 
1252 void
1253 vn_open_downgrade(
1254         vnode_t *vp,
1255         int filemode)
1256 {
1257         ASSERT(vp->v_type == VREG);
1258 
1259         if (filemode & FREAD) {
1260                 ASSERT(vp->v_rdcnt > 0);
1261                 atomic_add_32(&(vp->v_rdcnt), -1);
1262         }
1263         if (filemode & FWRITE) {
1264                 ASSERT(vp->v_wrcnt > 0);
1265                 atomic_add_32(&(vp->v_wrcnt), -1);
1266         }
1267 
1268 }
1269 
1270 int
1271 vn_create(
1272         char *pnamep,
1273         enum uio_seg seg,
1274         struct vattr *vap,
1275         enum vcexcl excl,
1276         int mode,
1277         struct vnode **vpp,
1278         enum create why,
1279         int flag,
1280         mode_t umask)
1281 {
1282         return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag,
1283             umask, NULL));
1284 }
1285 
1286 /*
1287  * Create a vnode (makenode).
1288  */
1289 int
1290 vn_createat(
1291         char *pnamep,
1292         enum uio_seg seg,
1293         struct vattr *vap,
1294         enum vcexcl excl,
1295         int mode,
1296         struct vnode **vpp,
1297         enum create why,
1298         int flag,
1299         mode_t umask,
1300         struct vnode *startvp)
1301 {
1302         struct vnode *dvp;      /* ptr to parent dir vnode */
1303         struct vnode *vp = NULL;
1304         struct pathname pn;
1305         int error;
1306         int in_crit = 0;
1307         struct vattr vattr;
1308         enum symfollow follow;
1309         int estale_retry = 0;
1310         uint32_t auditing = AU_AUDITING();
1311 
1312         ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
1313 
1314         /* symlink interpretation */
1315         if ((flag & FNOFOLLOW) || excl == EXCL)
1316                 follow = NO_FOLLOW;
1317         else
1318                 follow = FOLLOW;
1319         flag &= ~(FNOFOLLOW|FNOLINKS);
1320 
1321 top:
1322         /*
1323          * Lookup directory.
1324          * If new object is a file, call lower level to create it.
1325          * Note that it is up to the lower level to enforce exclusive
1326          * creation, if the file is already there.
1327          * This allows the lower level to do whatever
1328          * locking or protocol that is needed to prevent races.
1329          * If the new object is directory call lower level to make
1330          * the new directory, with "." and "..".
1331          */
1332         if (error = pn_get(pnamep, seg, &pn))
1333                 return (error);
1334         if (auditing)
1335                 audit_vncreate_start();
1336         dvp = NULL;
1337         *vpp = NULL;
1338         /*
1339          * lookup will find the parent directory for the vnode.
1340          * When it is done the pn holds the name of the entry
1341          * in the directory.
1342          * If this is a non-exclusive create we also find the node itself.
1343          */
1344         error = lookuppnat(&pn, NULL, follow, &dvp,
1345             (excl == EXCL) ? NULLVPP : vpp, startvp);
1346         if (error) {
1347                 pn_free(&pn);
1348                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1349                         goto top;
1350                 if (why == CRMKDIR && error == EINVAL)
1351                         error = EEXIST;         /* SVID */
1352                 return (error);
1353         }
1354 
1355         if (why != CRMKNOD)
1356                 vap->va_mode &= ~VSVTX;
1357 
1358         /*
1359          * If default ACLs are defined for the directory don't apply the
1360          * umask if umask is passed.
1361          */
1362 
1363         if (umask) {
1364 
1365                 vsecattr_t vsec;
1366 
1367                 vsec.vsa_aclcnt = 0;
1368                 vsec.vsa_aclentp = NULL;
1369                 vsec.vsa_dfaclcnt = 0;
1370                 vsec.vsa_dfaclentp = NULL;
1371                 vsec.vsa_mask = VSA_DFACLCNT;
1372                 error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL);
1373                 /*
1374                  * If error is ENOSYS then treat it as no error
1375                  * Don't want to force all file systems to support
1376                  * aclent_t style of ACL's.
1377                  */
1378                 if (error == ENOSYS)
1379                         error = 0;
1380                 if (error) {
1381                         if (*vpp != NULL)
1382                                 VN_RELE(*vpp);
1383                         goto out;
1384                 } else {
1385                         /*
1386                          * Apply the umask if no default ACLs.
1387                          */
1388                         if (vsec.vsa_dfaclcnt == 0)
1389                                 vap->va_mode &= ~umask;
1390 
1391                         /*
1392                          * VOP_GETSECATTR() may have allocated memory for
1393                          * ACLs we didn't request, so double-check and
1394                          * free it if necessary.
1395                          */
1396                         if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL)
1397                                 kmem_free((caddr_t)vsec.vsa_aclentp,
1398                                     vsec.vsa_aclcnt * sizeof (aclent_t));
1399                         if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL)
1400                                 kmem_free((caddr_t)vsec.vsa_dfaclentp,
1401                                     vsec.vsa_dfaclcnt * sizeof (aclent_t));
1402                 }
1403         }
1404 
1405         /*
1406          * In general we want to generate EROFS if the file system is
1407          * readonly.  However, POSIX (IEEE Std. 1003.1) section 5.3.1
1408          * documents the open system call, and it says that O_CREAT has no
1409          * effect if the file already exists.  Bug 1119649 states
1410          * that open(path, O_CREAT, ...) fails when attempting to open an
1411          * existing file on a read only file system.  Thus, the first part
1412          * of the following if statement has 3 checks:
1413          *      if the file exists &&
1414          *              it is being open with write access &&
1415          *              the file system is read only
1416          *      then generate EROFS
1417          */
1418         if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) ||
1419             (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) {
1420                 if (*vpp)
1421                         VN_RELE(*vpp);
1422                 error = EROFS;
1423         } else if (excl == NONEXCL && *vpp != NULL) {
1424                 vnode_t *rvp;
1425 
1426                 /*
1427                  * File already exists.  If a mandatory lock has been
1428                  * applied, return error.
1429                  */
1430                 vp = *vpp;
1431                 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1432                         rvp = vp;
1433                 if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) {
1434                         nbl_start_crit(vp, RW_READER);
1435                         in_crit = 1;
1436                 }
1437                 if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) {
1438                         vattr.va_mask = AT_MODE|AT_SIZE;
1439                         if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) {
1440                                 goto out;
1441                         }
1442                         if (MANDLOCK(vp, vattr.va_mode)) {
1443                                 error = EAGAIN;
1444                                 goto out;
1445                         }
1446                         /*
1447                          * File cannot be truncated if non-blocking mandatory
1448                          * locks are currently on the file.
1449                          */
1450                         if ((vap->va_mask & AT_SIZE) && in_crit) {
1451                                 u_offset_t offset;
1452                                 ssize_t length;
1453 
1454                                 offset = vap->va_size > vattr.va_size ?
1455                                     vattr.va_size : vap->va_size;
1456                                 length = vap->va_size > vattr.va_size ?
1457                                     vap->va_size - vattr.va_size :
1458                                     vattr.va_size - vap->va_size;
1459                                 if (nbl_conflict(vp, NBL_WRITE, offset,
1460                                     length, 0, NULL)) {
1461                                         error = EACCES;
1462                                         goto out;
1463                                 }
1464                         }
1465                 }
1466 
1467                 /*
1468                  * If the file is the root of a VFS, we've crossed a
1469                  * mount point and the "containing" directory that we
1470                  * acquired above (dvp) is irrelevant because it's in
1471                  * a different file system.  We apply VOP_CREATE to the
1472                  * target itself instead of to the containing directory
1473                  * and supply a null path name to indicate (conventionally)
1474                  * the node itself as the "component" of interest.
1475                  *
1476                  * The intercession of the file system is necessary to
1477                  * ensure that the appropriate permission checks are
1478                  * done.
1479                  */
1480                 if (vp->v_flag & VROOT) {
1481                         ASSERT(why != CRMKDIR);
1482                         error = VOP_CREATE(vp, "", vap, excl, mode, vpp,
1483                             CRED(), flag, NULL, NULL);
1484                         /*
1485                          * If the create succeeded, it will have created
1486                          * a new reference to the vnode.  Give up the
1487                          * original reference.  The assertion should not
1488                          * get triggered because NBMAND locks only apply to
1489                          * VREG files.  And if in_crit is non-zero for some
1490                          * reason, detect that here, rather than when we
1491                          * deference a null vp.
1492                          */
1493                         ASSERT(in_crit == 0);
1494                         VN_RELE(vp);
1495                         vp = NULL;
1496                         goto out;
1497                 }
1498 
1499                 /*
1500                  * Large File API - non-large open (FOFFMAX flag not set)
1501                  * of regular file fails if the file size exceeds MAXOFF32_T.
1502                  */
1503                 if (why != CRMKDIR &&
1504                     !(flag & FOFFMAX) &&
1505                     (vp->v_type == VREG)) {
1506                         vattr.va_mask = AT_SIZE;
1507                         if ((error = VOP_GETATTR(vp, &vattr, 0,
1508                             CRED(), NULL))) {
1509                                 goto out;
1510                         }
1511                         if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) {
1512                                 error = EOVERFLOW;
1513                                 goto out;
1514                         }
1515                 }
1516         }
1517 
1518         if (error == 0) {
1519                 /*
1520                  * Call mkdir() if specified, otherwise create().
1521                  */
1522                 int must_be_dir = pn_fixslash(&pn); /* trailing '/'? */
1523 
1524                 if (why == CRMKDIR)
1525                         /*
1526                          * N.B., if vn_createat() ever requests
1527                          * case-insensitive behavior then it will need
1528                          * to be passed to VOP_MKDIR().  VOP_CREATE()
1529                          * will already get it via "flag"
1530                          */
1531                         error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(),
1532                             NULL, 0, NULL);
1533                 else if (!must_be_dir)
1534                         error = VOP_CREATE(dvp, pn.pn_path, vap,
1535                             excl, mode, vpp, CRED(), flag, NULL, NULL);
1536                 else
1537                         error = ENOTDIR;
1538         }
1539 
1540 out:
1541 
1542         if (auditing)
1543                 audit_vncreate_finish(*vpp, error);
1544         if (in_crit) {
1545                 nbl_end_crit(vp);
1546                 in_crit = 0;
1547         }
1548         if (vp != NULL) {
1549                 VN_RELE(vp);
1550                 vp = NULL;
1551         }
1552         pn_free(&pn);
1553         VN_RELE(dvp);
1554         /*
1555          * The following clause was added to handle a problem
1556          * with NFS consistency.  It is possible that a lookup
1557          * of the file to be created succeeded, but the file
1558          * itself doesn't actually exist on the server.  This
1559          * is chiefly due to the DNLC containing an entry for
1560          * the file which has been removed on the server.  In
1561          * this case, we just start over.  If there was some
1562          * other cause for the ESTALE error, then the lookup
1563          * of the file will fail and the error will be returned
1564          * above instead of looping around from here.
1565          */
1566         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1567                 goto top;
1568         return (error);
1569 }
1570 
1571 int
1572 vn_link(char *from, char *to, enum uio_seg seg)
1573 {
1574         return (vn_linkat(NULL, from, NO_FOLLOW, NULL, to, seg));
1575 }
1576 
1577 int
1578 vn_linkat(vnode_t *fstartvp, char *from, enum symfollow follow,
1579     vnode_t *tstartvp, char *to, enum uio_seg seg)
1580 {
1581         struct vnode *fvp;              /* from vnode ptr */
1582         struct vnode *tdvp;             /* to directory vnode ptr */
1583         struct pathname pn;
1584         int error;
1585         struct vattr vattr;
1586         dev_t fsid;
1587         int estale_retry = 0;
1588         uint32_t auditing = AU_AUDITING();
1589 
1590 top:
1591         fvp = tdvp = NULL;
1592         if (error = pn_get(to, seg, &pn))
1593                 return (error);
1594         if (auditing && fstartvp != NULL)
1595                 audit_setfsat_path(1);
1596         if (error = lookupnameat(from, seg, follow, NULLVPP, &fvp, fstartvp))
1597                 goto out;
1598         if (auditing && tstartvp != NULL)
1599                 audit_setfsat_path(3);
1600         if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP, tstartvp))
1601                 goto out;
1602         /*
1603          * Make sure both source vnode and target directory vnode are
1604          * in the same vfs and that it is writeable.
1605          */
1606         vattr.va_mask = AT_FSID;
1607         if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL))
1608                 goto out;
1609         fsid = vattr.va_fsid;
1610         vattr.va_mask = AT_FSID;
1611         if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL))
1612                 goto out;
1613         if (fsid != vattr.va_fsid) {
1614                 error = EXDEV;
1615                 goto out;
1616         }
1617         if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) {
1618                 error = EROFS;
1619                 goto out;
1620         }
1621         /*
1622          * Do the link.
1623          */
1624         (void) pn_fixslash(&pn);
1625         error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0);
1626 out:
1627         pn_free(&pn);
1628         if (fvp)
1629                 VN_RELE(fvp);
1630         if (tdvp)
1631                 VN_RELE(tdvp);
1632         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1633                 goto top;
1634         return (error);
1635 }
1636 
1637 int
1638 vn_rename(char *from, char *to, enum uio_seg seg)
1639 {
1640         return (vn_renameat(NULL, from, NULL, to, seg));
1641 }
1642 
1643 int
1644 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp,
1645                 char *tname, enum uio_seg seg)
1646 {
1647         int error;
1648         struct vattr vattr;
1649         struct pathname fpn;            /* from pathname */
1650         struct pathname tpn;            /* to pathname */
1651         dev_t fsid;
1652         int in_crit_src, in_crit_targ;
1653         vnode_t *fromvp, *fvp;
1654         vnode_t *tovp, *targvp;
1655         int estale_retry = 0;
1656         uint32_t auditing = AU_AUDITING();
1657 
1658 top:
1659         fvp = fromvp = tovp = targvp = NULL;
1660         in_crit_src = in_crit_targ = 0;
1661         /*
1662          * Get to and from pathnames.
1663          */
1664         if (error = pn_get(fname, seg, &fpn))
1665                 return (error);
1666         if (error = pn_get(tname, seg, &tpn)) {
1667                 pn_free(&fpn);
1668                 return (error);
1669         }
1670 
1671         /*
1672          * First we need to resolve the correct directories
1673          * The passed in directories may only be a starting point,
1674          * but we need the real directories the file(s) live in.
1675          * For example the fname may be something like usr/lib/sparc
1676          * and we were passed in the / directory, but we need to
1677          * use the lib directory for the rename.
1678          */
1679 
1680         if (auditing && fdvp != NULL)
1681                 audit_setfsat_path(1);
1682         /*
1683          * Lookup to and from directories.
1684          */
1685         if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) {
1686                 goto out;
1687         }
1688 
1689         /*
1690          * Make sure there is an entry.
1691          */
1692         if (fvp == NULL) {
1693                 error = ENOENT;
1694                 goto out;
1695         }
1696 
1697         if (auditing && tdvp != NULL)
1698                 audit_setfsat_path(3);
1699         if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) {
1700                 goto out;
1701         }
1702 
1703         /*
1704          * Make sure both the from vnode directory and the to directory
1705          * are in the same vfs and the to directory is writable.
1706          * We check fsid's, not vfs pointers, so loopback fs works.
1707          */
1708         if (fromvp != tovp) {
1709                 vattr.va_mask = AT_FSID;
1710                 if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL))
1711                         goto out;
1712                 fsid = vattr.va_fsid;
1713                 vattr.va_mask = AT_FSID;
1714                 if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL))
1715                         goto out;
1716                 if (fsid != vattr.va_fsid) {
1717                         error = EXDEV;
1718                         goto out;
1719                 }
1720         }
1721 
1722         if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) {
1723                 error = EROFS;
1724                 goto out;
1725         }
1726 
1727         if (targvp && (fvp != targvp)) {
1728                 nbl_start_crit(targvp, RW_READER);
1729                 in_crit_targ = 1;
1730                 if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
1731                         error = EACCES;
1732                         goto out;
1733                 }
1734         }
1735 
1736         if (nbl_need_check(fvp)) {
1737                 nbl_start_crit(fvp, RW_READER);
1738                 in_crit_src = 1;
1739                 if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) {
1740                         error = EACCES;
1741                         goto out;
1742                 }
1743         }
1744 
1745         /*
1746          * Do the rename.
1747          */
1748         (void) pn_fixslash(&tpn);
1749         error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(),
1750             NULL, 0);
1751 
1752 out:
1753         pn_free(&fpn);
1754         pn_free(&tpn);
1755         if (in_crit_src)
1756                 nbl_end_crit(fvp);
1757         if (in_crit_targ)
1758                 nbl_end_crit(targvp);
1759         if (fromvp)
1760                 VN_RELE(fromvp);
1761         if (tovp)
1762                 VN_RELE(tovp);
1763         if (targvp)
1764                 VN_RELE(targvp);
1765         if (fvp)
1766                 VN_RELE(fvp);
1767         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1768                 goto top;
1769         return (error);
1770 }
1771 
1772 /*
1773  * Remove a file or directory.
1774  */
1775 int
1776 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag)
1777 {
1778         return (vn_removeat(NULL, fnamep, seg, dirflag));
1779 }
1780 
1781 int
1782 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag)
1783 {
1784         struct vnode *vp;               /* entry vnode */
1785         struct vnode *dvp;              /* ptr to parent dir vnode */
1786         struct vnode *coveredvp;
1787         struct pathname pn;             /* name of entry */
1788         enum vtype vtype;
1789         int error;
1790         struct vfs *vfsp;
1791         struct vfs *dvfsp;      /* ptr to parent dir vfs */
1792         int in_crit = 0;
1793         int estale_retry = 0;
1794 
1795 top:
1796         if (error = pn_get(fnamep, seg, &pn))
1797                 return (error);
1798         dvp = vp = NULL;
1799         if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) {
1800                 pn_free(&pn);
1801                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1802                         goto top;
1803                 return (error);
1804         }
1805 
1806         /*
1807          * Make sure there is an entry.
1808          */
1809         if (vp == NULL) {
1810                 error = ENOENT;
1811                 goto out;
1812         }
1813 
1814         vfsp = vp->v_vfsp;
1815         dvfsp = dvp->v_vfsp;
1816 
1817         /*
1818          * If the named file is the root of a mounted filesystem, fail,
1819          * unless it's marked unlinkable.  In that case, unmount the
1820          * filesystem and proceed to unlink the covered vnode.  (If the
1821          * covered vnode is a directory, use rmdir instead of unlink,
1822          * to avoid file system corruption.)
1823          */
1824         if (vp->v_flag & VROOT) {
1825                 if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) {
1826                         error = EBUSY;
1827                         goto out;
1828                 }
1829 
1830                 /*
1831                  * Namefs specific code starts here.
1832                  */
1833 
1834                 if (dirflag == RMDIRECTORY) {
1835                         /*
1836                          * User called rmdir(2) on a file that has
1837                          * been namefs mounted on top of.  Since
1838                          * namefs doesn't allow directories to
1839                          * be mounted on other files we know
1840                          * vp is not of type VDIR so fail to operation.
1841                          */
1842                         error = ENOTDIR;
1843                         goto out;
1844                 }
1845 
1846                 /*
1847                  * If VROOT is still set after grabbing vp->v_lock,
1848                  * noone has finished nm_unmount so far and coveredvp
1849                  * is valid.
1850                  * If we manage to grab vn_vfswlock(coveredvp) before releasing
1851                  * vp->v_lock, any race window is eliminated.
1852                  */
1853 
1854                 mutex_enter(&vp->v_lock);
1855                 if ((vp->v_flag & VROOT) == 0) {
1856                         /* Someone beat us to the unmount */
1857                         mutex_exit(&vp->v_lock);
1858                         error = EBUSY;
1859                         goto out;
1860                 }
1861                 vfsp = vp->v_vfsp;
1862                 coveredvp = vfsp->vfs_vnodecovered;
1863                 ASSERT(coveredvp);
1864                 /*
1865                  * Note: Implementation of vn_vfswlock shows that ordering of
1866                  * v_lock / vn_vfswlock is not an issue here.
1867                  */
1868                 error = vn_vfswlock(coveredvp);
1869                 mutex_exit(&vp->v_lock);
1870 
1871                 if (error)
1872                         goto out;
1873 
1874                 VN_HOLD(coveredvp);
1875                 VN_RELE(vp);
1876                 error = dounmount(vfsp, 0, CRED());
1877 
1878                 /*
1879                  * Unmounted the namefs file system; now get
1880                  * the object it was mounted over.
1881                  */
1882                 vp = coveredvp;
1883                 /*
1884                  * If namefs was mounted over a directory, then
1885                  * we want to use rmdir() instead of unlink().
1886                  */
1887                 if (vp->v_type == VDIR)
1888                         dirflag = RMDIRECTORY;
1889 
1890                 if (error)
1891                         goto out;
1892         }
1893 
1894         /*
1895          * Make sure filesystem is writeable.
1896          * We check the parent directory's vfs in case this is an lofs vnode.
1897          */
1898         if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) {
1899                 error = EROFS;
1900                 goto out;
1901         }
1902 
1903         vtype = vp->v_type;
1904 
1905         /*
1906          * If there is the possibility of an nbmand share reservation, make
1907          * sure it's okay to remove the file.  Keep a reference to the
1908          * vnode, so that we can exit the nbl critical region after
1909          * calling VOP_REMOVE.
1910          * If there is no possibility of an nbmand share reservation,
1911          * release the vnode reference now.  Filesystems like NFS may
1912          * behave differently if there is an extra reference, so get rid of
1913          * this one.  Fortunately, we can't have nbmand mounts on NFS
1914          * filesystems.
1915          */
1916         if (nbl_need_check(vp)) {
1917                 nbl_start_crit(vp, RW_READER);
1918                 in_crit = 1;
1919                 if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
1920                         error = EACCES;
1921                         goto out;
1922                 }
1923         } else {
1924                 VN_RELE(vp);
1925                 vp = NULL;
1926         }
1927 
1928         if (dirflag == RMDIRECTORY) {
1929                 /*
1930                  * Caller is using rmdir(2), which can only be applied to
1931                  * directories.
1932                  */
1933                 if (vtype != VDIR) {
1934                         error = ENOTDIR;
1935                 } else {
1936                         vnode_t *cwd;
1937                         proc_t *pp = curproc;
1938 
1939                         mutex_enter(&pp->p_lock);
1940                         cwd = PTOU(pp)->u_cdir;
1941                         VN_HOLD(cwd);
1942                         mutex_exit(&pp->p_lock);
1943                         error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(),
1944                             NULL, 0);
1945                         VN_RELE(cwd);
1946                 }
1947         } else {
1948                 /*
1949                  * Unlink(2) can be applied to anything.
1950                  */
1951                 error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
1952         }
1953 
1954 out:
1955         pn_free(&pn);
1956         if (in_crit) {
1957                 nbl_end_crit(vp);
1958                 in_crit = 0;
1959         }
1960         if (vp != NULL)
1961                 VN_RELE(vp);
1962         if (dvp != NULL)
1963                 VN_RELE(dvp);
1964         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1965                 goto top;
1966         return (error);
1967 }
1968 
1969 /*
1970  * Utility function to compare equality of vnodes.
1971  * Compare the underlying real vnodes, if there are underlying vnodes.
1972  * This is a more thorough comparison than the VN_CMP() macro provides.
1973  */
1974 int
1975 vn_compare(vnode_t *vp1, vnode_t *vp2)
1976 {
1977         vnode_t *realvp;
1978 
1979         if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0)
1980                 vp1 = realvp;
1981         if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0)
1982                 vp2 = realvp;
1983         return (VN_CMP(vp1, vp2));
1984 }
1985 
1986 /*
1987  * The number of locks to hash into.  This value must be a power
1988  * of 2 minus 1 and should probably also be prime.
1989  */
1990 #define NUM_BUCKETS     1023
1991 
1992 struct  vn_vfslocks_bucket {
1993         kmutex_t vb_lock;
1994         vn_vfslocks_entry_t *vb_list;
1995         char pad[64 - sizeof (kmutex_t) - sizeof (void *)];
1996 };
1997 
1998 /*
1999  * Total number of buckets will be NUM_BUCKETS + 1 .
2000  */
2001 
2002 #pragma align   64(vn_vfslocks_buckets)
2003 static  struct vn_vfslocks_bucket       vn_vfslocks_buckets[NUM_BUCKETS + 1];
2004 
2005 #define VN_VFSLOCKS_SHIFT       9
2006 
2007 #define VN_VFSLOCKS_HASH(vfsvpptr)      \
2008         ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS)
2009 
2010 /*
2011  * vn_vfslocks_getlock() uses an HASH scheme to generate
2012  * rwstlock using vfs/vnode pointer passed to it.
2013  *
2014  * vn_vfslocks_rele() releases a reference in the
2015  * HASH table which allows the entry allocated by
2016  * vn_vfslocks_getlock() to be freed at a later
2017  * stage when the refcount drops to zero.
2018  */
2019 
2020 vn_vfslocks_entry_t *
2021 vn_vfslocks_getlock(void *vfsvpptr)
2022 {
2023         struct vn_vfslocks_bucket *bp;
2024         vn_vfslocks_entry_t *vep;
2025         vn_vfslocks_entry_t *tvep;
2026 
2027         ASSERT(vfsvpptr != NULL);
2028         bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)];
2029 
2030         mutex_enter(&bp->vb_lock);
2031         for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2032                 if (vep->ve_vpvfs == vfsvpptr) {
2033                         vep->ve_refcnt++;
2034                         mutex_exit(&bp->vb_lock);
2035                         return (vep);
2036                 }
2037         }
2038         mutex_exit(&bp->vb_lock);
2039         vep = kmem_alloc(sizeof (*vep), KM_SLEEP);
2040         rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL);
2041         vep->ve_vpvfs = (char *)vfsvpptr;
2042         vep->ve_refcnt = 1;
2043         mutex_enter(&bp->vb_lock);
2044         for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) {
2045                 if (tvep->ve_vpvfs == vfsvpptr) {
2046                         tvep->ve_refcnt++;
2047                         mutex_exit(&bp->vb_lock);
2048 
2049                         /*
2050                          * There is already an entry in the hash
2051                          * destroy what we just allocated.
2052                          */
2053                         rwst_destroy(&vep->ve_lock);
2054                         kmem_free(vep, sizeof (*vep));
2055                         return (tvep);
2056                 }
2057         }
2058         vep->ve_next = bp->vb_list;
2059         bp->vb_list = vep;
2060         mutex_exit(&bp->vb_lock);
2061         return (vep);
2062 }
2063 
2064 void
2065 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent)
2066 {
2067         struct vn_vfslocks_bucket *bp;
2068         vn_vfslocks_entry_t *vep;
2069         vn_vfslocks_entry_t *pvep;
2070 
2071         ASSERT(vepent != NULL);
2072         ASSERT(vepent->ve_vpvfs != NULL);
2073 
2074         bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)];
2075 
2076         mutex_enter(&bp->vb_lock);
2077         vepent->ve_refcnt--;
2078 
2079         if ((int32_t)vepent->ve_refcnt < 0)
2080                 cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative");
2081 
2082         if (vepent->ve_refcnt == 0) {
2083                 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2084                         if (vep->ve_vpvfs == vepent->ve_vpvfs) {
2085                                 if (bp->vb_list == vep)
2086                                         bp->vb_list = vep->ve_next;
2087                                 else {
2088                                         /* LINTED */
2089                                         pvep->ve_next = vep->ve_next;
2090                                 }
2091                                 mutex_exit(&bp->vb_lock);
2092                                 rwst_destroy(&vep->ve_lock);
2093                                 kmem_free(vep, sizeof (*vep));
2094                                 return;
2095                         }
2096                         pvep = vep;
2097                 }
2098                 cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found");
2099         }
2100         mutex_exit(&bp->vb_lock);
2101 }
2102 
2103 /*
2104  * vn_vfswlock_wait is used to implement a lock which is logically a writers
2105  * lock protecting the v_vfsmountedhere field.
2106  * vn_vfswlock_wait has been modified to be similar to vn_vfswlock,
2107  * except that it blocks to acquire the lock VVFSLOCK.
2108  *
2109  * traverse() and routines re-implementing part of traverse (e.g. autofs)
2110  * need to hold this lock. mount(), vn_rename(), vn_remove() and so on
2111  * need the non-blocking version of the writers lock i.e. vn_vfswlock
2112  */
2113 int
2114 vn_vfswlock_wait(vnode_t *vp)
2115 {
2116         int retval;
2117         vn_vfslocks_entry_t *vpvfsentry;
2118         ASSERT(vp != NULL);
2119 
2120         vpvfsentry = vn_vfslocks_getlock(vp);
2121         retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER);
2122 
2123         if (retval == EINTR) {
2124                 vn_vfslocks_rele(vpvfsentry);
2125                 return (EINTR);
2126         }
2127         return (retval);
2128 }
2129 
2130 int
2131 vn_vfsrlock_wait(vnode_t *vp)
2132 {
2133         int retval;
2134         vn_vfslocks_entry_t *vpvfsentry;
2135         ASSERT(vp != NULL);
2136 
2137         vpvfsentry = vn_vfslocks_getlock(vp);
2138         retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER);
2139 
2140         if (retval == EINTR) {
2141                 vn_vfslocks_rele(vpvfsentry);
2142                 return (EINTR);
2143         }
2144 
2145         return (retval);
2146 }
2147 
2148 
2149 /*
2150  * vn_vfswlock is used to implement a lock which is logically a writers lock
2151  * protecting the v_vfsmountedhere field.
2152  */
2153 int
2154 vn_vfswlock(vnode_t *vp)
2155 {
2156         vn_vfslocks_entry_t *vpvfsentry;
2157 
2158         /*
2159          * If vp is NULL then somebody is trying to lock the covered vnode
2160          * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2161          * only happen when unmounting /.  Since that operation will fail
2162          * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2163          */
2164         if (vp == NULL)
2165                 return (EBUSY);
2166 
2167         vpvfsentry = vn_vfslocks_getlock(vp);
2168 
2169         if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER))
2170                 return (0);
2171 
2172         vn_vfslocks_rele(vpvfsentry);
2173         return (EBUSY);
2174 }
2175 
2176 int
2177 vn_vfsrlock(vnode_t *vp)
2178 {
2179         vn_vfslocks_entry_t *vpvfsentry;
2180 
2181         /*
2182          * If vp is NULL then somebody is trying to lock the covered vnode
2183          * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2184          * only happen when unmounting /.  Since that operation will fail
2185          * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2186          */
2187         if (vp == NULL)
2188                 return (EBUSY);
2189 
2190         vpvfsentry = vn_vfslocks_getlock(vp);
2191 
2192         if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER))
2193                 return (0);
2194 
2195         vn_vfslocks_rele(vpvfsentry);
2196         return (EBUSY);
2197 }
2198 
2199 void
2200 vn_vfsunlock(vnode_t *vp)
2201 {
2202         vn_vfslocks_entry_t *vpvfsentry;
2203 
2204         /*
2205          * ve_refcnt needs to be decremented twice.
2206          * 1. To release refernce after a call to vn_vfslocks_getlock()
2207          * 2. To release the reference from the locking routines like
2208          *    vn_vfsrlock/vn_vfswlock etc,.
2209          */
2210         vpvfsentry = vn_vfslocks_getlock(vp);
2211         vn_vfslocks_rele(vpvfsentry);
2212 
2213         rwst_exit(&vpvfsentry->ve_lock);
2214         vn_vfslocks_rele(vpvfsentry);
2215 }
2216 
2217 int
2218 vn_vfswlock_held(vnode_t *vp)
2219 {
2220         int held;
2221         vn_vfslocks_entry_t *vpvfsentry;
2222 
2223         ASSERT(vp != NULL);
2224 
2225         vpvfsentry = vn_vfslocks_getlock(vp);
2226         held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER);
2227 
2228         vn_vfslocks_rele(vpvfsentry);
2229         return (held);
2230 }
2231 
2232 
2233 int
2234 vn_make_ops(
2235         const char *name,                       /* Name of file system */
2236         const fs_operation_def_t *templ,        /* Operation specification */
2237         vnodeops_t **actual)                    /* Return the vnodeops */
2238 {
2239         int unused_ops;
2240         int error;
2241 
2242         *actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP);
2243 
2244         (*actual)->vnop_name = name;
2245 
2246         error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ);
2247         if (error) {
2248                 kmem_free(*actual, sizeof (vnodeops_t));
2249         }
2250 
2251 #if DEBUG
2252         if (unused_ops != 0)
2253                 cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied "
2254                     "but not used", name, unused_ops);
2255 #endif
2256 
2257         return (error);
2258 }
2259 
2260 /*
2261  * Free the vnodeops created as a result of vn_make_ops()
2262  */
2263 void
2264 vn_freevnodeops(vnodeops_t *vnops)
2265 {
2266         kmem_free(vnops, sizeof (vnodeops_t));
2267 }
2268 
2269 /*
2270  * Vnode cache.
2271  */
2272 
2273 /* ARGSUSED */
2274 static int
2275 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
2276 {
2277         struct vnode *vp;
2278 
2279         vp = buf;
2280 
2281         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
2282         mutex_init(&vp->v_vsd_lock, NULL, MUTEX_DEFAULT, NULL);
2283         cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
2284         rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
2285         vp->v_femhead = NULL;        /* Must be done before vn_reinit() */
2286         vp->v_path = NULL;
2287         vp->v_mpssdata = NULL;
2288         vp->v_vsd = NULL;
2289         vp->v_fopdata = NULL;
2290 
2291         return (0);
2292 }
2293 
2294 /* ARGSUSED */
2295 static void
2296 vn_cache_destructor(void *buf, void *cdrarg)
2297 {
2298         struct vnode *vp;
2299 
2300         vp = buf;
2301 
2302         rw_destroy(&vp->v_nbllock);
2303         cv_destroy(&vp->v_cv);
2304         mutex_destroy(&vp->v_vsd_lock);
2305         mutex_destroy(&vp->v_lock);
2306 }
2307 
2308 void
2309 vn_create_cache(void)
2310 {
2311         /* LINTED */
2312         ASSERT((1 << VNODE_ALIGN_LOG2) ==
2313             P2ROUNDUP(sizeof (struct vnode), VNODE_ALIGN));
2314         vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode),
2315             VNODE_ALIGN, vn_cache_constructor, vn_cache_destructor, NULL, NULL,
2316             NULL, 0);
2317 }
2318 
2319 void
2320 vn_destroy_cache(void)
2321 {
2322         kmem_cache_destroy(vn_cache);
2323 }
2324 
2325 /*
2326  * Used by file systems when fs-specific nodes (e.g., ufs inodes) are
2327  * cached by the file system and vnodes remain associated.
2328  */
2329 void
2330 vn_recycle(vnode_t *vp)
2331 {
2332         ASSERT(vp->v_pages == NULL);
2333 
2334         /*
2335          * XXX - This really belongs in vn_reinit(), but we have some issues
2336          * with the counts.  Best to have it here for clean initialization.
2337          */
2338         vp->v_rdcnt = 0;
2339         vp->v_wrcnt = 0;
2340         vp->v_mmap_read = 0;
2341         vp->v_mmap_write = 0;
2342 
2343         /*
2344          * If FEM was in use, make sure everything gets cleaned up
2345          * NOTE: vp->v_femhead is initialized to NULL in the vnode
2346          * constructor.
2347          */
2348         if (vp->v_femhead) {
2349                 /* XXX - There should be a free_femhead() that does all this */
2350                 ASSERT(vp->v_femhead->femh_list == NULL);
2351                 mutex_destroy(&vp->v_femhead->femh_lock);
2352                 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2353                 vp->v_femhead = NULL;
2354         }
2355         if (vp->v_path) {
2356                 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2357                 vp->v_path = NULL;
2358         }
2359 
2360         if (vp->v_fopdata != NULL) {
2361                 free_fopdata(vp);
2362         }
2363         vp->v_mpssdata = NULL;
2364         vsd_free(vp);
2365 }
2366 
2367 /*
2368  * Used to reset the vnode fields including those that are directly accessible
2369  * as well as those which require an accessor function.
2370  *
2371  * Does not initialize:
2372  *      synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv
2373  *      v_data (since FS-nodes and vnodes point to each other and should
2374  *              be updated simultaneously)
2375  *      v_op (in case someone needs to make a VOP call on this object)
2376  */
2377 void
2378 vn_reinit(vnode_t *vp)
2379 {
2380         vp->v_count = 1;
2381         vp->v_count_dnlc = 0;
2382         vp->v_vfsp = NULL;
2383         vp->v_stream = NULL;
2384         vp->v_vfsmountedhere = NULL;
2385         vp->v_flag = 0;
2386         vp->v_type = VNON;
2387         vp->v_rdev = NODEV;
2388 
2389         vp->v_filocks = NULL;
2390         vp->v_shrlocks = NULL;
2391         vp->v_pages = NULL;
2392 
2393         vp->v_locality = NULL;
2394         vp->v_xattrdir = NULL;
2395 
2396         /* Handles v_femhead, v_path, and the r/w/map counts */
2397         vn_recycle(vp);
2398 }
2399 
2400 vnode_t *
2401 vn_alloc(int kmflag)
2402 {
2403         vnode_t *vp;
2404 
2405         vp = kmem_cache_alloc(vn_cache, kmflag);
2406 
2407         if (vp != NULL) {
2408                 vp->v_femhead = NULL;        /* Must be done before vn_reinit() */
2409                 vp->v_fopdata = NULL;
2410                 vn_reinit(vp);
2411         }
2412 
2413         return (vp);
2414 }
2415 
2416 void
2417 vn_free(vnode_t *vp)
2418 {
2419         ASSERT(vp->v_shrlocks == NULL);
2420         ASSERT(vp->v_filocks == NULL);
2421 
2422         /*
2423          * Some file systems call vn_free() with v_count of zero,
2424          * some with v_count of 1.  In any case, the value should
2425          * never be anything else.
2426          */
2427         ASSERT((vp->v_count == 0) || (vp->v_count == 1));
2428         ASSERT(vp->v_count_dnlc == 0);
2429         if (vp->v_path != NULL) {
2430                 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2431                 vp->v_path = NULL;
2432         }
2433 
2434         /* If FEM was in use, make sure everything gets cleaned up */
2435         if (vp->v_femhead) {
2436                 /* XXX - There should be a free_femhead() that does all this */
2437                 ASSERT(vp->v_femhead->femh_list == NULL);
2438                 mutex_destroy(&vp->v_femhead->femh_lock);
2439                 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2440                 vp->v_femhead = NULL;
2441         }
2442 
2443         if (vp->v_fopdata != NULL) {
2444                 free_fopdata(vp);
2445         }
2446         vp->v_mpssdata = NULL;
2447         vsd_free(vp);
2448         kmem_cache_free(vn_cache, vp);
2449 }
2450 
2451 /*
2452  * vnode status changes, should define better states than 1, 0.
2453  */
2454 void
2455 vn_reclaim(vnode_t *vp)
2456 {
2457         vfs_t   *vfsp = vp->v_vfsp;
2458 
2459         if (vfsp == NULL ||
2460             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2461                 return;
2462         }
2463         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED);
2464 }
2465 
2466 void
2467 vn_idle(vnode_t *vp)
2468 {
2469         vfs_t   *vfsp = vp->v_vfsp;
2470 
2471         if (vfsp == NULL ||
2472             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2473                 return;
2474         }
2475         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED);
2476 }
2477 void
2478 vn_exists(vnode_t *vp)
2479 {
2480         vfs_t   *vfsp = vp->v_vfsp;
2481 
2482         if (vfsp == NULL ||
2483             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2484                 return;
2485         }
2486         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS);
2487 }
2488 
2489 void
2490 vn_invalid(vnode_t *vp)
2491 {
2492         vfs_t   *vfsp = vp->v_vfsp;
2493 
2494         if (vfsp == NULL ||
2495             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2496                 return;
2497         }
2498         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED);
2499 }
2500 
2501 /* Vnode event notification */
2502 
2503 int
2504 vnevent_support(vnode_t *vp, caller_context_t *ct)
2505 {
2506         if (vp == NULL)
2507                 return (EINVAL);
2508 
2509         return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct));
2510 }
2511 
2512 void
2513 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2514 {
2515         if (vp == NULL || vp->v_femhead == NULL) {
2516                 return;
2517         }
2518         (void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
2519 }
2520 
2521 void
2522 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
2523     caller_context_t *ct)
2524 {
2525         if (vp == NULL || vp->v_femhead == NULL) {
2526                 return;
2527         }
2528         (void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct);
2529 }
2530 
2531 void
2532 vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct)
2533 {
2534         if (vp == NULL || vp->v_femhead == NULL) {
2535                 return;
2536         }
2537         (void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct);
2538 }
2539 
2540 void
2541 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2542 {
2543         if (vp == NULL || vp->v_femhead == NULL) {
2544                 return;
2545         }
2546         (void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct);
2547 }
2548 
2549 void
2550 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2551 {
2552         if (vp == NULL || vp->v_femhead == NULL) {
2553                 return;
2554         }
2555         (void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct);
2556 }
2557 
2558 void
2559 vnevent_create(vnode_t *vp, caller_context_t *ct)
2560 {
2561         if (vp == NULL || vp->v_femhead == NULL) {
2562                 return;
2563         }
2564         (void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct);
2565 }
2566 
2567 void
2568 vnevent_link(vnode_t *vp, caller_context_t *ct)
2569 {
2570         if (vp == NULL || vp->v_femhead == NULL) {
2571                 return;
2572         }
2573         (void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct);
2574 }
2575 
2576 void
2577 vnevent_mountedover(vnode_t *vp, caller_context_t *ct)
2578 {
2579         if (vp == NULL || vp->v_femhead == NULL) {
2580                 return;
2581         }
2582         (void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct);
2583 }
2584 
2585 /*
2586  * Vnode accessors.
2587  */
2588 
2589 int
2590 vn_is_readonly(vnode_t *vp)
2591 {
2592         return (vp->v_vfsp->vfs_flag & VFS_RDONLY);
2593 }
2594 
2595 int
2596 vn_has_flocks(vnode_t *vp)
2597 {
2598         return (vp->v_filocks != NULL);
2599 }
2600 
2601 int
2602 vn_has_mandatory_locks(vnode_t *vp, int mode)
2603 {
2604         return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode)));
2605 }
2606 
2607 int
2608 vn_has_cached_data(vnode_t *vp)
2609 {
2610         return (vp->v_pages != NULL);
2611 }
2612 
2613 /*
2614  * Return 0 if the vnode in question shouldn't be permitted into a zone via
2615  * zone_enter(2).
2616  */
2617 int
2618 vn_can_change_zones(vnode_t *vp)
2619 {
2620         struct vfssw *vswp;
2621         int allow = 1;
2622         vnode_t *rvp;
2623 
2624         if (nfs_global_client_only != 0)
2625                 return (1);
2626 
2627         /*
2628          * We always want to look at the underlying vnode if there is one.
2629          */
2630         if (VOP_REALVP(vp, &rvp, NULL) != 0)
2631                 rvp = vp;
2632         /*
2633          * Some pseudo filesystems (including doorfs) don't actually register
2634          * their vfsops_t, so the following may return NULL; we happily let
2635          * such vnodes switch zones.
2636          */
2637         vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp));
2638         if (vswp != NULL) {
2639                 if (vswp->vsw_flag & VSW_NOTZONESAFE)
2640                         allow = 0;
2641                 vfs_unrefvfssw(vswp);
2642         }
2643         return (allow);
2644 }
2645 
2646 /*
2647  * Return nonzero if the vnode is a mount point, zero if not.
2648  */
2649 int
2650 vn_ismntpt(vnode_t *vp)
2651 {
2652         return (vp->v_vfsmountedhere != NULL);
2653 }
2654 
2655 /* Retrieve the vfs (if any) mounted on this vnode */
2656 vfs_t *
2657 vn_mountedvfs(vnode_t *vp)
2658 {
2659         return (vp->v_vfsmountedhere);
2660 }
2661 
2662 /*
2663  * Return nonzero if the vnode is referenced by the dnlc, zero if not.
2664  */
2665 int
2666 vn_in_dnlc(vnode_t *vp)
2667 {
2668         return (vp->v_count_dnlc > 0);
2669 }
2670 
2671 /*
2672  * vn_has_other_opens() checks whether a particular file is opened by more than
2673  * just the caller and whether the open is for read and/or write.
2674  * This routine is for calling after the caller has already called VOP_OPEN()
2675  * and the caller wishes to know if they are the only one with it open for
2676  * the mode(s) specified.
2677  *
2678  * Vnode counts are only kept on regular files (v_type=VREG).
2679  */
2680 int
2681 vn_has_other_opens(
2682         vnode_t *vp,
2683         v_mode_t mode)
2684 {
2685 
2686         ASSERT(vp != NULL);
2687 
2688         switch (mode) {
2689         case V_WRITE:
2690                 if (vp->v_wrcnt > 1)
2691                         return (V_TRUE);
2692                 break;
2693         case V_RDORWR:
2694                 if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1))
2695                         return (V_TRUE);
2696                 break;
2697         case V_RDANDWR:
2698                 if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1))
2699                         return (V_TRUE);
2700                 break;
2701         case V_READ:
2702                 if (vp->v_rdcnt > 1)
2703                         return (V_TRUE);
2704                 break;
2705         }
2706 
2707         return (V_FALSE);
2708 }
2709 
2710 /*
2711  * vn_is_opened() checks whether a particular file is opened and
2712  * whether the open is for read and/or write.
2713  *
2714  * Vnode counts are only kept on regular files (v_type=VREG).
2715  */
2716 int
2717 vn_is_opened(
2718         vnode_t *vp,
2719         v_mode_t mode)
2720 {
2721 
2722         ASSERT(vp != NULL);
2723 
2724         switch (mode) {
2725         case V_WRITE:
2726                 if (vp->v_wrcnt)
2727                         return (V_TRUE);
2728                 break;
2729         case V_RDANDWR:
2730                 if (vp->v_rdcnt && vp->v_wrcnt)
2731                         return (V_TRUE);
2732                 break;
2733         case V_RDORWR:
2734                 if (vp->v_rdcnt || vp->v_wrcnt)
2735                         return (V_TRUE);
2736                 break;
2737         case V_READ:
2738                 if (vp->v_rdcnt)
2739                         return (V_TRUE);
2740                 break;
2741         }
2742 
2743         return (V_FALSE);
2744 }
2745 
2746 /*
2747  * vn_is_mapped() checks whether a particular file is mapped and whether
2748  * the file is mapped read and/or write.
2749  */
2750 int
2751 vn_is_mapped(
2752         vnode_t *vp,
2753         v_mode_t mode)
2754 {
2755 
2756         ASSERT(vp != NULL);
2757 
2758 #if !defined(_LP64)
2759         switch (mode) {
2760         /*
2761          * The atomic_add_64_nv functions force atomicity in the
2762          * case of 32 bit architectures. Otherwise the 64 bit values
2763          * require two fetches. The value of the fields may be
2764          * (potentially) changed between the first fetch and the
2765          * second
2766          */
2767         case V_WRITE:
2768                 if (atomic_add_64_nv((&(vp->v_mmap_write)), 0))
2769                         return (V_TRUE);
2770                 break;
2771         case V_RDANDWR:
2772                 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) &&
2773                     (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2774                         return (V_TRUE);
2775                 break;
2776         case V_RDORWR:
2777                 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) ||
2778                     (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2779                         return (V_TRUE);
2780                 break;
2781         case V_READ:
2782                 if (atomic_add_64_nv((&(vp->v_mmap_read)), 0))
2783                         return (V_TRUE);
2784                 break;
2785         }
2786 #else
2787         switch (mode) {
2788         case V_WRITE:
2789                 if (vp->v_mmap_write)
2790                         return (V_TRUE);
2791                 break;
2792         case V_RDANDWR:
2793                 if (vp->v_mmap_read && vp->v_mmap_write)
2794                         return (V_TRUE);
2795                 break;
2796         case V_RDORWR:
2797                 if (vp->v_mmap_read || vp->v_mmap_write)
2798                         return (V_TRUE);
2799                 break;
2800         case V_READ:
2801                 if (vp->v_mmap_read)
2802                         return (V_TRUE);
2803                 break;
2804         }
2805 #endif
2806 
2807         return (V_FALSE);
2808 }
2809 
2810 /*
2811  * Set the operations vector for a vnode.
2812  *
2813  * FEM ensures that the v_femhead pointer is filled in before the
2814  * v_op pointer is changed.  This means that if the v_femhead pointer
2815  * is NULL, and the v_op field hasn't changed since before which checked
2816  * the v_femhead pointer; then our update is ok - we are not racing with
2817  * FEM.
2818  */
2819 void
2820 vn_setops(vnode_t *vp, vnodeops_t *vnodeops)
2821 {
2822         vnodeops_t      *op;
2823 
2824         ASSERT(vp != NULL);
2825         ASSERT(vnodeops != NULL);
2826 
2827         op = vp->v_op;
2828         membar_consumer();
2829         /*
2830          * If vp->v_femhead == NULL, then we'll call casptr() to do the
2831          * compare-and-swap on vp->v_op.  If either fails, then FEM is
2832          * in effect on the vnode and we need to have FEM deal with it.
2833          */
2834         if (vp->v_femhead != NULL || casptr(&vp->v_op, op, vnodeops) != op) {
2835                 fem_setvnops(vp, vnodeops);
2836         }
2837 }
2838 
2839 /*
2840  * Retrieve the operations vector for a vnode
2841  * As with vn_setops(above); make sure we aren't racing with FEM.
2842  * FEM sets the v_op to a special, internal, vnodeops that wouldn't
2843  * make sense to the callers of this routine.
2844  */
2845 vnodeops_t *
2846 vn_getops(vnode_t *vp)
2847 {
2848         vnodeops_t      *op;
2849 
2850         ASSERT(vp != NULL);
2851 
2852         op = vp->v_op;
2853         membar_consumer();
2854         if (vp->v_femhead == NULL && op == vp->v_op) {
2855                 return (op);
2856         } else {
2857                 return (fem_getvnops(vp));
2858         }
2859 }
2860 
2861 /*
2862  * Returns non-zero (1) if the vnodeops matches that of the vnode.
2863  * Returns zero (0) if not.
2864  */
2865 int
2866 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops)
2867 {
2868         return (vn_getops(vp) == vnodeops);
2869 }
2870 
2871 /*
2872  * Returns non-zero (1) if the specified operation matches the
2873  * corresponding operation for that the vnode.
2874  * Returns zero (0) if not.
2875  */
2876 
2877 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0))
2878 
2879 int
2880 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp)
2881 {
2882         const fs_operation_trans_def_t *otdp;
2883         fs_generic_func_p *loc = NULL;
2884         vnodeops_t      *vop = vn_getops(vp);
2885 
2886         ASSERT(vopname != NULL);
2887 
2888         for (otdp = vn_ops_table; otdp->name != NULL; otdp++) {
2889                 if (MATCHNAME(otdp->name, vopname)) {
2890                         loc = (fs_generic_func_p *)
2891                             ((char *)(vop) + otdp->offset);
2892                         break;
2893                 }
2894         }
2895 
2896         return ((loc != NULL) && (*loc == funcp));
2897 }
2898 
2899 /*
2900  * fs_new_caller_id() needs to return a unique ID on a given local system.
2901  * The IDs do not need to survive across reboots.  These are primarily
2902  * used so that (FEM) monitors can detect particular callers (such as
2903  * the NFS server) to a given vnode/vfs operation.
2904  */
2905 u_longlong_t
2906 fs_new_caller_id()
2907 {
2908         static uint64_t next_caller_id = 0LL; /* First call returns 1 */
2909 
2910         return ((u_longlong_t)atomic_add_64_nv(&next_caller_id, 1));
2911 }
2912 
2913 /*
2914  * Given a starting vnode and a path, updates the path in the target vnode in
2915  * a safe manner.  If the vnode already has path information embedded, then the
2916  * cached path is left untouched.
2917  */
2918 
2919 size_t max_vnode_path = 4 * MAXPATHLEN;
2920 
2921 void
2922 vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
2923     const char *path, size_t plen)
2924 {
2925         char    *rpath;
2926         vnode_t *base;
2927         size_t  rpathlen, rpathalloc;
2928         int     doslash = 1;
2929 
2930         if (*path == '/') {
2931                 base = rootvp;
2932                 path++;
2933                 plen--;
2934         } else {
2935                 base = startvp;
2936         }
2937 
2938         /*
2939          * We cannot grab base->v_lock while we hold vp->v_lock because of
2940          * the potential for deadlock.
2941          */
2942         mutex_enter(&base->v_lock);
2943         if (base->v_path == NULL) {
2944                 mutex_exit(&base->v_lock);
2945                 return;
2946         }
2947 
2948         rpathlen = strlen(base->v_path);
2949         rpathalloc = rpathlen + plen + 1;
2950         /* Avoid adding a slash if there's already one there */
2951         if (base->v_path[rpathlen-1] == '/')
2952                 doslash = 0;
2953         else
2954                 rpathalloc++;
2955 
2956         /*
2957          * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held,
2958          * so we must do this dance.  If, by chance, something changes the path,
2959          * just give up since there is no real harm.
2960          */
2961         mutex_exit(&base->v_lock);
2962 
2963         /* Paths should stay within reason */
2964         if (rpathalloc > max_vnode_path)
2965                 return;
2966 
2967         rpath = kmem_alloc(rpathalloc, KM_SLEEP);
2968 
2969         mutex_enter(&base->v_lock);
2970         if (base->v_path == NULL || strlen(base->v_path) != rpathlen) {
2971                 mutex_exit(&base->v_lock);
2972                 kmem_free(rpath, rpathalloc);
2973                 return;
2974         }
2975         bcopy(base->v_path, rpath, rpathlen);
2976         mutex_exit(&base->v_lock);
2977 
2978         if (doslash)
2979                 rpath[rpathlen++] = '/';
2980         bcopy(path, rpath + rpathlen, plen);
2981         rpath[rpathlen + plen] = '\0';
2982 
2983         mutex_enter(&vp->v_lock);
2984         if (vp->v_path != NULL) {
2985                 mutex_exit(&vp->v_lock);
2986                 kmem_free(rpath, rpathalloc);
2987         } else {
2988                 vp->v_path = rpath;
2989                 mutex_exit(&vp->v_lock);
2990         }
2991 }
2992 
2993 /*
2994  * Sets the path to the vnode to be the given string, regardless of current
2995  * context.  The string must be a complete path from rootdir.  This is only used
2996  * by fsop_root() for setting the path based on the mountpoint.
2997  */
2998 void
2999 vn_setpath_str(struct vnode *vp, const char *str, size_t len)
3000 {
3001         char *buf = kmem_alloc(len + 1, KM_SLEEP);
3002 
3003         mutex_enter(&vp->v_lock);
3004         if (vp->v_path != NULL) {
3005                 mutex_exit(&vp->v_lock);
3006                 kmem_free(buf, len + 1);
3007                 return;
3008         }
3009 
3010         vp->v_path = buf;
3011         bcopy(str, vp->v_path, len);
3012         vp->v_path[len] = '\0';
3013 
3014         mutex_exit(&vp->v_lock);
3015 }
3016 
3017 /*
3018  * Called from within filesystem's vop_rename() to handle renames once the
3019  * target vnode is available.
3020  */
3021 void
3022 vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len)
3023 {
3024         char *tmp;
3025 
3026         mutex_enter(&vp->v_lock);
3027         tmp = vp->v_path;
3028         vp->v_path = NULL;
3029         mutex_exit(&vp->v_lock);
3030         vn_setpath(rootdir, dvp, vp, nm, len);
3031         if (tmp != NULL)
3032                 kmem_free(tmp, strlen(tmp) + 1);
3033 }
3034 
3035 /*
3036  * Similar to vn_setpath_str(), this function sets the path of the destination
3037  * vnode to the be the same as the source vnode.
3038  */
3039 void
3040 vn_copypath(struct vnode *src, struct vnode *dst)
3041 {
3042         char *buf;
3043         int alloc;
3044 
3045         mutex_enter(&src->v_lock);
3046         if (src->v_path == NULL) {
3047                 mutex_exit(&src->v_lock);
3048                 return;
3049         }
3050         alloc = strlen(src->v_path) + 1;
3051 
3052         /* avoid kmem_alloc() with lock held */
3053         mutex_exit(&src->v_lock);
3054         buf = kmem_alloc(alloc, KM_SLEEP);
3055         mutex_enter(&src->v_lock);
3056         if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) {
3057                 mutex_exit(&src->v_lock);
3058                 kmem_free(buf, alloc);
3059                 return;
3060         }
3061         bcopy(src->v_path, buf, alloc);
3062         mutex_exit(&src->v_lock);
3063 
3064         mutex_enter(&dst->v_lock);
3065         if (dst->v_path != NULL) {
3066                 mutex_exit(&dst->v_lock);
3067                 kmem_free(buf, alloc);
3068                 return;
3069         }
3070         dst->v_path = buf;
3071         mutex_exit(&dst->v_lock);
3072 }
3073 
3074 /*
3075  * XXX Private interface for segvn routines that handle vnode
3076  * large page segments.
3077  *
3078  * return 1 if vp's file system VOP_PAGEIO() implementation
3079  * can be safely used instead of VOP_GETPAGE() for handling
3080  * pagefaults against regular non swap files. VOP_PAGEIO()
3081  * interface is considered safe here if its implementation
3082  * is very close to VOP_GETPAGE() implementation.
3083  * e.g. It zero's out the part of the page beyond EOF. Doesn't
3084  * panic if there're file holes but instead returns an error.
3085  * Doesn't assume file won't be changed by user writes, etc.
3086  *
3087  * return 0 otherwise.
3088  *
3089  * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs.
3090  */
3091 int
3092 vn_vmpss_usepageio(vnode_t *vp)
3093 {
3094         vfs_t   *vfsp = vp->v_vfsp;
3095         char *fsname = vfssw[vfsp->vfs_fstype].vsw_name;
3096         char *pageio_ok_fss[] = {"ufs", "nfs", NULL};
3097         char **fsok = pageio_ok_fss;
3098 
3099         if (fsname == NULL) {
3100                 return (0);
3101         }
3102 
3103         for (; *fsok; fsok++) {
3104                 if (strcmp(*fsok, fsname) == 0) {
3105                         return (1);
3106                 }
3107         }
3108         return (0);
3109 }
3110 
3111 /* VOP_XXX() macros call the corresponding fop_xxx() function */
3112 
3113 int
3114 fop_open(
3115         vnode_t **vpp,
3116         int mode,
3117         cred_t *cr,
3118         caller_context_t *ct)
3119 {
3120         int ret;
3121         vnode_t *vp = *vpp;
3122 
3123         VN_HOLD(vp);
3124         /*
3125          * Adding to the vnode counts before calling open
3126          * avoids the need for a mutex. It circumvents a race
3127          * condition where a query made on the vnode counts results in a
3128          * false negative. The inquirer goes away believing the file is
3129          * not open when there is an open on the file already under way.
3130          *
3131          * The counts are meant to prevent NFS from granting a delegation
3132          * when it would be dangerous to do so.
3133          *
3134          * The vnode counts are only kept on regular files
3135          */
3136         if ((*vpp)->v_type == VREG) {
3137                 if (mode & FREAD)
3138                         atomic_add_32(&((*vpp)->v_rdcnt), 1);
3139                 if (mode & FWRITE)
3140                         atomic_add_32(&((*vpp)->v_wrcnt), 1);
3141         }
3142 
3143         VOPXID_MAP_CR(vp, cr);
3144 
3145         ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct);
3146 
3147         if (ret) {
3148                 /*
3149                  * Use the saved vp just in case the vnode ptr got trashed
3150                  * by the error.
3151                  */
3152                 VOPSTATS_UPDATE(vp, open);
3153                 if ((vp->v_type == VREG) && (mode & FREAD))
3154                         atomic_add_32(&(vp->v_rdcnt), -1);
3155                 if ((vp->v_type == VREG) && (mode & FWRITE))
3156                         atomic_add_32(&(vp->v_wrcnt), -1);
3157         } else {
3158                 /*
3159                  * Some filesystems will return a different vnode,
3160                  * but the same path was still used to open it.
3161                  * So if we do change the vnode and need to
3162                  * copy over the path, do so here, rather than special
3163                  * casing each filesystem. Adjust the vnode counts to
3164                  * reflect the vnode switch.
3165                  */
3166                 VOPSTATS_UPDATE(*vpp, open);
3167                 if (*vpp != vp && *vpp != NULL) {
3168                         vn_copypath(vp, *vpp);
3169                         if (((*vpp)->v_type == VREG) && (mode & FREAD))
3170                                 atomic_add_32(&((*vpp)->v_rdcnt), 1);
3171                         if ((vp->v_type == VREG) && (mode & FREAD))
3172                                 atomic_add_32(&(vp->v_rdcnt), -1);
3173                         if (((*vpp)->v_type == VREG) && (mode & FWRITE))
3174                                 atomic_add_32(&((*vpp)->v_wrcnt), 1);
3175                         if ((vp->v_type == VREG) && (mode & FWRITE))
3176                                 atomic_add_32(&(vp->v_wrcnt), -1);
3177                 }
3178         }
3179         VN_RELE(vp);
3180         return (ret);
3181 }
3182 
3183 int
3184 fop_close(
3185         vnode_t *vp,
3186         int flag,
3187         int count,
3188         offset_t offset,
3189         cred_t *cr,
3190         caller_context_t *ct)
3191 {
3192         int err;
3193 
3194         VOPXID_MAP_CR(vp, cr);
3195 
3196         err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct);
3197         VOPSTATS_UPDATE(vp, close);
3198         /*
3199          * Check passed in count to handle possible dups. Vnode counts are only
3200          * kept on regular files
3201          */
3202         if ((vp->v_type == VREG) && (count == 1))  {
3203                 if (flag & FREAD) {
3204                         ASSERT(vp->v_rdcnt > 0);
3205                         atomic_add_32(&(vp->v_rdcnt), -1);
3206                 }
3207                 if (flag & FWRITE) {
3208                         ASSERT(vp->v_wrcnt > 0);
3209                         atomic_add_32(&(vp->v_wrcnt), -1);
3210                 }
3211         }
3212         return (err);
3213 }
3214 
3215 int
3216 fop_read(
3217         vnode_t *vp,
3218         uio_t *uiop,
3219         int ioflag,
3220         cred_t *cr,
3221         caller_context_t *ct)
3222 {
3223         int     err;
3224         ssize_t resid_start = uiop->uio_resid;
3225 
3226         VOPXID_MAP_CR(vp, cr);
3227 
3228         err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
3229         VOPSTATS_UPDATE_IO(vp, read,
3230             read_bytes, (resid_start - uiop->uio_resid));
3231         return (err);
3232 }
3233 
3234 int
3235 fop_write(
3236         vnode_t *vp,
3237         uio_t *uiop,
3238         int ioflag,
3239         cred_t *cr,
3240         caller_context_t *ct)
3241 {
3242         int     err;
3243         ssize_t resid_start = uiop->uio_resid;
3244 
3245         VOPXID_MAP_CR(vp, cr);
3246 
3247         err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
3248         VOPSTATS_UPDATE_IO(vp, write,
3249             write_bytes, (resid_start - uiop->uio_resid));
3250         return (err);
3251 }
3252 
3253 int
3254 fop_ioctl(
3255         vnode_t *vp,
3256         int cmd,
3257         intptr_t arg,
3258         int flag,
3259         cred_t *cr,
3260         int *rvalp,
3261         caller_context_t *ct)
3262 {
3263         int     err;
3264 
3265         VOPXID_MAP_CR(vp, cr);
3266 
3267         err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct);
3268         VOPSTATS_UPDATE(vp, ioctl);
3269         return (err);
3270 }
3271 
3272 int
3273 fop_setfl(
3274         vnode_t *vp,
3275         int oflags,
3276         int nflags,
3277         cred_t *cr,
3278         caller_context_t *ct)
3279 {
3280         int     err;
3281 
3282         VOPXID_MAP_CR(vp, cr);
3283 
3284         err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct);
3285         VOPSTATS_UPDATE(vp, setfl);
3286         return (err);
3287 }
3288 
3289 int
3290 fop_getattr(
3291         vnode_t *vp,
3292         vattr_t *vap,
3293         int flags,
3294         cred_t *cr,
3295         caller_context_t *ct)
3296 {
3297         int     err;
3298 
3299         VOPXID_MAP_CR(vp, cr);
3300 
3301         /*
3302          * If this file system doesn't understand the xvattr extensions
3303          * then turn off the xvattr bit.
3304          */
3305         if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3306                 vap->va_mask &= ~AT_XVATTR;
3307         }
3308 
3309         /*
3310          * We're only allowed to skip the ACL check iff we used a 32 bit
3311          * ACE mask with VOP_ACCESS() to determine permissions.
3312          */
3313         if ((flags & ATTR_NOACLCHECK) &&
3314             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3315                 return (EINVAL);
3316         }
3317         err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct);
3318         VOPSTATS_UPDATE(vp, getattr);
3319         return (err);
3320 }
3321 
3322 int
3323 fop_setattr(
3324         vnode_t *vp,
3325         vattr_t *vap,
3326         int flags,
3327         cred_t *cr,
3328         caller_context_t *ct)
3329 {
3330         int     err;
3331 
3332         VOPXID_MAP_CR(vp, cr);
3333 
3334         /*
3335          * If this file system doesn't understand the xvattr extensions
3336          * then turn off the xvattr bit.
3337          */
3338         if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3339                 vap->va_mask &= ~AT_XVATTR;
3340         }
3341 
3342         /*
3343          * We're only allowed to skip the ACL check iff we used a 32 bit
3344          * ACE mask with VOP_ACCESS() to determine permissions.
3345          */
3346         if ((flags & ATTR_NOACLCHECK) &&
3347             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3348                 return (EINVAL);
3349         }
3350         err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct);
3351         VOPSTATS_UPDATE(vp, setattr);
3352         return (err);
3353 }
3354 
3355 int
3356 fop_access(
3357         vnode_t *vp,
3358         int mode,
3359         int flags,
3360         cred_t *cr,
3361         caller_context_t *ct)
3362 {
3363         int     err;
3364 
3365         if ((flags & V_ACE_MASK) &&
3366             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3367                 return (EINVAL);
3368         }
3369 
3370         VOPXID_MAP_CR(vp, cr);
3371 
3372         err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct);
3373         VOPSTATS_UPDATE(vp, access);
3374         return (err);
3375 }
3376 
3377 int
3378 fop_lookup(
3379         vnode_t *dvp,
3380         char *nm,
3381         vnode_t **vpp,
3382         pathname_t *pnp,
3383         int flags,
3384         vnode_t *rdir,
3385         cred_t *cr,
3386         caller_context_t *ct,
3387         int *deflags,           /* Returned per-dirent flags */
3388         pathname_t *ppnp)       /* Returned case-preserved name in directory */
3389 {
3390         int ret;
3391 
3392         /*
3393          * If this file system doesn't support case-insensitive access
3394          * and said access is requested, fail quickly.  It is required
3395          * that if the vfs supports case-insensitive lookup, it also
3396          * supports extended dirent flags.
3397          */
3398         if (flags & FIGNORECASE &&
3399             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3400             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3401                 return (EINVAL);
3402 
3403         VOPXID_MAP_CR(dvp, cr);
3404 
3405         if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) {
3406                 ret = xattr_dir_lookup(dvp, vpp, flags, cr);
3407         } else {
3408                 ret = (*(dvp)->v_op->vop_lookup)
3409                     (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp);
3410         }
3411         if (ret == 0 && *vpp) {
3412                 VOPSTATS_UPDATE(*vpp, lookup);
3413                 if ((*vpp)->v_path == NULL) {
3414                         vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm));
3415                 }
3416         }
3417 
3418         return (ret);
3419 }
3420 
3421 int
3422 fop_create(
3423         vnode_t *dvp,
3424         char *name,
3425         vattr_t *vap,
3426         vcexcl_t excl,
3427         int mode,
3428         vnode_t **vpp,
3429         cred_t *cr,
3430         int flags,
3431         caller_context_t *ct,
3432         vsecattr_t *vsecp)      /* ACL to set during create */
3433 {
3434         int ret;
3435 
3436         if (vsecp != NULL &&
3437             vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3438                 return (EINVAL);
3439         }
3440         /*
3441          * If this file system doesn't support case-insensitive access
3442          * and said access is requested, fail quickly.
3443          */
3444         if (flags & FIGNORECASE &&
3445             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3446             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3447                 return (EINVAL);
3448 
3449         VOPXID_MAP_CR(dvp, cr);
3450 
3451         ret = (*(dvp)->v_op->vop_create)
3452             (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
3453         if (ret == 0 && *vpp) {
3454                 VOPSTATS_UPDATE(*vpp, create);
3455                 if ((*vpp)->v_path == NULL) {
3456                         vn_setpath(rootdir, dvp, *vpp, name, strlen(name));
3457                 }
3458         }
3459 
3460         return (ret);
3461 }
3462 
3463 int
3464 fop_remove(
3465         vnode_t *dvp,
3466         char *nm,
3467         cred_t *cr,
3468         caller_context_t *ct,
3469         int flags)
3470 {
3471         int     err;
3472 
3473         /*
3474          * If this file system doesn't support case-insensitive access
3475          * and said access is requested, fail quickly.
3476          */
3477         if (flags & FIGNORECASE &&
3478             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3479             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3480                 return (EINVAL);
3481 
3482         VOPXID_MAP_CR(dvp, cr);
3483 
3484         err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags);
3485         VOPSTATS_UPDATE(dvp, remove);
3486         return (err);
3487 }
3488 
3489 int
3490 fop_link(
3491         vnode_t *tdvp,
3492         vnode_t *svp,
3493         char *tnm,
3494         cred_t *cr,
3495         caller_context_t *ct,
3496         int flags)
3497 {
3498         int     err;
3499 
3500         /*
3501          * If the target file system doesn't support case-insensitive access
3502          * and said access is requested, fail quickly.
3503          */
3504         if (flags & FIGNORECASE &&
3505             (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3506             vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3507                 return (EINVAL);
3508 
3509         VOPXID_MAP_CR(tdvp, cr);
3510 
3511         err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags);
3512         VOPSTATS_UPDATE(tdvp, link);
3513         return (err);
3514 }
3515 
3516 int
3517 fop_rename(
3518         vnode_t *sdvp,
3519         char *snm,
3520         vnode_t *tdvp,
3521         char *tnm,
3522         cred_t *cr,
3523         caller_context_t *ct,
3524         int flags)
3525 {
3526         int     err;
3527 
3528         /*
3529          * If the file system involved does not support
3530          * case-insensitive access and said access is requested, fail
3531          * quickly.
3532          */
3533         if (flags & FIGNORECASE &&
3534             ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3535             vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)))
3536                 return (EINVAL);
3537 
3538         VOPXID_MAP_CR(tdvp, cr);
3539 
3540         err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags);
3541         VOPSTATS_UPDATE(sdvp, rename);
3542         return (err);
3543 }
3544 
3545 int
3546 fop_mkdir(
3547         vnode_t *dvp,
3548         char *dirname,
3549         vattr_t *vap,
3550         vnode_t **vpp,
3551         cred_t *cr,
3552         caller_context_t *ct,
3553         int flags,
3554         vsecattr_t *vsecp)      /* ACL to set during create */
3555 {
3556         int ret;
3557 
3558         if (vsecp != NULL &&
3559             vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3560                 return (EINVAL);
3561         }
3562         /*
3563          * If this file system doesn't support case-insensitive access
3564          * and said access is requested, fail quickly.
3565          */
3566         if (flags & FIGNORECASE &&
3567             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3568             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3569                 return (EINVAL);
3570 
3571         VOPXID_MAP_CR(dvp, cr);
3572 
3573         ret = (*(dvp)->v_op->vop_mkdir)
3574             (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
3575         if (ret == 0 && *vpp) {
3576                 VOPSTATS_UPDATE(*vpp, mkdir);
3577                 if ((*vpp)->v_path == NULL) {
3578                         vn_setpath(rootdir, dvp, *vpp, dirname,
3579                             strlen(dirname));
3580                 }
3581         }
3582 
3583         return (ret);
3584 }
3585 
3586 int
3587 fop_rmdir(
3588         vnode_t *dvp,
3589         char *nm,
3590         vnode_t *cdir,
3591         cred_t *cr,
3592         caller_context_t *ct,
3593         int flags)
3594 {
3595         int     err;
3596 
3597         /*
3598          * If this file system doesn't support case-insensitive access
3599          * and said access is requested, fail quickly.
3600          */
3601         if (flags & FIGNORECASE &&
3602             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3603             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3604                 return (EINVAL);
3605 
3606         VOPXID_MAP_CR(dvp, cr);
3607 
3608         err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags);
3609         VOPSTATS_UPDATE(dvp, rmdir);
3610         return (err);
3611 }
3612 
3613 int
3614 fop_readdir(
3615         vnode_t *vp,
3616         uio_t *uiop,
3617         cred_t *cr,
3618         int *eofp,
3619         caller_context_t *ct,
3620         int flags)
3621 {
3622         int     err;
3623         ssize_t resid_start = uiop->uio_resid;
3624 
3625         /*
3626          * If this file system doesn't support retrieving directory
3627          * entry flags and said access is requested, fail quickly.
3628          */
3629         if (flags & V_RDDIR_ENTFLAGS &&
3630             vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0)
3631                 return (EINVAL);
3632 
3633         VOPXID_MAP_CR(vp, cr);
3634 
3635         err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags);
3636         VOPSTATS_UPDATE_IO(vp, readdir,
3637             readdir_bytes, (resid_start - uiop->uio_resid));
3638         return (err);
3639 }
3640 
3641 int
3642 fop_symlink(
3643         vnode_t *dvp,
3644         char *linkname,
3645         vattr_t *vap,
3646         char *target,
3647         cred_t *cr,
3648         caller_context_t *ct,
3649         int flags)
3650 {
3651         int     err;
3652         xvattr_t xvattr;
3653 
3654         /*
3655          * If this file system doesn't support case-insensitive access
3656          * and said access is requested, fail quickly.
3657          */
3658         if (flags & FIGNORECASE &&
3659             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3660             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3661                 return (EINVAL);
3662 
3663         VOPXID_MAP_CR(dvp, cr);
3664 
3665         /* check for reparse point */
3666         if ((vfs_has_feature(dvp->v_vfsp, VFSFT_REPARSE)) &&
3667             (strncmp(target, FS_REPARSE_TAG_STR,
3668             strlen(FS_REPARSE_TAG_STR)) == 0)) {
3669                 if (!fs_reparse_mark(target, vap, &xvattr))
3670                         vap = (vattr_t *)&xvattr;
3671         }
3672 
3673         err = (*(dvp)->v_op->vop_symlink)
3674             (dvp, linkname, vap, target, cr, ct, flags);
3675         VOPSTATS_UPDATE(dvp, symlink);
3676         return (err);
3677 }
3678 
3679 int
3680 fop_readlink(
3681         vnode_t *vp,
3682         uio_t *uiop,
3683         cred_t *cr,
3684         caller_context_t *ct)
3685 {
3686         int     err;
3687 
3688         VOPXID_MAP_CR(vp, cr);
3689 
3690         err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct);
3691         VOPSTATS_UPDATE(vp, readlink);
3692         return (err);
3693 }
3694 
3695 int
3696 fop_fsync(
3697         vnode_t *vp,
3698         int syncflag,
3699         cred_t *cr,
3700         caller_context_t *ct)
3701 {
3702         int     err;
3703 
3704         VOPXID_MAP_CR(vp, cr);
3705 
3706         err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct);
3707         VOPSTATS_UPDATE(vp, fsync);
3708         return (err);
3709 }
3710 
3711 void
3712 fop_inactive(
3713         vnode_t *vp,
3714         cred_t *cr,
3715         caller_context_t *ct)
3716 {
3717         /* Need to update stats before vop call since we may lose the vnode */
3718         VOPSTATS_UPDATE(vp, inactive);
3719 
3720         VOPXID_MAP_CR(vp, cr);
3721 
3722         (*(vp)->v_op->vop_inactive)(vp, cr, ct);
3723 }
3724 
3725 int
3726 fop_fid(
3727         vnode_t *vp,
3728         fid_t *fidp,
3729         caller_context_t *ct)
3730 {
3731         int     err;
3732 
3733         err = (*(vp)->v_op->vop_fid)(vp, fidp, ct);
3734         VOPSTATS_UPDATE(vp, fid);
3735         return (err);
3736 }
3737 
3738 int
3739 fop_rwlock(
3740         vnode_t *vp,
3741         int write_lock,
3742         caller_context_t *ct)
3743 {
3744         int     ret;
3745 
3746         ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct));
3747         VOPSTATS_UPDATE(vp, rwlock);
3748         return (ret);
3749 }
3750 
3751 void
3752 fop_rwunlock(
3753         vnode_t *vp,
3754         int write_lock,
3755         caller_context_t *ct)
3756 {
3757         (*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct);
3758         VOPSTATS_UPDATE(vp, rwunlock);
3759 }
3760 
3761 int
3762 fop_seek(
3763         vnode_t *vp,
3764         offset_t ooff,
3765         offset_t *noffp,
3766         caller_context_t *ct)
3767 {
3768         int     err;
3769 
3770         err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct);
3771         VOPSTATS_UPDATE(vp, seek);
3772         return (err);
3773 }
3774 
3775 int
3776 fop_cmp(
3777         vnode_t *vp1,
3778         vnode_t *vp2,
3779         caller_context_t *ct)
3780 {
3781         int     err;
3782 
3783         err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct);
3784         VOPSTATS_UPDATE(vp1, cmp);
3785         return (err);
3786 }
3787 
3788 int
3789 fop_frlock(
3790         vnode_t *vp,
3791         int cmd,
3792         flock64_t *bfp,
3793         int flag,
3794         offset_t offset,
3795         struct flk_callback *flk_cbp,
3796         cred_t *cr,
3797         caller_context_t *ct)
3798 {
3799         int     err;
3800 
3801         VOPXID_MAP_CR(vp, cr);
3802 
3803         err = (*(vp)->v_op->vop_frlock)
3804             (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3805         VOPSTATS_UPDATE(vp, frlock);
3806         return (err);
3807 }
3808 
3809 int
3810 fop_space(
3811         vnode_t *vp,
3812         int cmd,
3813         flock64_t *bfp,
3814         int flag,
3815         offset_t offset,
3816         cred_t *cr,
3817         caller_context_t *ct)
3818 {
3819         int     err;
3820 
3821         VOPXID_MAP_CR(vp, cr);
3822 
3823         err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct);
3824         VOPSTATS_UPDATE(vp, space);
3825         return (err);
3826 }
3827 
3828 int
3829 fop_realvp(
3830         vnode_t *vp,
3831         vnode_t **vpp,
3832         caller_context_t *ct)
3833 {
3834         int     err;
3835 
3836         err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct);
3837         VOPSTATS_UPDATE(vp, realvp);
3838         return (err);
3839 }
3840 
3841 int
3842 fop_getpage(
3843         vnode_t *vp,
3844         offset_t off,
3845         size_t len,
3846         uint_t *protp,
3847         page_t **plarr,
3848         size_t plsz,
3849         struct seg *seg,
3850         caddr_t addr,
3851         enum seg_rw rw,
3852         cred_t *cr,
3853         caller_context_t *ct)
3854 {
3855         int     err;
3856 
3857         VOPXID_MAP_CR(vp, cr);
3858 
3859         err = (*(vp)->v_op->vop_getpage)
3860             (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct);
3861         VOPSTATS_UPDATE(vp, getpage);
3862         return (err);
3863 }
3864 
3865 int
3866 fop_putpage(
3867         vnode_t *vp,
3868         offset_t off,
3869         size_t len,
3870         int flags,
3871         cred_t *cr,
3872         caller_context_t *ct)
3873 {
3874         int     err;
3875 
3876         VOPXID_MAP_CR(vp, cr);
3877 
3878         err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct);
3879         VOPSTATS_UPDATE(vp, putpage);
3880         return (err);
3881 }
3882 
3883 int
3884 fop_map(
3885         vnode_t *vp,
3886         offset_t off,
3887         struct as *as,
3888         caddr_t *addrp,
3889         size_t len,
3890         uchar_t prot,
3891         uchar_t maxprot,
3892         uint_t flags,
3893         cred_t *cr,
3894         caller_context_t *ct)
3895 {
3896         int     err;
3897 
3898         VOPXID_MAP_CR(vp, cr);
3899 
3900         err = (*(vp)->v_op->vop_map)
3901             (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct);
3902         VOPSTATS_UPDATE(vp, map);
3903         return (err);
3904 }
3905 
3906 int
3907 fop_addmap(
3908         vnode_t *vp,
3909         offset_t off,
3910         struct as *as,
3911         caddr_t addr,
3912         size_t len,
3913         uchar_t prot,
3914         uchar_t maxprot,
3915         uint_t flags,
3916         cred_t *cr,
3917         caller_context_t *ct)
3918 {
3919         int error;
3920         u_longlong_t delta;
3921 
3922         VOPXID_MAP_CR(vp, cr);
3923 
3924         error = (*(vp)->v_op->vop_addmap)
3925             (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3926 
3927         if ((!error) && (vp->v_type == VREG)) {
3928                 delta = (u_longlong_t)btopr(len);
3929                 /*
3930                  * If file is declared MAP_PRIVATE, it can't be written back
3931                  * even if open for write. Handle as read.
3932                  */
3933                 if (flags & MAP_PRIVATE) {
3934                         atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3935                             (int64_t)delta);
3936                 } else {
3937                         /*
3938                          * atomic_add_64 forces the fetch of a 64 bit value to
3939                          * be atomic on 32 bit machines
3940                          */
3941                         if (maxprot & PROT_WRITE)
3942                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
3943                                     (int64_t)delta);
3944                         if (maxprot & PROT_READ)
3945                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3946                                     (int64_t)delta);
3947                         if (maxprot & PROT_EXEC)
3948                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3949                                     (int64_t)delta);
3950                 }
3951         }
3952         VOPSTATS_UPDATE(vp, addmap);
3953         return (error);
3954 }
3955 
3956 int
3957 fop_delmap(
3958         vnode_t *vp,
3959         offset_t off,
3960         struct as *as,
3961         caddr_t addr,
3962         size_t len,
3963         uint_t prot,
3964         uint_t maxprot,
3965         uint_t flags,
3966         cred_t *cr,
3967         caller_context_t *ct)
3968 {
3969         int error;
3970         u_longlong_t delta;
3971 
3972         VOPXID_MAP_CR(vp, cr);
3973 
3974         error = (*(vp)->v_op->vop_delmap)
3975             (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
3976 
3977         /*
3978          * NFS calls into delmap twice, the first time
3979          * it simply establishes a callback mechanism and returns EAGAIN
3980          * while the real work is being done upon the second invocation.
3981          * We have to detect this here and only decrement the counts upon
3982          * the second delmap request.
3983          */
3984         if ((error != EAGAIN) && (vp->v_type == VREG)) {
3985 
3986                 delta = (u_longlong_t)btopr(len);
3987 
3988                 if (flags & MAP_PRIVATE) {
3989                         atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
3990                             (int64_t)(-delta));
3991                 } else {
3992                         /*
3993                          * atomic_add_64 forces the fetch of a 64 bit value
3994                          * to be atomic on 32 bit machines
3995                          */
3996                         if (maxprot & PROT_WRITE)
3997                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
3998                                     (int64_t)(-delta));
3999                         if (maxprot & PROT_READ)
4000                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4001                                     (int64_t)(-delta));
4002                         if (maxprot & PROT_EXEC)
4003                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4004                                     (int64_t)(-delta));
4005                 }
4006         }
4007         VOPSTATS_UPDATE(vp, delmap);
4008         return (error);
4009 }
4010 
4011 
4012 int
4013 fop_poll(
4014         vnode_t *vp,
4015         short events,
4016         int anyyet,
4017         short *reventsp,
4018         struct pollhead **phpp,
4019         caller_context_t *ct)
4020 {
4021         int     err;
4022 
4023         err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct);
4024         VOPSTATS_UPDATE(vp, poll);
4025         return (err);
4026 }
4027 
4028 int
4029 fop_dump(
4030         vnode_t *vp,
4031         caddr_t addr,
4032         offset_t lbdn,
4033         offset_t dblks,
4034         caller_context_t *ct)
4035 {
4036         int     err;
4037 
4038         /* ensure lbdn and dblks can be passed safely to bdev_dump */
4039         if ((lbdn != (daddr_t)lbdn) || (dblks != (int)dblks))
4040                 return (EIO);
4041 
4042         err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct);
4043         VOPSTATS_UPDATE(vp, dump);
4044         return (err);
4045 }
4046 
4047 int
4048 fop_pathconf(
4049         vnode_t *vp,
4050         int cmd,
4051         ulong_t *valp,
4052         cred_t *cr,
4053         caller_context_t *ct)
4054 {
4055         int     err;
4056 
4057         VOPXID_MAP_CR(vp, cr);
4058 
4059         err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct);
4060         VOPSTATS_UPDATE(vp, pathconf);
4061         return (err);
4062 }
4063 
4064 int
4065 fop_pageio(
4066         vnode_t *vp,
4067         struct page *pp,
4068         u_offset_t io_off,
4069         size_t io_len,
4070         int flags,
4071         cred_t *cr,
4072         caller_context_t *ct)
4073 {
4074         int     err;
4075 
4076         VOPXID_MAP_CR(vp, cr);
4077 
4078         err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct);
4079         VOPSTATS_UPDATE(vp, pageio);
4080         return (err);
4081 }
4082 
4083 int
4084 fop_dumpctl(
4085         vnode_t *vp,
4086         int action,
4087         offset_t *blkp,
4088         caller_context_t *ct)
4089 {
4090         int     err;
4091         err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct);
4092         VOPSTATS_UPDATE(vp, dumpctl);
4093         return (err);
4094 }
4095 
4096 void
4097 fop_dispose(
4098         vnode_t *vp,
4099         page_t *pp,
4100         int flag,
4101         int dn,
4102         cred_t *cr,
4103         caller_context_t *ct)
4104 {
4105         /* Must do stats first since it's possible to lose the vnode */
4106         VOPSTATS_UPDATE(vp, dispose);
4107 
4108         VOPXID_MAP_CR(vp, cr);
4109 
4110         (*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct);
4111 }
4112 
4113 int
4114 fop_setsecattr(
4115         vnode_t *vp,
4116         vsecattr_t *vsap,
4117         int flag,
4118         cred_t *cr,
4119         caller_context_t *ct)
4120 {
4121         int     err;
4122 
4123         VOPXID_MAP_CR(vp, cr);
4124 
4125         /*
4126          * We're only allowed to skip the ACL check iff we used a 32 bit
4127          * ACE mask with VOP_ACCESS() to determine permissions.
4128          */
4129         if ((flag & ATTR_NOACLCHECK) &&
4130             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4131                 return (EINVAL);
4132         }
4133         err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct);
4134         VOPSTATS_UPDATE(vp, setsecattr);
4135         return (err);
4136 }
4137 
4138 int
4139 fop_getsecattr(
4140         vnode_t *vp,
4141         vsecattr_t *vsap,
4142         int flag,
4143         cred_t *cr,
4144         caller_context_t *ct)
4145 {
4146         int     err;
4147 
4148         /*
4149          * We're only allowed to skip the ACL check iff we used a 32 bit
4150          * ACE mask with VOP_ACCESS() to determine permissions.
4151          */
4152         if ((flag & ATTR_NOACLCHECK) &&
4153             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4154                 return (EINVAL);
4155         }
4156 
4157         VOPXID_MAP_CR(vp, cr);
4158 
4159         err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct);
4160         VOPSTATS_UPDATE(vp, getsecattr);
4161         return (err);
4162 }
4163 
4164 int
4165 fop_shrlock(
4166         vnode_t *vp,
4167         int cmd,
4168         struct shrlock *shr,
4169         int flag,
4170         cred_t *cr,
4171         caller_context_t *ct)
4172 {
4173         int     err;
4174 
4175         VOPXID_MAP_CR(vp, cr);
4176 
4177         err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct);
4178         VOPSTATS_UPDATE(vp, shrlock);
4179         return (err);
4180 }
4181 
4182 int
4183 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm,
4184     caller_context_t *ct)
4185 {
4186         int     err;
4187 
4188         err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct);
4189         VOPSTATS_UPDATE(vp, vnevent);
4190         return (err);
4191 }
4192 
4193 int
4194 fop_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *uiop, cred_t *cr,
4195     caller_context_t *ct)
4196 {
4197         int err;
4198 
4199         if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4200                 return (ENOTSUP);
4201         err = (*(vp)->v_op->vop_reqzcbuf)(vp, ioflag, uiop, cr, ct);
4202         VOPSTATS_UPDATE(vp, reqzcbuf);
4203         return (err);
4204 }
4205 
4206 int
4207 fop_retzcbuf(vnode_t *vp, xuio_t *uiop, cred_t *cr, caller_context_t *ct)
4208 {
4209         int err;
4210 
4211         if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4212                 return (ENOTSUP);
4213         err = (*(vp)->v_op->vop_retzcbuf)(vp, uiop, cr, ct);
4214         VOPSTATS_UPDATE(vp, retzcbuf);
4215         return (err);
4216 }
4217 
4218 /*
4219  * Default destructor
4220  *      Needed because NULL destructor means that the key is unused
4221  */
4222 /* ARGSUSED */
4223 void
4224 vsd_defaultdestructor(void *value)
4225 {}
4226 
4227 /*
4228  * Create a key (index into per vnode array)
4229  *      Locks out vsd_create, vsd_destroy, and vsd_free
4230  *      May allocate memory with lock held
4231  */
4232 void
4233 vsd_create(uint_t *keyp, void (*destructor)(void *))
4234 {
4235         int     i;
4236         uint_t  nkeys;
4237 
4238         /*
4239          * if key is allocated, do nothing
4240          */
4241         mutex_enter(&vsd_lock);
4242         if (*keyp) {
4243                 mutex_exit(&vsd_lock);
4244                 return;
4245         }
4246         /*
4247          * find an unused key
4248          */
4249         if (destructor == NULL)
4250                 destructor = vsd_defaultdestructor;
4251 
4252         for (i = 0; i < vsd_nkeys; ++i)
4253                 if (vsd_destructor[i] == NULL)
4254                         break;
4255 
4256         /*
4257          * if no unused keys, increase the size of the destructor array
4258          */
4259         if (i == vsd_nkeys) {
4260                 if ((nkeys = (vsd_nkeys << 1)) == 0)
4261                         nkeys = 1;
4262                 vsd_destructor =
4263                     (void (**)(void *))vsd_realloc((void *)vsd_destructor,
4264                     (size_t)(vsd_nkeys * sizeof (void (*)(void *))),
4265                     (size_t)(nkeys * sizeof (void (*)(void *))));
4266                 vsd_nkeys = nkeys;
4267         }
4268 
4269         /*
4270          * allocate the next available unused key
4271          */
4272         vsd_destructor[i] = destructor;
4273         *keyp = i + 1;
4274 
4275         /* create vsd_list, if it doesn't exist */
4276         if (vsd_list == NULL) {
4277                 vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
4278                 list_create(vsd_list, sizeof (struct vsd_node),
4279                     offsetof(struct vsd_node, vs_nodes));
4280         }
4281 
4282         mutex_exit(&vsd_lock);
4283 }
4284 
4285 /*
4286  * Destroy a key
4287  *
4288  * Assumes that the caller is preventing vsd_set and vsd_get
4289  * Locks out vsd_create, vsd_destroy, and vsd_free
4290  * May free memory with lock held
4291  */
4292 void
4293 vsd_destroy(uint_t *keyp)
4294 {
4295         uint_t key;
4296         struct vsd_node *vsd;
4297 
4298         /*
4299          * protect the key namespace and our destructor lists
4300          */
4301         mutex_enter(&vsd_lock);
4302         key = *keyp;
4303         *keyp = 0;
4304 
4305         ASSERT(key <= vsd_nkeys);
4306 
4307         /*
4308          * if the key is valid
4309          */
4310         if (key != 0) {
4311                 uint_t k = key - 1;
4312                 /*
4313                  * for every vnode with VSD, call key's destructor
4314                  */
4315                 for (vsd = list_head(vsd_list); vsd != NULL;
4316                     vsd = list_next(vsd_list, vsd)) {
4317                         /*
4318                          * no VSD for key in this vnode
4319                          */
4320                         if (key > vsd->vs_nkeys)
4321                                 continue;
4322                         /*
4323                          * call destructor for key
4324                          */
4325                         if (vsd->vs_value[k] && vsd_destructor[k])
4326                                 (*vsd_destructor[k])(vsd->vs_value[k]);
4327                         /*
4328                          * reset value for key
4329                          */
4330                         vsd->vs_value[k] = NULL;
4331                 }
4332                 /*
4333                  * actually free the key (NULL destructor == unused)
4334                  */
4335                 vsd_destructor[k] = NULL;
4336         }
4337 
4338         mutex_exit(&vsd_lock);
4339 }
4340 
4341 /*
4342  * Quickly return the per vnode value that was stored with the specified key
4343  * Assumes the caller is protecting key from vsd_create and vsd_destroy
4344  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4345  */
4346 void *
4347 vsd_get(vnode_t *vp, uint_t key)
4348 {
4349         struct vsd_node *vsd;
4350 
4351         ASSERT(vp != NULL);
4352         ASSERT(mutex_owned(&vp->v_vsd_lock));
4353 
4354         vsd = vp->v_vsd;
4355 
4356         if (key && vsd != NULL && key <= vsd->vs_nkeys)
4357                 return (vsd->vs_value[key - 1]);
4358         return (NULL);
4359 }
4360 
4361 /*
4362  * Set a per vnode value indexed with the specified key
4363  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4364  */
4365 int
4366 vsd_set(vnode_t *vp, uint_t key, void *value)
4367 {
4368         struct vsd_node *vsd;
4369 
4370         ASSERT(vp != NULL);
4371         ASSERT(mutex_owned(&vp->v_vsd_lock));
4372 
4373         if (key == 0)
4374                 return (EINVAL);
4375 
4376         vsd = vp->v_vsd;
4377         if (vsd == NULL)
4378                 vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP);
4379 
4380         /*
4381          * If the vsd was just allocated, vs_nkeys will be 0, so the following
4382          * code won't happen and we will continue down and allocate space for
4383          * the vs_value array.
4384          * If the caller is replacing one value with another, then it is up
4385          * to the caller to free/rele/destroy the previous value (if needed).
4386          */
4387         if (key <= vsd->vs_nkeys) {
4388                 vsd->vs_value[key - 1] = value;
4389                 return (0);
4390         }
4391 
4392         ASSERT(key <= vsd_nkeys);
4393 
4394         if (vsd->vs_nkeys == 0) {
4395                 mutex_enter(&vsd_lock);     /* lock out vsd_destroy() */
4396                 /*
4397                  * Link onto list of all VSD nodes.
4398                  */
4399                 list_insert_head(vsd_list, vsd);
4400                 mutex_exit(&vsd_lock);
4401         }
4402 
4403         /*
4404          * Allocate vnode local storage and set the value for key
4405          */
4406         vsd->vs_value = vsd_realloc(vsd->vs_value,
4407             vsd->vs_nkeys * sizeof (void *),
4408             key * sizeof (void *));
4409         vsd->vs_nkeys = key;
4410         vsd->vs_value[key - 1] = value;
4411 
4412         return (0);
4413 }
4414 
4415 /*
4416  * Called from vn_free() to run the destructor function for each vsd
4417  *      Locks out vsd_create and vsd_destroy
4418  *      Assumes that the destructor *DOES NOT* use vsd
4419  */
4420 void
4421 vsd_free(vnode_t *vp)
4422 {
4423         int i;
4424         struct vsd_node *vsd = vp->v_vsd;
4425 
4426         if (vsd == NULL)
4427                 return;
4428 
4429         if (vsd->vs_nkeys == 0) {
4430                 kmem_free(vsd, sizeof (*vsd));
4431                 vp->v_vsd = NULL;
4432                 return;
4433         }
4434 
4435         /*
4436          * lock out vsd_create and vsd_destroy, call
4437          * the destructor, and mark the value as destroyed.
4438          */
4439         mutex_enter(&vsd_lock);
4440 
4441         for (i = 0; i < vsd->vs_nkeys; i++) {
4442                 if (vsd->vs_value[i] && vsd_destructor[i])
4443                         (*vsd_destructor[i])(vsd->vs_value[i]);
4444                 vsd->vs_value[i] = NULL;
4445         }
4446 
4447         /*
4448          * remove from linked list of VSD nodes
4449          */
4450         list_remove(vsd_list, vsd);
4451 
4452         mutex_exit(&vsd_lock);
4453 
4454         /*
4455          * free up the VSD
4456          */
4457         kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *));
4458         kmem_free(vsd, sizeof (struct vsd_node));
4459         vp->v_vsd = NULL;
4460 }
4461 
4462 /*
4463  * realloc
4464  */
4465 static void *
4466 vsd_realloc(void *old, size_t osize, size_t nsize)
4467 {
4468         void *new;
4469 
4470         new = kmem_zalloc(nsize, KM_SLEEP);
4471         if (old) {
4472                 bcopy(old, new, osize);
4473                 kmem_free(old, osize);
4474         }
4475         return (new);
4476 }
4477 
4478 /*
4479  * Setup the extensible system attribute for creating a reparse point.
4480  * The symlink data 'target' is validated for proper format of a reparse
4481  * string and a check also made to make sure the symlink data does not
4482  * point to an existing file.
4483  *
4484  * return 0 if ok else -1.
4485  */
4486 static int
4487 fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr)
4488 {
4489         xoptattr_t *xoap;
4490 
4491         if ((!target) || (!vap) || (!xvattr))
4492                 return (-1);
4493 
4494         /* validate reparse string */
4495         if (reparse_validate((const char *)target))
4496                 return (-1);
4497 
4498         xva_init(xvattr);
4499         xvattr->xva_vattr = *vap;
4500         xvattr->xva_vattr.va_mask |= AT_XVATTR;
4501         xoap = xva_getxoptattr(xvattr);
4502         ASSERT(xoap);
4503         XVA_SET_REQ(xvattr, XAT_REPARSE);
4504         xoap->xoa_reparse = 1;
4505 
4506         return (0);
4507 }
4508 
4509 /*
4510  * Function to check whether a symlink is a reparse point.
4511  * Return B_TRUE if it is a reparse point, else return B_FALSE
4512  */
4513 boolean_t
4514 vn_is_reparse(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4515 {
4516         xvattr_t xvattr;
4517         xoptattr_t *xoap;
4518 
4519         if ((vp->v_type != VLNK) ||
4520             !(vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR)))
4521                 return (B_FALSE);
4522 
4523         xva_init(&xvattr);
4524         xoap = xva_getxoptattr(&xvattr);
4525         ASSERT(xoap);
4526         XVA_SET_REQ(&xvattr, XAT_REPARSE);
4527 
4528         if (VOP_GETATTR(vp, &xvattr.xva_vattr, 0, cr, ct))
4529                 return (B_FALSE);
4530 
4531         if ((!(xvattr.xva_vattr.va_mask & AT_XVATTR)) ||
4532             (!(XVA_ISSET_RTN(&xvattr, XAT_REPARSE))))
4533                 return (B_FALSE);
4534 
4535         return (xoap->xoa_reparse ? B_TRUE : B_FALSE);
4536 }