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  * Copyright 2020 Joyent, Inc.
  25  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  26  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  27  * Copyright 2021 OmniOS Community Edition (OmniOSce) Association.
  28  */
  29 
  30 /*      Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
  31 /*        All Rights Reserved   */
  32 
  33 /*
  34  * University Copyright- Copyright (c) 1982, 1986, 1988
  35  * The Regents of the University of California
  36  * All Rights Reserved
  37  *
  38  * University Acknowledgment- Portions of this document are derived from
  39  * software developed by the University of California, Berkeley, and its
  40  * contributors.
  41  */
  42 
  43 #include <sys/types.h>
  44 #include <sys/param.h>
  45 #include <sys/t_lock.h>
  46 #include <sys/errno.h>
  47 #include <sys/cred.h>
  48 #include <sys/user.h>
  49 #include <sys/uio.h>
  50 #include <sys/file.h>
  51 #include <sys/pathname.h>
  52 #include <sys/vfs.h>
  53 #include <sys/vfs_opreg.h>
  54 #include <sys/vnode.h>
  55 #include <sys/filio.h>
  56 #include <sys/rwstlock.h>
  57 #include <sys/fem.h>
  58 #include <sys/stat.h>
  59 #include <sys/mode.h>
  60 #include <sys/conf.h>
  61 #include <sys/sysmacros.h>
  62 #include <sys/cmn_err.h>
  63 #include <sys/systm.h>
  64 #include <sys/kmem.h>
  65 #include <sys/debug.h>
  66 #include <c2/audit.h>
  67 #include <sys/acl.h>
  68 #include <sys/nbmlock.h>
  69 #include <sys/fcntl.h>
  70 #include <fs/fs_subr.h>
  71 #include <sys/taskq.h>
  72 #include <fs/fs_reparse.h>
  73 #include <sys/time.h>
  74 #include <sys/sdt.h>
  75 
  76 /* Determine if this vnode is a file that is read-only */
  77 #define ISROFILE(vp)    \
  78         ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \
  79             (vp)->v_type != VFIFO && vn_is_readonly(vp))
  80 
  81 /* Tunable via /etc/system; used only by admin/install */
  82 int nfs_global_client_only;
  83 
  84 /*
  85  * Array of vopstats_t for per-FS-type vopstats.  This array has the same
  86  * number of entries as and parallel to the vfssw table.  (Arguably, it could
  87  * be part of the vfssw table.)  Once it's initialized, it's accessed using
  88  * the same fstype index that is used to index into the vfssw table.
  89  */
  90 vopstats_t **vopstats_fstype;
  91 
  92 /* vopstats initialization template used for fast initialization via bcopy() */
  93 static vopstats_t *vs_templatep;
  94 
  95 /* Kmem cache handle for vsk_anchor_t allocations */
  96 kmem_cache_t *vsk_anchor_cache;
  97 
  98 /* file events cleanup routine */
  99 extern void free_fopdata(vnode_t *);
 100 
 101 /*
 102  * Root of AVL tree for the kstats associated with vopstats.  Lock protects
 103  * updates to vsktat_tree.
 104  */
 105 avl_tree_t      vskstat_tree;
 106 kmutex_t        vskstat_tree_lock;
 107 
 108 /* Global variable which enables/disables the vopstats collection */
 109 int vopstats_enabled = 1;
 110 
 111 /* Global used for empty/invalid v_path */
 112 char *vn_vpath_empty = "";
 113 
 114 /*
 115  * forward declarations for internal vnode specific data (vsd)
 116  */
 117 static void *vsd_realloc(void *, size_t, size_t);
 118 
 119 /*
 120  * forward declarations for reparse point functions
 121  */
 122 static int fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr);
 123 
 124 /*
 125  * VSD -- VNODE SPECIFIC DATA
 126  * The v_data pointer is typically used by a file system to store a
 127  * pointer to the file system's private node (e.g. ufs inode, nfs rnode).
 128  * However, there are times when additional project private data needs
 129  * to be stored separately from the data (node) pointed to by v_data.
 130  * This additional data could be stored by the file system itself or
 131  * by a completely different kernel entity.  VSD provides a way for
 132  * callers to obtain a key and store a pointer to private data associated
 133  * with a vnode.
 134  *
 135  * Callers are responsible for protecting the vsd by holding v_vsd_lock
 136  * for calls to vsd_set() and vsd_get().
 137  */
 138 
 139 /*
 140  * vsd_lock protects:
 141  *   vsd_nkeys - creation and deletion of vsd keys
 142  *   vsd_list - insertion and deletion of vsd_node in the vsd_list
 143  *   vsd_destructor - adding and removing destructors to the list
 144  */
 145 static kmutex_t         vsd_lock;
 146 static uint_t           vsd_nkeys;       /* size of destructor array */
 147 /* list of vsd_node's */
 148 static list_t *vsd_list = NULL;
 149 /* per-key destructor funcs */
 150 static void             (**vsd_destructor)(void *);
 151 
 152 /*
 153  * The following is the common set of actions needed to update the
 154  * vopstats structure from a vnode op.  Both VOPSTATS_UPDATE() and
 155  * VOPSTATS_UPDATE_IO() do almost the same thing, except for the
 156  * recording of the bytes transferred.  Since the code is similar
 157  * but small, it is nearly a duplicate.  Consequently any changes
 158  * to one may need to be reflected in the other.
 159  * Rundown of the variables:
 160  * vp - Pointer to the vnode
 161  * counter - Partial name structure member to update in vopstats for counts
 162  * bytecounter - Partial name structure member to update in vopstats for bytes
 163  * bytesval - Value to update in vopstats for bytes
 164  * fstype - Index into vsanchor_fstype[], same as index into vfssw[]
 165  * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i])
 166  */
 167 
 168 #define VOPSTATS_UPDATE(vp, counter) {                                  \
 169         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 170         if (vfsp && vfsp->vfs_implp &&                                       \
 171             (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {   \
 172                 vopstats_t *vsp = &vfsp->vfs_vopstats;                   \
 173                 uint64_t *stataddr = &(vsp->n##counter.value.ui64);      \
 174                 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
 175                     size_t, uint64_t *);                                \
 176                 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr);     \
 177                 (*stataddr)++;                                          \
 178                 if ((vsp = vfsp->vfs_fstypevsp) != NULL) {           \
 179                         vsp->n##counter.value.ui64++;                        \
 180                 }                                                       \
 181         }                                                               \
 182 }
 183 
 184 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) {        \
 185         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 186         if (vfsp && vfsp->vfs_implp &&                                       \
 187             (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) {   \
 188                 vopstats_t *vsp = &vfsp->vfs_vopstats;                   \
 189                 uint64_t *stataddr = &(vsp->n##counter.value.ui64);      \
 190                 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
 191                     size_t, uint64_t *);                                \
 192                 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \
 193                 (*stataddr)++;                                          \
 194                 vsp->bytecounter.value.ui64 += bytesval;             \
 195                 if ((vsp = vfsp->vfs_fstypevsp) != NULL) {           \
 196                         vsp->n##counter.value.ui64++;                        \
 197                         vsp->bytecounter.value.ui64 += bytesval;     \
 198                 }                                                       \
 199         }                                                               \
 200 }
 201 
 202 /*
 203  * If the filesystem does not support XIDs map credential
 204  * If the vfsp is NULL, perhaps we should also map?
 205  */
 206 #define VOPXID_MAP_CR(vp, cr)   {                                       \
 207         vfs_t *vfsp = (vp)->v_vfsp;                                  \
 208         if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0)             \
 209                 cr = crgetmapped(cr);                                   \
 210         }
 211 
 212 #define VOP_LATENCY_10MS        10000000
 213 #define VOP_LATENCY_100MS       100000000
 214 #define VOP_LATENCY_1S          1000000000
 215 #define VOP_LATENCY_10S         10000000000
 216 
 217 /*
 218  * Convert stat(2) formats to vnode types and vice versa.  (Knows about
 219  * numerical order of S_IFMT and vnode types.)
 220  */
 221 enum vtype iftovt_tab[] = {
 222         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
 223         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
 224 };
 225 
 226 ushort_t vttoif_tab[] = {
 227         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO,
 228         S_IFDOOR, 0, S_IFSOCK, S_IFPORT, 0
 229 };
 230 
 231 /*
 232  * The system vnode cache.
 233  */
 234 
 235 kmem_cache_t *vn_cache;
 236 
 237 
 238 /*
 239  * Vnode operations vector.
 240  */
 241 
 242 static const fs_operation_trans_def_t vn_ops_table[] = {
 243         VOPNAME_OPEN, offsetof(struct vnodeops, vop_open),
 244             fs_nosys, fs_nosys,
 245 
 246         VOPNAME_CLOSE, offsetof(struct vnodeops, vop_close),
 247             fs_nosys, fs_nosys,
 248 
 249         VOPNAME_READ, offsetof(struct vnodeops, vop_read),
 250             fs_nosys, fs_nosys,
 251 
 252         VOPNAME_WRITE, offsetof(struct vnodeops, vop_write),
 253             fs_nosys, fs_nosys,
 254 
 255         VOPNAME_IOCTL, offsetof(struct vnodeops, vop_ioctl),
 256             fs_nosys, fs_nosys,
 257 
 258         VOPNAME_SETFL, offsetof(struct vnodeops, vop_setfl),
 259             fs_setfl, fs_nosys,
 260 
 261         VOPNAME_GETATTR, offsetof(struct vnodeops, vop_getattr),
 262             fs_nosys, fs_nosys,
 263 
 264         VOPNAME_SETATTR, offsetof(struct vnodeops, vop_setattr),
 265             fs_nosys, fs_nosys,
 266 
 267         VOPNAME_ACCESS, offsetof(struct vnodeops, vop_access),
 268             fs_nosys, fs_nosys,
 269 
 270         VOPNAME_LOOKUP, offsetof(struct vnodeops, vop_lookup),
 271             fs_nosys, fs_nosys,
 272 
 273         VOPNAME_CREATE, offsetof(struct vnodeops, vop_create),
 274             fs_nosys, fs_nosys,
 275 
 276         VOPNAME_REMOVE, offsetof(struct vnodeops, vop_remove),
 277             fs_nosys, fs_nosys,
 278 
 279         VOPNAME_LINK, offsetof(struct vnodeops, vop_link),
 280             fs_nosys, fs_nosys,
 281 
 282         VOPNAME_RENAME, offsetof(struct vnodeops, vop_rename),
 283             fs_nosys, fs_nosys,
 284 
 285         VOPNAME_MKDIR, offsetof(struct vnodeops, vop_mkdir),
 286             fs_nosys, fs_nosys,
 287 
 288         VOPNAME_RMDIR, offsetof(struct vnodeops, vop_rmdir),
 289             fs_nosys, fs_nosys,
 290 
 291         VOPNAME_READDIR, offsetof(struct vnodeops, vop_readdir),
 292             fs_nosys, fs_nosys,
 293 
 294         VOPNAME_SYMLINK, offsetof(struct vnodeops, vop_symlink),
 295             fs_nosys, fs_nosys,
 296 
 297         VOPNAME_READLINK, offsetof(struct vnodeops, vop_readlink),
 298             fs_nosys, fs_nosys,
 299 
 300         VOPNAME_FSYNC, offsetof(struct vnodeops, vop_fsync),
 301             fs_nosys, fs_nosys,
 302 
 303         VOPNAME_INACTIVE, offsetof(struct vnodeops, vop_inactive),
 304             fs_nosys, fs_nosys,
 305 
 306         VOPNAME_FID, offsetof(struct vnodeops, vop_fid),
 307             fs_nosys, fs_nosys,
 308 
 309         VOPNAME_RWLOCK, offsetof(struct vnodeops, vop_rwlock),
 310             fs_rwlock, fs_rwlock,
 311 
 312         VOPNAME_RWUNLOCK, offsetof(struct vnodeops, vop_rwunlock),
 313             (fs_generic_func_p)(uintptr_t)fs_rwunlock,
 314             (fs_generic_func_p)(uintptr_t)fs_rwunlock,  /* no errors allowed */
 315 
 316         VOPNAME_SEEK, offsetof(struct vnodeops, vop_seek),
 317             fs_nosys, fs_nosys,
 318 
 319         VOPNAME_CMP, offsetof(struct vnodeops, vop_cmp),
 320             fs_cmp, fs_cmp,             /* no errors allowed */
 321 
 322         VOPNAME_FRLOCK, offsetof(struct vnodeops, vop_frlock),
 323             fs_frlock, fs_nosys,
 324 
 325         VOPNAME_SPACE, offsetof(struct vnodeops, vop_space),
 326             fs_nosys, fs_nosys,
 327 
 328         VOPNAME_REALVP, offsetof(struct vnodeops, vop_realvp),
 329             fs_nosys, fs_nosys,
 330 
 331         VOPNAME_GETPAGE, offsetof(struct vnodeops, vop_getpage),
 332             fs_nosys, fs_nosys,
 333 
 334         VOPNAME_PUTPAGE, offsetof(struct vnodeops, vop_putpage),
 335             fs_nosys, fs_nosys,
 336 
 337         VOPNAME_MAP, offsetof(struct vnodeops, vop_map),
 338             (fs_generic_func_p) fs_nosys_map,
 339             (fs_generic_func_p) fs_nosys_map,
 340 
 341         VOPNAME_ADDMAP, offsetof(struct vnodeops, vop_addmap),
 342             (fs_generic_func_p) fs_nosys_addmap,
 343             (fs_generic_func_p) fs_nosys_addmap,
 344 
 345         VOPNAME_DELMAP, offsetof(struct vnodeops, vop_delmap),
 346             fs_nosys, fs_nosys,
 347 
 348         VOPNAME_POLL, offsetof(struct vnodeops, vop_poll),
 349             (fs_generic_func_p) fs_poll, (fs_generic_func_p) fs_nosys_poll,
 350 
 351         VOPNAME_DUMP, offsetof(struct vnodeops, vop_dump),
 352             fs_nosys, fs_nosys,
 353 
 354         VOPNAME_PATHCONF, offsetof(struct vnodeops, vop_pathconf),
 355             fs_pathconf, fs_nosys,
 356 
 357         VOPNAME_PAGEIO, offsetof(struct vnodeops, vop_pageio),
 358             fs_nosys, fs_nosys,
 359 
 360         VOPNAME_DUMPCTL, offsetof(struct vnodeops, vop_dumpctl),
 361             fs_nosys, fs_nosys,
 362 
 363         VOPNAME_DISPOSE, offsetof(struct vnodeops, vop_dispose),
 364             (fs_generic_func_p)(uintptr_t)fs_dispose,
 365             (fs_generic_func_p)(uintptr_t)fs_nodispose,
 366 
 367         VOPNAME_SETSECATTR, offsetof(struct vnodeops, vop_setsecattr),
 368             fs_nosys, fs_nosys,
 369 
 370         VOPNAME_GETSECATTR, offsetof(struct vnodeops, vop_getsecattr),
 371             fs_fab_acl, fs_nosys,
 372 
 373         VOPNAME_SHRLOCK, offsetof(struct vnodeops, vop_shrlock),
 374             fs_shrlock, fs_nosys,
 375 
 376         VOPNAME_VNEVENT, offsetof(struct vnodeops, vop_vnevent),
 377             (fs_generic_func_p) fs_vnevent_nosupport,
 378             (fs_generic_func_p) fs_vnevent_nosupport,
 379 
 380         VOPNAME_REQZCBUF, offsetof(struct vnodeops, vop_reqzcbuf),
 381             fs_nosys, fs_nosys,
 382 
 383         VOPNAME_RETZCBUF, offsetof(struct vnodeops, vop_retzcbuf),
 384             fs_nosys, fs_nosys,
 385 
 386         NULL, 0, NULL, NULL
 387 };
 388 
 389 /* Extensible attribute (xva) routines. */
 390 
 391 /*
 392  * Zero out the structure, set the size of the requested/returned bitmaps,
 393  * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer
 394  * to the returned attributes array.
 395  */
 396 void
 397 xva_init(xvattr_t *xvap)
 398 {
 399         bzero(xvap, sizeof (xvattr_t));
 400         xvap->xva_mapsize = XVA_MAPSIZE;
 401         xvap->xva_magic = XVA_MAGIC;
 402         xvap->xva_vattr.va_mask = AT_XVATTR;
 403         xvap->xva_rtnattrmapp = &(xvap->xva_rtnattrmap)[0];
 404 }
 405 
 406 /*
 407  * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t
 408  * structure.  Otherwise, returns NULL.
 409  */
 410 xoptattr_t *
 411 xva_getxoptattr(xvattr_t *xvap)
 412 {
 413         xoptattr_t *xoap = NULL;
 414         if (xvap->xva_vattr.va_mask & AT_XVATTR)
 415                 xoap = &xvap->xva_xoptattrs;
 416         return (xoap);
 417 }
 418 
 419 /*
 420  * Used by the AVL routines to compare two vsk_anchor_t structures in the tree.
 421  * We use the f_fsid reported by VFS_STATVFS() since we use that for the
 422  * kstat name.
 423  */
 424 static int
 425 vska_compar(const void *n1, const void *n2)
 426 {
 427         int ret;
 428         ulong_t p1 = ((vsk_anchor_t *)n1)->vsk_fsid;
 429         ulong_t p2 = ((vsk_anchor_t *)n2)->vsk_fsid;
 430 
 431         if (p1 < p2) {
 432                 ret = -1;
 433         } else if (p1 > p2) {
 434                 ret = 1;
 435         } else {
 436                 ret = 0;
 437         }
 438 
 439         return (ret);
 440 }
 441 
 442 /*
 443  * Used to create a single template which will be bcopy()ed to a newly
 444  * allocated vsanchor_combo_t structure in new_vsanchor(), below.
 445  */
 446 static vopstats_t *
 447 create_vopstats_template()
 448 {
 449         vopstats_t              *vsp;
 450 
 451         vsp = kmem_alloc(sizeof (vopstats_t), KM_SLEEP);
 452         bzero(vsp, sizeof (*vsp));      /* Start fresh */
 453 
 454         /* VOP_OPEN */
 455         kstat_named_init(&vsp->nopen, "nopen", KSTAT_DATA_UINT64);
 456         /* VOP_CLOSE */
 457         kstat_named_init(&vsp->nclose, "nclose", KSTAT_DATA_UINT64);
 458         /* VOP_READ I/O */
 459         kstat_named_init(&vsp->nread, "nread", KSTAT_DATA_UINT64);
 460         kstat_named_init(&vsp->read_bytes, "read_bytes", KSTAT_DATA_UINT64);
 461         /* VOP_WRITE I/O */
 462         kstat_named_init(&vsp->nwrite, "nwrite", KSTAT_DATA_UINT64);
 463         kstat_named_init(&vsp->write_bytes, "write_bytes", KSTAT_DATA_UINT64);
 464         /* VOP_IOCTL */
 465         kstat_named_init(&vsp->nioctl, "nioctl", KSTAT_DATA_UINT64);
 466         /* VOP_SETFL */
 467         kstat_named_init(&vsp->nsetfl, "nsetfl", KSTAT_DATA_UINT64);
 468         /* VOP_GETATTR */
 469         kstat_named_init(&vsp->ngetattr, "ngetattr", KSTAT_DATA_UINT64);
 470         /* VOP_SETATTR */
 471         kstat_named_init(&vsp->nsetattr, "nsetattr", KSTAT_DATA_UINT64);
 472         /* VOP_ACCESS */
 473         kstat_named_init(&vsp->naccess, "naccess", KSTAT_DATA_UINT64);
 474         /* VOP_LOOKUP */
 475         kstat_named_init(&vsp->nlookup, "nlookup", KSTAT_DATA_UINT64);
 476         /* VOP_CREATE */
 477         kstat_named_init(&vsp->ncreate, "ncreate", KSTAT_DATA_UINT64);
 478         /* VOP_REMOVE */
 479         kstat_named_init(&vsp->nremove, "nremove", KSTAT_DATA_UINT64);
 480         /* VOP_LINK */
 481         kstat_named_init(&vsp->nlink, "nlink", KSTAT_DATA_UINT64);
 482         /* VOP_RENAME */
 483         kstat_named_init(&vsp->nrename, "nrename", KSTAT_DATA_UINT64);
 484         /* VOP_MKDIR */
 485         kstat_named_init(&vsp->nmkdir, "nmkdir", KSTAT_DATA_UINT64);
 486         /* VOP_RMDIR */
 487         kstat_named_init(&vsp->nrmdir, "nrmdir", KSTAT_DATA_UINT64);
 488         /* VOP_READDIR I/O */
 489         kstat_named_init(&vsp->nreaddir, "nreaddir", KSTAT_DATA_UINT64);
 490         kstat_named_init(&vsp->readdir_bytes, "readdir_bytes",
 491             KSTAT_DATA_UINT64);
 492         /* VOP_SYMLINK */
 493         kstat_named_init(&vsp->nsymlink, "nsymlink", KSTAT_DATA_UINT64);
 494         /* VOP_READLINK */
 495         kstat_named_init(&vsp->nreadlink, "nreadlink", KSTAT_DATA_UINT64);
 496         /* VOP_FSYNC */
 497         kstat_named_init(&vsp->nfsync, "nfsync", KSTAT_DATA_UINT64);
 498         /* VOP_INACTIVE */
 499         kstat_named_init(&vsp->ninactive, "ninactive", KSTAT_DATA_UINT64);
 500         /* VOP_FID */
 501         kstat_named_init(&vsp->nfid, "nfid", KSTAT_DATA_UINT64);
 502         /* VOP_RWLOCK */
 503         kstat_named_init(&vsp->nrwlock, "nrwlock", KSTAT_DATA_UINT64);
 504         /* VOP_RWUNLOCK */
 505         kstat_named_init(&vsp->nrwunlock, "nrwunlock", KSTAT_DATA_UINT64);
 506         /* VOP_SEEK */
 507         kstat_named_init(&vsp->nseek, "nseek", KSTAT_DATA_UINT64);
 508         /* VOP_CMP */
 509         kstat_named_init(&vsp->ncmp, "ncmp", KSTAT_DATA_UINT64);
 510         /* VOP_FRLOCK */
 511         kstat_named_init(&vsp->nfrlock, "nfrlock", KSTAT_DATA_UINT64);
 512         /* VOP_SPACE */
 513         kstat_named_init(&vsp->nspace, "nspace", KSTAT_DATA_UINT64);
 514         /* VOP_REALVP */
 515         kstat_named_init(&vsp->nrealvp, "nrealvp", KSTAT_DATA_UINT64);
 516         /* VOP_GETPAGE */
 517         kstat_named_init(&vsp->ngetpage, "ngetpage", KSTAT_DATA_UINT64);
 518         /* VOP_PUTPAGE */
 519         kstat_named_init(&vsp->nputpage, "nputpage", KSTAT_DATA_UINT64);
 520         /* VOP_MAP */
 521         kstat_named_init(&vsp->nmap, "nmap", KSTAT_DATA_UINT64);
 522         /* VOP_ADDMAP */
 523         kstat_named_init(&vsp->naddmap, "naddmap", KSTAT_DATA_UINT64);
 524         /* VOP_DELMAP */
 525         kstat_named_init(&vsp->ndelmap, "ndelmap", KSTAT_DATA_UINT64);
 526         /* VOP_POLL */
 527         kstat_named_init(&vsp->npoll, "npoll", KSTAT_DATA_UINT64);
 528         /* VOP_DUMP */
 529         kstat_named_init(&vsp->ndump, "ndump", KSTAT_DATA_UINT64);
 530         /* VOP_PATHCONF */
 531         kstat_named_init(&vsp->npathconf, "npathconf", KSTAT_DATA_UINT64);
 532         /* VOP_PAGEIO */
 533         kstat_named_init(&vsp->npageio, "npageio", KSTAT_DATA_UINT64);
 534         /* VOP_DUMPCTL */
 535         kstat_named_init(&vsp->ndumpctl, "ndumpctl", KSTAT_DATA_UINT64);
 536         /* VOP_DISPOSE */
 537         kstat_named_init(&vsp->ndispose, "ndispose", KSTAT_DATA_UINT64);
 538         /* VOP_SETSECATTR */
 539         kstat_named_init(&vsp->nsetsecattr, "nsetsecattr", KSTAT_DATA_UINT64);
 540         /* VOP_GETSECATTR */
 541         kstat_named_init(&vsp->ngetsecattr, "ngetsecattr", KSTAT_DATA_UINT64);
 542         /* VOP_SHRLOCK */
 543         kstat_named_init(&vsp->nshrlock, "nshrlock", KSTAT_DATA_UINT64);
 544         /* VOP_VNEVENT */
 545         kstat_named_init(&vsp->nvnevent, "nvnevent", KSTAT_DATA_UINT64);
 546         /* VOP_REQZCBUF */
 547         kstat_named_init(&vsp->nreqzcbuf, "nreqzcbuf", KSTAT_DATA_UINT64);
 548         /* VOP_RETZCBUF */
 549         kstat_named_init(&vsp->nretzcbuf, "nretzcbuf", KSTAT_DATA_UINT64);
 550 
 551         return (vsp);
 552 }
 553 
 554 /*
 555  * Creates a kstat structure associated with a vopstats structure.
 556  */
 557 kstat_t *
 558 new_vskstat(char *ksname, vopstats_t *vsp)
 559 {
 560         kstat_t         *ksp;
 561 
 562         if (!vopstats_enabled) {
 563                 return (NULL);
 564         }
 565 
 566         ksp = kstat_create("unix", 0, ksname, "misc", KSTAT_TYPE_NAMED,
 567             sizeof (vopstats_t)/sizeof (kstat_named_t),
 568             KSTAT_FLAG_VIRTUAL|KSTAT_FLAG_WRITABLE);
 569         if (ksp) {
 570                 ksp->ks_data = vsp;
 571                 kstat_install(ksp);
 572         }
 573 
 574         return (ksp);
 575 }
 576 
 577 /*
 578  * Called from vfsinit() to initialize the support mechanisms for vopstats
 579  */
 580 void
 581 vopstats_startup()
 582 {
 583         if (!vopstats_enabled)
 584                 return;
 585 
 586         /*
 587          * Creates the AVL tree which holds per-vfs vopstat anchors.  This
 588          * is necessary since we need to check if a kstat exists before we
 589          * attempt to create it.  Also, initialize its lock.
 590          */
 591         avl_create(&vskstat_tree, vska_compar, sizeof (vsk_anchor_t),
 592             offsetof(vsk_anchor_t, vsk_node));
 593         mutex_init(&vskstat_tree_lock, NULL, MUTEX_DEFAULT, NULL);
 594 
 595         vsk_anchor_cache = kmem_cache_create("vsk_anchor_cache",
 596             sizeof (vsk_anchor_t), sizeof (uintptr_t), NULL, NULL, NULL,
 597             NULL, NULL, 0);
 598 
 599         /*
 600          * Set up the array of pointers for the vopstats-by-FS-type.
 601          * The entries will be allocated/initialized as each file system
 602          * goes through modload/mod_installfs.
 603          */
 604         vopstats_fstype = (vopstats_t **)kmem_zalloc(
 605             (sizeof (vopstats_t *) * nfstype), KM_SLEEP);
 606 
 607         /* Set up the global vopstats initialization template */
 608         vs_templatep = create_vopstats_template();
 609 }
 610 
 611 /*
 612  * We need to have the all of the counters zeroed.
 613  * The initialization of the vopstats_t includes on the order of
 614  * 50 calls to kstat_named_init().  Rather that do that on every call,
 615  * we do it once in a template (vs_templatep) then bcopy it over.
 616  */
 617 void
 618 initialize_vopstats(vopstats_t *vsp)
 619 {
 620         if (vsp == NULL)
 621                 return;
 622 
 623         bcopy(vs_templatep, vsp, sizeof (vopstats_t));
 624 }
 625 
 626 /*
 627  * If possible, determine which vopstats by fstype to use and
 628  * return a pointer to the caller.
 629  */
 630 vopstats_t *
 631 get_fstype_vopstats(vfs_t *vfsp, struct vfssw *vswp)
 632 {
 633         int             fstype = 0;     /* Index into vfssw[] */
 634         vopstats_t      *vsp = NULL;
 635 
 636         if (vfsp == NULL || (vfsp->vfs_flag & VFS_STATS) == 0 ||
 637             !vopstats_enabled)
 638                 return (NULL);
 639         /*
 640          * Set up the fstype.  We go to so much trouble because all versions
 641          * of NFS use the same fstype in their vfs even though they have
 642          * distinct entries in the vfssw[] table.
 643          * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry.
 644          */
 645         if (vswp) {
 646                 fstype = vswp - vfssw;  /* Gets us the index */
 647         } else {
 648                 fstype = vfsp->vfs_fstype;
 649         }
 650 
 651         /*
 652          * Point to the per-fstype vopstats. The only valid values are
 653          * non-zero positive values less than the number of vfssw[] table
 654          * entries.
 655          */
 656         if (fstype > 0 && fstype < nfstype) {
 657                 vsp = vopstats_fstype[fstype];
 658         }
 659 
 660         return (vsp);
 661 }
 662 
 663 /*
 664  * Generate a kstat name, create the kstat structure, and allocate a
 665  * vsk_anchor_t to hold it together.  Return the pointer to the vsk_anchor_t
 666  * to the caller.  This must only be called from a mount.
 667  */
 668 vsk_anchor_t *
 669 get_vskstat_anchor(vfs_t *vfsp)
 670 {
 671         char            kstatstr[KSTAT_STRLEN]; /* kstat name for vopstats */
 672         statvfs64_t     statvfsbuf;             /* Needed to find f_fsid */
 673         vsk_anchor_t    *vskp = NULL;           /* vfs <--> kstat anchor */
 674         kstat_t         *ksp;                   /* Ptr to new kstat */
 675         avl_index_t     where;                  /* Location in the AVL tree */
 676 
 677         if (vfsp == NULL || vfsp->vfs_implp == NULL ||
 678             (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
 679                 return (NULL);
 680 
 681         /* Need to get the fsid to build a kstat name */
 682         if (VFS_STATVFS(vfsp, &statvfsbuf) == 0) {
 683                 /* Create a name for our kstats based on fsid */
 684                 (void) snprintf(kstatstr, KSTAT_STRLEN, "%s%lx",
 685                     VOPSTATS_STR, statvfsbuf.f_fsid);
 686 
 687                 /* Allocate and initialize the vsk_anchor_t */
 688                 vskp = kmem_cache_alloc(vsk_anchor_cache, KM_SLEEP);
 689                 bzero(vskp, sizeof (*vskp));
 690                 vskp->vsk_fsid = statvfsbuf.f_fsid;
 691 
 692                 mutex_enter(&vskstat_tree_lock);
 693                 if (avl_find(&vskstat_tree, vskp, &where) == NULL) {
 694                         avl_insert(&vskstat_tree, vskp, where);
 695                         mutex_exit(&vskstat_tree_lock);
 696 
 697                         /*
 698                          * Now that we've got the anchor in the AVL
 699                          * tree, we can create the kstat.
 700                          */
 701                         ksp = new_vskstat(kstatstr, &vfsp->vfs_vopstats);
 702                         if (ksp) {
 703                                 vskp->vsk_ksp = ksp;
 704                         }
 705                 } else {
 706                         /* Oops, found one! Release memory and lock. */
 707                         mutex_exit(&vskstat_tree_lock);
 708                         kmem_cache_free(vsk_anchor_cache, vskp);
 709                         vskp = NULL;
 710                 }
 711         }
 712         return (vskp);
 713 }
 714 
 715 /*
 716  * We're in the process of tearing down the vfs and need to cleanup
 717  * the data structures associated with the vopstats. Must only be called
 718  * from dounmount().
 719  */
 720 void
 721 teardown_vopstats(vfs_t *vfsp)
 722 {
 723         vsk_anchor_t    *vskap;
 724         avl_index_t     where;
 725 
 726         if (vfsp == NULL || vfsp->vfs_implp == NULL ||
 727             (vfsp->vfs_flag & VFS_STATS) == 0 || !vopstats_enabled)
 728                 return;
 729 
 730         /* This is a safe check since VFS_STATS must be set (see above) */
 731         if ((vskap = vfsp->vfs_vskap) == NULL)
 732                 return;
 733 
 734         /* Whack the pointer right away */
 735         vfsp->vfs_vskap = NULL;
 736 
 737         /* Lock the tree, remove the node, and delete the kstat */
 738         mutex_enter(&vskstat_tree_lock);
 739         if (avl_find(&vskstat_tree, vskap, &where)) {
 740                 avl_remove(&vskstat_tree, vskap);
 741         }
 742 
 743         if (vskap->vsk_ksp) {
 744                 kstat_delete(vskap->vsk_ksp);
 745         }
 746         mutex_exit(&vskstat_tree_lock);
 747 
 748         kmem_cache_free(vsk_anchor_cache, vskap);
 749 }
 750 
 751 /*
 752  * Read or write a vnode.  Called from kernel code.
 753  */
 754 int
 755 vn_rdwr(
 756         enum uio_rw rw,
 757         struct vnode *vp,
 758         caddr_t base,
 759         ssize_t len,
 760         offset_t offset,
 761         enum uio_seg seg,
 762         int ioflag,
 763         rlim64_t ulimit,        /* meaningful only if rw is UIO_WRITE */
 764         cred_t *cr,
 765         ssize_t *residp)
 766 {
 767         struct uio uio;
 768         struct iovec iov;
 769         int error;
 770         int in_crit = 0;
 771 
 772         if (rw == UIO_WRITE && ISROFILE(vp))
 773                 return (EROFS);
 774 
 775         if (len < 0)
 776                 return (EIO);
 777 
 778         VOPXID_MAP_CR(vp, cr);
 779 
 780         iov.iov_base = base;
 781         iov.iov_len = len;
 782         uio.uio_iov = &iov;
 783         uio.uio_iovcnt = 1;
 784         uio.uio_loffset = offset;
 785         uio.uio_segflg = (short)seg;
 786         uio.uio_resid = len;
 787         uio.uio_llimit = ulimit;
 788 
 789         /*
 790          * We have to enter the critical region before calling VOP_RWLOCK
 791          * to avoid a deadlock with ufs.
 792          */
 793         if (nbl_need_check(vp)) {
 794                 int svmand;
 795 
 796                 nbl_start_crit(vp, RW_READER);
 797                 in_crit = 1;
 798                 error = nbl_svmand(vp, cr, &svmand);
 799                 if (error != 0)
 800                         goto done;
 801                 if (nbl_conflict(vp, rw == UIO_WRITE ? NBL_WRITE : NBL_READ,
 802                     uio.uio_offset, uio.uio_resid, svmand, NULL)) {
 803                         error = EACCES;
 804                         goto done;
 805                 }
 806         }
 807 
 808         (void) VOP_RWLOCK(vp,
 809             rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
 810         if (rw == UIO_WRITE) {
 811                 uio.uio_fmode = FWRITE;
 812                 uio.uio_extflg = UIO_COPY_DEFAULT;
 813                 error = VOP_WRITE(vp, &uio, ioflag, cr, NULL);
 814         } else {
 815                 uio.uio_fmode = FREAD;
 816                 uio.uio_extflg = UIO_COPY_CACHED;
 817                 error = VOP_READ(vp, &uio, ioflag, cr, NULL);
 818         }
 819         VOP_RWUNLOCK(vp,
 820             rw == UIO_WRITE ? V_WRITELOCK_TRUE : V_WRITELOCK_FALSE, NULL);
 821         if (residp)
 822                 *residp = uio.uio_resid;
 823         else if (uio.uio_resid)
 824                 error = EIO;
 825 
 826 done:
 827         if (in_crit)
 828                 nbl_end_crit(vp);
 829         return (error);
 830 }
 831 
 832 /*
 833  * Release a vnode.  Call VOP_INACTIVE on last reference or
 834  * decrement reference count.
 835  *
 836  * To avoid race conditions, the v_count is left at 1 for
 837  * the call to VOP_INACTIVE. This prevents another thread
 838  * from reclaiming and releasing the vnode *before* the
 839  * VOP_INACTIVE routine has a chance to destroy the vnode.
 840  * We can't have more than 1 thread calling VOP_INACTIVE
 841  * on a vnode.
 842  */
 843 void
 844 vn_rele(vnode_t *vp)
 845 {
 846         mutex_enter(&vp->v_lock);
 847         if (vp->v_count == 1) {
 848                 mutex_exit(&vp->v_lock);
 849                 VOP_INACTIVE(vp, CRED(), NULL);
 850                 return;
 851         }
 852         else{
 853                 VERIFY(vp->v_count > 0);
 854         }
 855         VN_RELE_LOCKED(vp);
 856         mutex_exit(&vp->v_lock);
 857 }
 858 
 859 void
 860 vn_phantom_rele(vnode_t *vp)
 861 {
 862         mutex_enter(&vp->v_lock);
 863 
 864         vp->v_phantom_count--;
 865         DTRACE_PROBE1(vn__phantom_rele, vnode_t *, vp);
 866         if (vp->v_count == 1) {
 867                 ASSERT0(vp->v_phantom_count);
 868                 mutex_exit(&vp->v_lock);
 869                 VOP_INACTIVE(vp, CRED(), NULL);
 870                 return;
 871         }else{
 872                 VERIFY(vp->v_count > 0);
 873                 VERIFY3U(vp->v_count, >=, vp->v_phantom_count);
 874         }
 875         VN_RELE_LOCKED(vp);
 876         mutex_exit(&vp->v_lock);
 877 }
 878 
 879 /*
 880  * Return the number of non-phantom holds. Things such as portfs will use
 881  * phantom holds to prevent it from blocking filesystems from mounting over
 882  * watched directories.
 883  */
 884 uint_t
 885 vn_count(vnode_t *vp)
 886 {
 887         ASSERT(MUTEX_HELD(&vp->v_lock));
 888         return (vp->v_count - vp->v_phantom_count);
 889 }
 890 
 891 /*
 892  * Release a vnode referenced by the DNLC. Multiple DNLC references are treated
 893  * as a single reference, so v_count is not decremented until the last DNLC hold
 894  * is released. This makes it possible to distinguish vnodes that are referenced
 895  * only by the DNLC.
 896  */
 897 void
 898 vn_rele_dnlc(vnode_t *vp)
 899 {
 900         mutex_enter(&vp->v_lock);
 901 
 902         if (--vp->v_count_dnlc == 0) {
 903                 if (vp->v_count == 1) {
 904                         mutex_exit(&vp->v_lock);
 905                         VOP_INACTIVE(vp, CRED(), NULL);
 906                         return;
 907                 }
 908                 VN_RELE_LOCKED(vp);
 909         }else{
 910                 VERIFY((vp->v_count > 0) && (vp->v_count_dnlc > 0));
 911         }
 912         mutex_exit(&vp->v_lock);
 913 }
 914 
 915 /*
 916  * Like vn_rele() except that it clears v_stream under v_lock.
 917  * This is used by sockfs when it dismantles the association between
 918  * the sockfs node and the vnode in the underlying file system.
 919  * v_lock has to be held to prevent a thread coming through the lookupname
 920  * path from accessing a stream head that is going away.
 921  */
 922 void
 923 vn_rele_stream(vnode_t *vp)
 924 {
 925         mutex_enter(&vp->v_lock);
 926 
 927         vp->v_stream = NULL;
 928         if (vp->v_count == 1) {
 929                 mutex_exit(&vp->v_lock);
 930                 VOP_INACTIVE(vp, CRED(), NULL);
 931                 return;
 932         }
 933         else{
 934                 VERIFY(vp->v_count > 0);
 935         }
 936         VN_RELE_LOCKED(vp);
 937         mutex_exit(&vp->v_lock);
 938 }
 939 
 940 static void
 941 vn_rele_inactive(vnode_t *vp)
 942 {
 943         VOP_INACTIVE(vp, CRED(), NULL);
 944 }
 945 
 946 /*
 947  * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it
 948  * asynchronously using a taskq. This can avoid deadlocks caused by re-entering
 949  * the file system as a result of releasing the vnode. Note, file systems
 950  * already have to handle the race where the vnode is incremented before the
 951  * inactive routine is called and does its locking.
 952  *
 953  * Warning: Excessive use of this routine can lead to performance problems.
 954  * This is because taskqs throttle back allocation if too many are created.
 955  */
 956 void
 957 vn_rele_async(vnode_t *vp, taskq_t *taskq)
 958 {
 959         mutex_enter(&vp->v_lock);
 960         if (vp->v_count == 1) {
 961                 mutex_exit(&vp->v_lock);
 962                 VERIFY(taskq_dispatch(taskq, (task_func_t *)vn_rele_inactive,
 963                     vp, TQ_SLEEP) != TASKQID_INVALID);
 964                 return;
 965         }
 966         else{
 967                 VERIFY(vp->v_count > 0);
 968         }
 969         VN_RELE_LOCKED(vp);
 970         mutex_exit(&vp->v_lock);
 971 }
 972 
 973 int
 974 vn_open(
 975         char *pnamep,
 976         enum uio_seg seg,
 977         int filemode,
 978         int createmode,
 979         struct vnode **vpp,
 980         enum create crwhy,
 981         mode_t umask)
 982 {
 983         return (vn_openat(pnamep, seg, filemode, createmode, vpp, crwhy,
 984             umask, NULL, -1));
 985 }
 986 
 987 
 988 /*
 989  * Open/create a vnode.
 990  * This may be callable by the kernel, the only known use
 991  * of user context being that the current user credentials
 992  * are used for permissions.  crwhy is defined iff filemode & FCREAT.
 993  */
 994 int
 995 vn_openat(
 996         char *pnamep,
 997         enum uio_seg seg,
 998         int filemode,
 999         int createmode,
1000         struct vnode **vpp,
1001         enum create crwhy,
1002         mode_t umask,
1003         struct vnode *startvp,
1004         int fd)
1005 {
1006         struct vnode *vp;
1007         int mode;
1008         int accessflags;
1009         int error;
1010         int in_crit = 0;
1011         int open_done = 0;
1012         int shrlock_done = 0;
1013         struct vattr vattr;
1014         enum symfollow follow;
1015         int estale_retry = 0;
1016         struct shrlock shr;
1017         struct shr_locowner shr_own;
1018         boolean_t create;
1019 
1020         mode = 0;
1021         accessflags = 0;
1022         if (filemode & FREAD)
1023                 mode |= VREAD;
1024         if (filemode & (FWRITE|FTRUNC))
1025                 mode |= VWRITE;
1026         if (filemode & (FSEARCH|FEXEC|FXATTRDIROPEN))
1027                 mode |= VEXEC;
1028 
1029         /* symlink interpretation */
1030         if (filemode & FNOFOLLOW)
1031                 follow = NO_FOLLOW;
1032         else
1033                 follow = FOLLOW;
1034 
1035         if (filemode & FAPPEND)
1036                 accessflags |= V_APPEND;
1037 
1038         /*
1039          * We need to handle the case of FCREAT | FDIRECTORY and the case of
1040          * FEXCL. If all three are specified, then we always fail because we
1041          * cannot create a directory through this interface and FEXCL says we
1042          * need to fail the request if we can't create it. If, however, only
1043          * FCREAT | FDIRECTORY are specified, then we can treat this as the case
1044          * of opening a file that already exists. If it exists, we can do
1045          * something and if not, we fail. Effectively FCREAT | FDIRECTORY is
1046          * treated as FDIRECTORY.
1047          */
1048         if ((filemode & (FCREAT | FDIRECTORY | FEXCL)) ==
1049             (FCREAT | FDIRECTORY | FEXCL)) {
1050                 return (EINVAL);
1051         }
1052 
1053         if ((filemode & (FCREAT | FDIRECTORY)) == (FCREAT | FDIRECTORY)) {
1054                 create = B_FALSE;
1055         } else if ((filemode & FCREAT) != 0) {
1056                 create = B_TRUE;
1057         } else {
1058                 create = B_FALSE;
1059         }
1060 
1061 top:
1062         if (create) {
1063                 enum vcexcl excl;
1064 
1065                 /*
1066                  * Wish to create a file.
1067                  */
1068                 vattr.va_type = VREG;
1069                 vattr.va_mode = createmode;
1070                 vattr.va_mask = AT_TYPE|AT_MODE;
1071                 if (filemode & FTRUNC) {
1072                         vattr.va_size = 0;
1073                         vattr.va_mask |= AT_SIZE;
1074                 }
1075                 if (filemode & FEXCL)
1076                         excl = EXCL;
1077                 else
1078                         excl = NONEXCL;
1079 
1080                 if (error =
1081                     vn_createat(pnamep, seg, &vattr, excl, mode, &vp, crwhy,
1082                     (filemode & ~(FTRUNC|FEXCL)), umask, startvp))
1083                         return (error);
1084         } else {
1085                 /*
1086                  * Wish to open a file.  Just look it up.
1087                  */
1088                 if (error = lookupnameat(pnamep, seg, follow,
1089                     NULLVPP, &vp, startvp)) {
1090                         if ((error == ESTALE) &&
1091                             fs_need_estale_retry(estale_retry++))
1092                                 goto top;
1093                         return (error);
1094                 }
1095 
1096                 /*
1097                  * Get the attributes to check whether file is large.
1098                  * We do this only if the FOFFMAX flag is not set and
1099                  * only for regular files.
1100                  */
1101 
1102                 if (!(filemode & FOFFMAX) && (vp->v_type == VREG)) {
1103                         vattr.va_mask = AT_SIZE;
1104                         if ((error = VOP_GETATTR(vp, &vattr, 0,
1105                             CRED(), NULL))) {
1106                                 goto out;
1107                         }
1108                         if (vattr.va_size > (u_offset_t)MAXOFF32_T) {
1109                                 /*
1110                                  * Large File API - regular open fails
1111                                  * if FOFFMAX flag is set in file mode
1112                                  */
1113                                 error = EOVERFLOW;
1114                                 goto out;
1115                         }
1116                 }
1117                 /*
1118                  * Can't write directories, active texts, or
1119                  * read-only filesystems.  Can't truncate files
1120                  * on which mandatory locking is in effect.
1121                  */
1122                 if (filemode & (FWRITE|FTRUNC)) {
1123                         /*
1124                          * Allow writable directory if VDIROPEN flag is set.
1125                          */
1126                         if (vp->v_type == VDIR && !(vp->v_flag & VDIROPEN)) {
1127                                 error = EISDIR;
1128                                 goto out;
1129                         }
1130                         if (ISROFILE(vp)) {
1131                                 error = EROFS;
1132                                 goto out;
1133                         }
1134                         /*
1135                          * Can't truncate files on which
1136                          * sysv mandatory locking is in effect.
1137                          */
1138                         if (filemode & FTRUNC) {
1139                                 vnode_t *rvp;
1140 
1141                                 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1142                                         rvp = vp;
1143                                 if (rvp->v_filocks != NULL) {
1144                                         vattr.va_mask = AT_MODE;
1145                                         if ((error = VOP_GETATTR(vp,
1146                                             &vattr, 0, CRED(), NULL)) == 0 &&
1147                                             MANDLOCK(vp, vattr.va_mode))
1148                                                 error = EAGAIN;
1149                                 }
1150                         }
1151                         if (error)
1152                                 goto out;
1153                 }
1154                 /*
1155                  * Check permissions.
1156                  */
1157                 if (error = VOP_ACCESS(vp, mode, accessflags, CRED(), NULL))
1158                         goto out;
1159 
1160                 /*
1161                  * Require FSEARCH and FDIRECTORY to return a directory. Require
1162                  * FEXEC to return a regular file.
1163                  */
1164                 if ((filemode & (FSEARCH|FDIRECTORY)) != 0 &&
1165                     vp->v_type != VDIR) {
1166                         error = ENOTDIR;
1167                         goto out;
1168                 }
1169                 if ((filemode & FEXEC) && vp->v_type != VREG) {
1170                         error = ENOEXEC;        /* XXX: error code? */
1171                         goto out;
1172                 }
1173         }
1174 
1175         /*
1176          * Do remaining checks for FNOFOLLOW and FNOLINKS.
1177          */
1178         if ((filemode & FNOFOLLOW) && vp->v_type == VLNK) {
1179                 /*
1180                  * The __FLXPATH flag is a private interface for use by the lx
1181                  * brand in order to emulate open(O_NOFOLLOW|O_PATH) which,
1182                  * when a symbolic link is encountered, returns a file
1183                  * descriptor which references it.
1184                  * See uts/common/brand/lx/syscall/lx_open.c
1185                  *
1186                  * When this flag is set, VOP_OPEN() is not called (for a
1187                  * symlink, most filesystems will return ENOSYS anyway)
1188                  * and the link's vnode is returned to be linked to the
1189                  * file descriptor.
1190                  */
1191                 if ((filemode & __FLXPATH) == 0)
1192                         error = ELOOP;
1193                 goto out;
1194         }
1195         if (filemode & FNOLINKS) {
1196                 vattr.va_mask = AT_NLINK;
1197                 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))) {
1198                         goto out;
1199                 }
1200                 if (vattr.va_nlink != 1) {
1201                         error = EMLINK;
1202                         goto out;
1203                 }
1204         }
1205 
1206         /*
1207          * Opening a socket corresponding to the AF_UNIX pathname
1208          * in the filesystem name space is not supported.
1209          * However, VSOCK nodes in namefs are supported in order
1210          * to make fattach work for sockets.
1211          *
1212          * XXX This uses VOP_REALVP to distinguish between
1213          * an unopened namefs node (where VOP_REALVP returns a
1214          * different VSOCK vnode) and a VSOCK created by vn_create
1215          * in some file system (where VOP_REALVP would never return
1216          * a different vnode).
1217          */
1218         if (vp->v_type == VSOCK) {
1219                 struct vnode *nvp;
1220 
1221                 error = VOP_REALVP(vp, &nvp, NULL);
1222                 if (error != 0 || nvp == NULL || nvp == vp ||
1223                     nvp->v_type != VSOCK) {
1224                         error = EOPNOTSUPP;
1225                         goto out;
1226                 }
1227         }
1228 
1229         if ((vp->v_type == VREG) && nbl_need_check(vp)) {
1230                 /* get share reservation */
1231                 shr.s_access = 0;
1232                 if (filemode & FWRITE)
1233                         shr.s_access |= F_WRACC;
1234                 if (filemode & FREAD)
1235                         shr.s_access |= F_RDACC;
1236                 shr.s_deny = 0;
1237                 shr.s_sysid = 0;
1238                 shr.s_pid = ttoproc(curthread)->p_pid;
1239                 shr_own.sl_pid = shr.s_pid;
1240                 shr_own.sl_id = fd;
1241                 shr.s_own_len = sizeof (shr_own);
1242                 shr.s_owner = (caddr_t)&shr_own;
1243                 error = VOP_SHRLOCK(vp, F_SHARE_NBMAND, &shr, filemode, CRED(),
1244                     NULL);
1245                 if (error)
1246                         goto out;
1247                 shrlock_done = 1;
1248 
1249                 /* nbmand conflict check if truncating file */
1250                 if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1251                         nbl_start_crit(vp, RW_READER);
1252                         in_crit = 1;
1253 
1254                         vattr.va_mask = AT_SIZE;
1255                         if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
1256                                 goto out;
1257                         if (nbl_conflict(vp, NBL_WRITE, 0, vattr.va_size, 0,
1258                             NULL)) {
1259                                 error = EACCES;
1260                                 goto out;
1261                         }
1262                 }
1263         }
1264 
1265         /*
1266          * Do opening protocol.
1267          */
1268         error = VOP_OPEN(&vp, filemode, CRED(), NULL);
1269         if (error)
1270                 goto out;
1271         open_done = 1;
1272 
1273         /*
1274          * Truncate if required.
1275          */
1276         if ((filemode & FTRUNC) && !(filemode & FCREAT)) {
1277                 vattr.va_size = 0;
1278                 vattr.va_mask = AT_SIZE;
1279                 if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
1280                         goto out;
1281         }
1282 
1283         /*
1284          * Turn on directio, if requested.
1285          */
1286         if (filemode & FDIRECT) {
1287                 if ((error = VOP_IOCTL(vp, _FIODIRECTIO, DIRECTIO_ON, 0,
1288                     CRED(), NULL, NULL)) != 0) {
1289                         /*
1290                          * On Linux, O_DIRECT returns EINVAL when the file
1291                          * system does not support directio, so we'll do the
1292                          * same.
1293                          */
1294                         error = EINVAL;
1295                         goto out;
1296                 }
1297         }
1298 out:
1299         ASSERT(vp->v_count > 0);
1300 
1301         if (in_crit) {
1302                 nbl_end_crit(vp);
1303                 in_crit = 0;
1304         }
1305         if (error) {
1306                 if (open_done) {
1307                         (void) VOP_CLOSE(vp, filemode, 1, (offset_t)0, CRED(),
1308                             NULL);
1309                         open_done = 0;
1310                         shrlock_done = 0;
1311                 }
1312                 if (shrlock_done) {
1313                         (void) VOP_SHRLOCK(vp, F_UNSHARE, &shr, 0, CRED(),
1314                             NULL);
1315                         shrlock_done = 0;
1316                 }
1317 
1318                 /*
1319                  * The following clause was added to handle a problem
1320                  * with NFS consistency.  It is possible that a lookup
1321                  * of the file to be opened succeeded, but the file
1322                  * itself doesn't actually exist on the server.  This
1323                  * is chiefly due to the DNLC containing an entry for
1324                  * the file which has been removed on the server.  In
1325                  * this case, we just start over.  If there was some
1326                  * other cause for the ESTALE error, then the lookup
1327                  * of the file will fail and the error will be returned
1328                  * above instead of looping around from here.
1329                  */
1330                 VN_RELE(vp);
1331                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1332                         goto top;
1333         } else
1334                 *vpp = vp;
1335         return (error);
1336 }
1337 
1338 /*
1339  * The following two accessor functions are for the NFSv4 server.  Since there
1340  * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the
1341  * vnode open counts correct when a client "upgrades" an open or does an
1342  * open_downgrade.  In NFS, an upgrade or downgrade can not only change the
1343  * open mode (add or subtract read or write), but also change the share/deny
1344  * modes.  However, share reservations are not integrated with OPEN, yet, so
1345  * we need to handle each separately.  These functions are cleaner than having
1346  * the NFS server manipulate the counts directly, however, nobody else should
1347  * use these functions.
1348  */
1349 void
1350 vn_open_upgrade(
1351         vnode_t *vp,
1352         int filemode)
1353 {
1354         ASSERT(vp->v_type == VREG);
1355 
1356         if (filemode & FREAD)
1357                 atomic_inc_32(&vp->v_rdcnt);
1358         if (filemode & FWRITE)
1359                 atomic_inc_32(&vp->v_wrcnt);
1360 
1361 }
1362 
1363 void
1364 vn_open_downgrade(
1365         vnode_t *vp,
1366         int filemode)
1367 {
1368         ASSERT(vp->v_type == VREG);
1369 
1370         if (filemode & FREAD) {
1371                 ASSERT(vp->v_rdcnt > 0);
1372                 atomic_dec_32(&vp->v_rdcnt);
1373         }
1374         if (filemode & FWRITE) {
1375                 ASSERT(vp->v_wrcnt > 0);
1376                 atomic_dec_32(&vp->v_wrcnt);
1377         }
1378 
1379 }
1380 
1381 int
1382 vn_create(
1383         char *pnamep,
1384         enum uio_seg seg,
1385         struct vattr *vap,
1386         enum vcexcl excl,
1387         int mode,
1388         struct vnode **vpp,
1389         enum create why,
1390         int flag,
1391         mode_t umask)
1392 {
1393         return (vn_createat(pnamep, seg, vap, excl, mode, vpp, why, flag,
1394             umask, NULL));
1395 }
1396 
1397 /*
1398  * Create a vnode (makenode).
1399  */
1400 int
1401 vn_createat(
1402         char *pnamep,
1403         enum uio_seg seg,
1404         struct vattr *vap,
1405         enum vcexcl excl,
1406         int mode,
1407         struct vnode **vpp,
1408         enum create why,
1409         int flag,
1410         mode_t umask,
1411         struct vnode *startvp)
1412 {
1413         struct vnode *dvp;      /* ptr to parent dir vnode */
1414         struct vnode *vp = NULL;
1415         struct pathname pn;
1416         int error;
1417         int in_crit = 0;
1418         struct vattr vattr;
1419         enum symfollow follow;
1420         int estale_retry = 0;
1421         uint32_t auditing = AU_AUDITING();
1422 
1423         ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
1424 
1425         /* symlink interpretation */
1426         if ((flag & FNOFOLLOW) || excl == EXCL)
1427                 follow = NO_FOLLOW;
1428         else
1429                 follow = FOLLOW;
1430         flag &= ~(FNOFOLLOW|FNOLINKS);
1431 
1432 top:
1433         /*
1434          * Lookup directory.
1435          * If new object is a file, call lower level to create it.
1436          * Note that it is up to the lower level to enforce exclusive
1437          * creation, if the file is already there.
1438          * This allows the lower level to do whatever
1439          * locking or protocol that is needed to prevent races.
1440          * If the new object is directory call lower level to make
1441          * the new directory, with "." and "..".
1442          */
1443         if (error = pn_get(pnamep, seg, &pn))
1444                 return (error);
1445         if (auditing)
1446                 audit_vncreate_start();
1447         dvp = NULL;
1448         *vpp = NULL;
1449         /*
1450          * lookup will find the parent directory for the vnode.
1451          * When it is done the pn holds the name of the entry
1452          * in the directory.
1453          * If this is a non-exclusive create we also find the node itself.
1454          */
1455         error = lookuppnat(&pn, NULL, follow, &dvp,
1456             (excl == EXCL) ? NULLVPP : vpp, startvp);
1457         if (error) {
1458                 pn_free(&pn);
1459                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1460                         goto top;
1461                 if (why == CRMKDIR && error == EINVAL)
1462                         error = EEXIST;         /* SVID */
1463                 return (error);
1464         }
1465 
1466         if (why != CRMKNOD)
1467                 vap->va_mode &= ~VSVTX;
1468 
1469         /*
1470          * If default ACLs are defined for the directory don't apply the
1471          * umask if umask is passed.
1472          */
1473 
1474         if (umask) {
1475 
1476                 vsecattr_t vsec;
1477 
1478                 vsec.vsa_aclcnt = 0;
1479                 vsec.vsa_aclentp = NULL;
1480                 vsec.vsa_dfaclcnt = 0;
1481                 vsec.vsa_dfaclentp = NULL;
1482                 vsec.vsa_mask = VSA_DFACLCNT;
1483                 error = VOP_GETSECATTR(dvp, &vsec, 0, CRED(), NULL);
1484                 /*
1485                  * If error is ENOSYS then treat it as no error
1486                  * Don't want to force all file systems to support
1487                  * aclent_t style of ACL's.
1488                  */
1489                 if (error == ENOSYS)
1490                         error = 0;
1491                 if (error) {
1492                         if (*vpp != NULL)
1493                                 VN_RELE(*vpp);
1494                         goto out;
1495                 } else {
1496                         /*
1497                          * Apply the umask if no default ACLs.
1498                          */
1499                         if (vsec.vsa_dfaclcnt == 0)
1500                                 vap->va_mode &= ~umask;
1501 
1502                         /*
1503                          * VOP_GETSECATTR() may have allocated memory for
1504                          * ACLs we didn't request, so double-check and
1505                          * free it if necessary.
1506                          */
1507                         if (vsec.vsa_aclcnt && vsec.vsa_aclentp != NULL)
1508                                 kmem_free((caddr_t)vsec.vsa_aclentp,
1509                                     vsec.vsa_aclcnt * sizeof (aclent_t));
1510                         if (vsec.vsa_dfaclcnt && vsec.vsa_dfaclentp != NULL)
1511                                 kmem_free((caddr_t)vsec.vsa_dfaclentp,
1512                                     vsec.vsa_dfaclcnt * sizeof (aclent_t));
1513                 }
1514         }
1515 
1516         /*
1517          * In general we want to generate EROFS if the file system is
1518          * readonly.  However, POSIX (IEEE Std. 1003.1) section 5.3.1
1519          * documents the open system call, and it says that O_CREAT has no
1520          * effect if the file already exists.  Bug 1119649 states
1521          * that open(path, O_CREAT, ...) fails when attempting to open an
1522          * existing file on a read only file system.  Thus, the first part
1523          * of the following if statement has 3 checks:
1524          *      if the file exists &&
1525          *              it is being open with write access &&
1526          *              the file system is read only
1527          *      then generate EROFS
1528          */
1529         if ((*vpp != NULL && (mode & VWRITE) && ISROFILE(*vpp)) ||
1530             (*vpp == NULL && dvp->v_vfsp->vfs_flag & VFS_RDONLY)) {
1531                 if (*vpp)
1532                         VN_RELE(*vpp);
1533                 error = EROFS;
1534         } else if (excl == NONEXCL && *vpp != NULL) {
1535                 vnode_t *rvp;
1536 
1537                 /*
1538                  * File already exists.  If a mandatory lock has been
1539                  * applied, return error.
1540                  */
1541                 vp = *vpp;
1542                 if (VOP_REALVP(vp, &rvp, NULL) != 0)
1543                         rvp = vp;
1544                 if ((vap->va_mask & AT_SIZE) && nbl_need_check(vp)) {
1545                         nbl_start_crit(vp, RW_READER);
1546                         in_crit = 1;
1547                 }
1548                 if (rvp->v_filocks != NULL || rvp->v_shrlocks != NULL) {
1549                         vattr.va_mask = AT_MODE|AT_SIZE;
1550                         if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) {
1551                                 goto out;
1552                         }
1553                         if (MANDLOCK(vp, vattr.va_mode)) {
1554                                 error = EAGAIN;
1555                                 goto out;
1556                         }
1557                         /*
1558                          * File cannot be truncated if non-blocking mandatory
1559                          * locks are currently on the file.
1560                          */
1561                         if ((vap->va_mask & AT_SIZE) && in_crit) {
1562                                 u_offset_t offset;
1563                                 ssize_t length;
1564 
1565                                 offset = vap->va_size > vattr.va_size ?
1566                                     vattr.va_size : vap->va_size;
1567                                 length = vap->va_size > vattr.va_size ?
1568                                     vap->va_size - vattr.va_size :
1569                                     vattr.va_size - vap->va_size;
1570                                 if (nbl_conflict(vp, NBL_WRITE, offset,
1571                                     length, 0, NULL)) {
1572                                         error = EACCES;
1573                                         goto out;
1574                                 }
1575                         }
1576                 }
1577 
1578                 /*
1579                  * If the file is the root of a VFS, we've crossed a
1580                  * mount point and the "containing" directory that we
1581                  * acquired above (dvp) is irrelevant because it's in
1582                  * a different file system.  We apply VOP_CREATE to the
1583                  * target itself instead of to the containing directory
1584                  * and supply a null path name to indicate (conventionally)
1585                  * the node itself as the "component" of interest.
1586                  *
1587                  * The call to VOP_CREATE() is necessary to ensure
1588                  * that the appropriate permission checks are made,
1589                  * i.e. EISDIR, EACCES, etc.  We already know that vpp
1590                  * exists since we are in the else condition where this
1591                  * was checked.
1592                  */
1593                 if (vp->v_flag & VROOT) {
1594                         ASSERT(why != CRMKDIR);
1595                         error = VOP_CREATE(vp, "", vap, excl, mode, vpp,
1596                             CRED(), flag, NULL, NULL);
1597                         /*
1598                          * If the create succeeded, it will have created a
1599                          * new reference on a new vnode (*vpp) in the child
1600                          * file system, so we want to drop our reference on
1601                          * the old (vp) upon exit.
1602                          */
1603                         goto out;
1604                 }
1605 
1606                 /*
1607                  * Large File API - non-large open (FOFFMAX flag not set)
1608                  * of regular file fails if the file size exceeds MAXOFF32_T.
1609                  */
1610                 if (why != CRMKDIR &&
1611                     !(flag & FOFFMAX) &&
1612                     (vp->v_type == VREG)) {
1613                         vattr.va_mask = AT_SIZE;
1614                         if ((error = VOP_GETATTR(vp, &vattr, 0,
1615                             CRED(), NULL))) {
1616                                 goto out;
1617                         }
1618                         if ((vattr.va_size > (u_offset_t)MAXOFF32_T)) {
1619                                 error = EOVERFLOW;
1620                                 goto out;
1621                         }
1622                 }
1623         }
1624 
1625         if (error == 0) {
1626                 /*
1627                  * Call mkdir() if specified, otherwise create().
1628                  */
1629                 int must_be_dir = pn_fixslash(&pn); /* trailing '/'? */
1630 
1631                 if (why == CRMKDIR)
1632                         /*
1633                          * N.B., if vn_createat() ever requests
1634                          * case-insensitive behavior then it will need
1635                          * to be passed to VOP_MKDIR().  VOP_CREATE()
1636                          * will already get it via "flag"
1637                          */
1638                         error = VOP_MKDIR(dvp, pn.pn_path, vap, vpp, CRED(),
1639                             NULL, 0, NULL);
1640                 else if (!must_be_dir)
1641                         error = VOP_CREATE(dvp, pn.pn_path, vap,
1642                             excl, mode, vpp, CRED(), flag, NULL, NULL);
1643                 else
1644                         error = ENOTDIR;
1645         }
1646 
1647 out:
1648 
1649         if (auditing)
1650                 audit_vncreate_finish(*vpp, error);
1651         if (in_crit) {
1652                 nbl_end_crit(vp);
1653                 in_crit = 0;
1654         }
1655         if (vp != NULL) {
1656                 VN_RELE(vp);
1657                 vp = NULL;
1658         }
1659         pn_free(&pn);
1660         VN_RELE(dvp);
1661         /*
1662          * The following clause was added to handle a problem
1663          * with NFS consistency.  It is possible that a lookup
1664          * of the file to be created succeeded, but the file
1665          * itself doesn't actually exist on the server.  This
1666          * is chiefly due to the DNLC containing an entry for
1667          * the file which has been removed on the server.  In
1668          * this case, we just start over.  If there was some
1669          * other cause for the ESTALE error, then the lookup
1670          * of the file will fail and the error will be returned
1671          * above instead of looping around from here.
1672          */
1673         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1674                 goto top;
1675         return (error);
1676 }
1677 
1678 int
1679 vn_link(char *from, char *to, enum uio_seg seg)
1680 {
1681         return (vn_linkat(NULL, from, NO_FOLLOW, NULL, to, seg));
1682 }
1683 
1684 int
1685 vn_linkat(vnode_t *fstartvp, char *from, enum symfollow follow,
1686     vnode_t *tstartvp, char *to, enum uio_seg seg)
1687 {
1688         struct vnode *fvp;              /* from vnode ptr */
1689         struct vnode *tdvp;             /* to directory vnode ptr */
1690         struct pathname pn;
1691         int error;
1692         struct vattr vattr;
1693         dev_t fsid;
1694         int estale_retry = 0;
1695         uint32_t auditing = AU_AUDITING();
1696 
1697 top:
1698         fvp = tdvp = NULL;
1699         if (error = pn_get(to, seg, &pn))
1700                 return (error);
1701         if (auditing && fstartvp != NULL)
1702                 audit_setfsat_path(1);
1703         if (error = lookupnameat(from, seg, follow, NULLVPP, &fvp, fstartvp))
1704                 goto out;
1705         if (auditing && tstartvp != NULL)
1706                 audit_setfsat_path(3);
1707         if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &tdvp, NULLVPP, tstartvp))
1708                 goto out;
1709         /*
1710          * Make sure both source vnode and target directory vnode are
1711          * in the same vfs and that it is writeable.
1712          */
1713         vattr.va_mask = AT_FSID;
1714         if (error = VOP_GETATTR(fvp, &vattr, 0, CRED(), NULL))
1715                 goto out;
1716         fsid = vattr.va_fsid;
1717         vattr.va_mask = AT_FSID;
1718         if (error = VOP_GETATTR(tdvp, &vattr, 0, CRED(), NULL))
1719                 goto out;
1720         if (fsid != vattr.va_fsid) {
1721                 error = EXDEV;
1722                 goto out;
1723         }
1724         if (tdvp->v_vfsp->vfs_flag & VFS_RDONLY) {
1725                 error = EROFS;
1726                 goto out;
1727         }
1728         /*
1729          * Do the link.
1730          */
1731         (void) pn_fixslash(&pn);
1732         error = VOP_LINK(tdvp, fvp, pn.pn_path, CRED(), NULL, 0);
1733 out:
1734         pn_free(&pn);
1735         if (fvp)
1736                 VN_RELE(fvp);
1737         if (tdvp)
1738                 VN_RELE(tdvp);
1739         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1740                 goto top;
1741         return (error);
1742 }
1743 
1744 int
1745 vn_rename(char *from, char *to, enum uio_seg seg)
1746 {
1747         return (vn_renameat(NULL, from, NULL, to, seg));
1748 }
1749 
1750 int
1751 vn_renameat(vnode_t *fdvp, char *fname, vnode_t *tdvp,
1752     char *tname, enum uio_seg seg)
1753 {
1754         int error;
1755         struct vattr vattr;
1756         struct pathname fpn;            /* from pathname */
1757         struct pathname tpn;            /* to pathname */
1758         dev_t fsid;
1759         int in_crit_src, in_crit_targ;
1760         vnode_t *fromvp, *fvp;
1761         vnode_t *tovp, *targvp;
1762         int estale_retry = 0;
1763         uint32_t auditing = AU_AUDITING();
1764 
1765 top:
1766         fvp = fromvp = tovp = targvp = NULL;
1767         in_crit_src = in_crit_targ = 0;
1768         /*
1769          * Get to and from pathnames.
1770          */
1771         if (error = pn_get(fname, seg, &fpn))
1772                 return (error);
1773         if (error = pn_get(tname, seg, &tpn)) {
1774                 pn_free(&fpn);
1775                 return (error);
1776         }
1777 
1778         /*
1779          * First we need to resolve the correct directories
1780          * The passed in directories may only be a starting point,
1781          * but we need the real directories the file(s) live in.
1782          * For example the fname may be something like usr/lib/sparc
1783          * and we were passed in the / directory, but we need to
1784          * use the lib directory for the rename.
1785          */
1786 
1787         if (auditing && fdvp != NULL)
1788                 audit_setfsat_path(1);
1789         /*
1790          * Lookup to and from directories.
1791          */
1792         if (error = lookuppnat(&fpn, NULL, NO_FOLLOW, &fromvp, &fvp, fdvp)) {
1793                 goto out;
1794         }
1795 
1796         /*
1797          * Make sure there is an entry.
1798          */
1799         if (fvp == NULL) {
1800                 error = ENOENT;
1801                 goto out;
1802         }
1803 
1804         if (auditing && tdvp != NULL)
1805                 audit_setfsat_path(3);
1806         if (error = lookuppnat(&tpn, NULL, NO_FOLLOW, &tovp, &targvp, tdvp)) {
1807                 goto out;
1808         }
1809 
1810         /*
1811          * Make sure both the from vnode directory and the to directory
1812          * are in the same vfs and the to directory is writable.
1813          * We check fsid's, not vfs pointers, so loopback fs works.
1814          */
1815         if (fromvp != tovp) {
1816                 vattr.va_mask = AT_FSID;
1817                 if (error = VOP_GETATTR(fromvp, &vattr, 0, CRED(), NULL))
1818                         goto out;
1819                 fsid = vattr.va_fsid;
1820                 vattr.va_mask = AT_FSID;
1821                 if (error = VOP_GETATTR(tovp, &vattr, 0, CRED(), NULL))
1822                         goto out;
1823                 if (fsid != vattr.va_fsid) {
1824                         error = EXDEV;
1825                         goto out;
1826                 }
1827         }
1828 
1829         if (tovp->v_vfsp->vfs_flag & VFS_RDONLY) {
1830                 error = EROFS;
1831                 goto out;
1832         }
1833 
1834         /*
1835          * Make sure "from" vp is not a mount point.
1836          * Note, lookup did traverse() already, so
1837          * we'll be looking at the mounted FS root.
1838          * (but allow files like mnttab)
1839          */
1840         if ((fvp->v_flag & VROOT) != 0 && fvp->v_type == VDIR) {
1841                 error = EBUSY;
1842                 goto out;
1843         }
1844 
1845         if (targvp && (fvp != targvp)) {
1846                 nbl_start_crit(targvp, RW_READER);
1847                 in_crit_targ = 1;
1848                 if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
1849                         error = EACCES;
1850                         goto out;
1851                 }
1852         }
1853 
1854         if (nbl_need_check(fvp)) {
1855                 nbl_start_crit(fvp, RW_READER);
1856                 in_crit_src = 1;
1857                 if (nbl_conflict(fvp, NBL_RENAME, 0, 0, 0, NULL)) {
1858                         error = EACCES;
1859                         goto out;
1860                 }
1861         }
1862 
1863         /*
1864          * Do the rename.
1865          */
1866         (void) pn_fixslash(&tpn);
1867         error = VOP_RENAME(fromvp, fpn.pn_path, tovp, tpn.pn_path, CRED(),
1868             NULL, 0);
1869 
1870 out:
1871         pn_free(&fpn);
1872         pn_free(&tpn);
1873         if (in_crit_src)
1874                 nbl_end_crit(fvp);
1875         if (in_crit_targ)
1876                 nbl_end_crit(targvp);
1877         if (fromvp)
1878                 VN_RELE(fromvp);
1879         if (tovp)
1880                 VN_RELE(tovp);
1881         if (targvp)
1882                 VN_RELE(targvp);
1883         if (fvp)
1884                 VN_RELE(fvp);
1885         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1886                 goto top;
1887         return (error);
1888 }
1889 
1890 /*
1891  * Remove a file or directory.
1892  */
1893 int
1894 vn_remove(char *fnamep, enum uio_seg seg, enum rm dirflag)
1895 {
1896         return (vn_removeat(NULL, fnamep, seg, dirflag));
1897 }
1898 
1899 int
1900 vn_removeat(vnode_t *startvp, char *fnamep, enum uio_seg seg, enum rm dirflag)
1901 {
1902         struct vnode *vp;               /* entry vnode */
1903         struct vnode *dvp;              /* ptr to parent dir vnode */
1904         struct vnode *coveredvp;
1905         struct pathname pn;             /* name of entry */
1906         enum vtype vtype;
1907         int error;
1908         struct vfs *vfsp;
1909         struct vfs *dvfsp;      /* ptr to parent dir vfs */
1910         int in_crit = 0;
1911         int estale_retry = 0;
1912 
1913 top:
1914         if (error = pn_get(fnamep, seg, &pn))
1915                 return (error);
1916         dvp = vp = NULL;
1917         if (error = lookuppnat(&pn, NULL, NO_FOLLOW, &dvp, &vp, startvp)) {
1918                 pn_free(&pn);
1919                 if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
1920                         goto top;
1921                 return (error);
1922         }
1923 
1924         /*
1925          * Make sure there is an entry.
1926          */
1927         if (vp == NULL) {
1928                 error = ENOENT;
1929                 goto out;
1930         }
1931 
1932         vfsp = vp->v_vfsp;
1933         dvfsp = dvp->v_vfsp;
1934 
1935         /*
1936          * If the named file is the root of a mounted filesystem, fail,
1937          * unless it's marked unlinkable.  In that case, unmount the
1938          * filesystem and proceed to unlink the covered vnode.  (If the
1939          * covered vnode is a directory, use rmdir instead of unlink,
1940          * to avoid file system corruption.)
1941          */
1942         if (vp->v_flag & VROOT) {
1943                 if ((vfsp->vfs_flag & VFS_UNLINKABLE) == 0) {
1944                         error = EBUSY;
1945                         goto out;
1946                 }
1947 
1948                 /*
1949                  * Namefs specific code starts here.
1950                  */
1951 
1952                 if (dirflag == RMDIRECTORY) {
1953                         /*
1954                          * User called rmdir(2) on a file that has
1955                          * been namefs mounted on top of.  Since
1956                          * namefs doesn't allow directories to
1957                          * be mounted on other files we know
1958                          * vp is not of type VDIR so fail to operation.
1959                          */
1960                         error = ENOTDIR;
1961                         goto out;
1962                 }
1963 
1964                 /*
1965                  * If VROOT is still set after grabbing vp->v_lock,
1966                  * noone has finished nm_unmount so far and coveredvp
1967                  * is valid.
1968                  * If we manage to grab vn_vfswlock(coveredvp) before releasing
1969                  * vp->v_lock, any race window is eliminated.
1970                  */
1971 
1972                 mutex_enter(&vp->v_lock);
1973                 if ((vp->v_flag & VROOT) == 0) {
1974                         /* Someone beat us to the unmount */
1975                         mutex_exit(&vp->v_lock);
1976                         error = EBUSY;
1977                         goto out;
1978                 }
1979                 vfsp = vp->v_vfsp;
1980                 coveredvp = vfsp->vfs_vnodecovered;
1981                 ASSERT(coveredvp);
1982                 /*
1983                  * Note: Implementation of vn_vfswlock shows that ordering of
1984                  * v_lock / vn_vfswlock is not an issue here.
1985                  */
1986                 error = vn_vfswlock(coveredvp);
1987                 mutex_exit(&vp->v_lock);
1988 
1989                 if (error)
1990                         goto out;
1991 
1992                 VN_HOLD(coveredvp);
1993                 VN_RELE(vp);
1994                 error = dounmount(vfsp, 0, CRED());
1995 
1996                 /*
1997                  * Unmounted the namefs file system; now get
1998                  * the object it was mounted over.
1999                  */
2000                 vp = coveredvp;
2001                 /*
2002                  * If namefs was mounted over a directory, then
2003                  * we want to use rmdir() instead of unlink().
2004                  */
2005                 if (vp->v_type == VDIR)
2006                         dirflag = RMDIRECTORY;
2007 
2008                 if (error)
2009                         goto out;
2010         }
2011 
2012         /*
2013          * Make sure filesystem is writeable.
2014          * We check the parent directory's vfs in case this is an lofs vnode.
2015          */
2016         if (dvfsp && dvfsp->vfs_flag & VFS_RDONLY) {
2017                 error = EROFS;
2018                 goto out;
2019         }
2020 
2021         vtype = vp->v_type;
2022 
2023         /*
2024          * If there is the possibility of an nbmand share reservation, make
2025          * sure it's okay to remove the file.  Keep a reference to the
2026          * vnode, so that we can exit the nbl critical region after
2027          * calling VOP_REMOVE.
2028          * If there is no possibility of an nbmand share reservation,
2029          * release the vnode reference now.  Filesystems like NFS may
2030          * behave differently if there is an extra reference, so get rid of
2031          * this one.  Fortunately, we can't have nbmand mounts on NFS
2032          * filesystems.
2033          */
2034         if (nbl_need_check(vp)) {
2035                 nbl_start_crit(vp, RW_READER);
2036                 in_crit = 1;
2037                 if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
2038                         error = EACCES;
2039                         goto out;
2040                 }
2041         } else {
2042                 VN_RELE(vp);
2043                 vp = NULL;
2044         }
2045 
2046         if (dirflag == RMDIRECTORY) {
2047                 /*
2048                  * Caller is using rmdir(2), which can only be applied to
2049                  * directories.
2050                  */
2051                 if (vtype != VDIR) {
2052                         error = ENOTDIR;
2053                 } else {
2054                         vnode_t *cwd;
2055                         proc_t *pp = curproc;
2056 
2057                         mutex_enter(&pp->p_lock);
2058                         cwd = PTOU(pp)->u_cdir;
2059                         VN_HOLD(cwd);
2060                         mutex_exit(&pp->p_lock);
2061                         error = VOP_RMDIR(dvp, pn.pn_path, cwd, CRED(),
2062                             NULL, 0);
2063                         VN_RELE(cwd);
2064                 }
2065         } else {
2066                 /*
2067                  * Unlink(2) can be applied to anything.
2068                  */
2069                 error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
2070         }
2071 
2072 out:
2073         pn_free(&pn);
2074         if (in_crit) {
2075                 nbl_end_crit(vp);
2076                 in_crit = 0;
2077         }
2078         if (vp != NULL)
2079                 VN_RELE(vp);
2080         if (dvp != NULL)
2081                 VN_RELE(dvp);
2082         if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
2083                 goto top;
2084         return (error);
2085 }
2086 
2087 /*
2088  * Utility function to compare equality of vnodes.
2089  * Compare the underlying real vnodes, if there are underlying vnodes.
2090  * This is a more thorough comparison than the VN_CMP() macro provides.
2091  */
2092 int
2093 vn_compare(vnode_t *vp1, vnode_t *vp2)
2094 {
2095         vnode_t *realvp;
2096 
2097         if (vp1 != NULL && VOP_REALVP(vp1, &realvp, NULL) == 0)
2098                 vp1 = realvp;
2099         if (vp2 != NULL && VOP_REALVP(vp2, &realvp, NULL) == 0)
2100                 vp2 = realvp;
2101         return (VN_CMP(vp1, vp2));
2102 }
2103 
2104 /*
2105  * The number of locks to hash into.  This value must be a power
2106  * of 2 minus 1 and should probably also be prime.
2107  */
2108 #define NUM_BUCKETS     1023
2109 
2110 struct  vn_vfslocks_bucket {
2111         kmutex_t vb_lock;
2112         vn_vfslocks_entry_t *vb_list;
2113         char pad[64 - sizeof (kmutex_t) - sizeof (void *)];
2114 };
2115 
2116 /*
2117  * Total number of buckets will be NUM_BUCKETS + 1 .
2118  */
2119 
2120 #pragma align   64(vn_vfslocks_buckets)
2121 static  struct vn_vfslocks_bucket       vn_vfslocks_buckets[NUM_BUCKETS + 1];
2122 
2123 #define VN_VFSLOCKS_SHIFT       9
2124 
2125 #define VN_VFSLOCKS_HASH(vfsvpptr)      \
2126         ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS)
2127 
2128 /*
2129  * vn_vfslocks_getlock() uses an HASH scheme to generate
2130  * rwstlock using vfs/vnode pointer passed to it.
2131  *
2132  * vn_vfslocks_rele() releases a reference in the
2133  * HASH table which allows the entry allocated by
2134  * vn_vfslocks_getlock() to be freed at a later
2135  * stage when the refcount drops to zero.
2136  */
2137 
2138 vn_vfslocks_entry_t *
2139 vn_vfslocks_getlock(void *vfsvpptr)
2140 {
2141         struct vn_vfslocks_bucket *bp;
2142         vn_vfslocks_entry_t *vep;
2143         vn_vfslocks_entry_t *tvep;
2144 
2145         ASSERT(vfsvpptr != NULL);
2146         bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vfsvpptr)];
2147 
2148         mutex_enter(&bp->vb_lock);
2149         for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2150                 if (vep->ve_vpvfs == vfsvpptr) {
2151                         vep->ve_refcnt++;
2152                         mutex_exit(&bp->vb_lock);
2153                         return (vep);
2154                 }
2155         }
2156         mutex_exit(&bp->vb_lock);
2157         vep = kmem_alloc(sizeof (*vep), KM_SLEEP);
2158         rwst_init(&vep->ve_lock, NULL, RW_DEFAULT, NULL);
2159         vep->ve_vpvfs = (char *)vfsvpptr;
2160         vep->ve_refcnt = 1;
2161         mutex_enter(&bp->vb_lock);
2162         for (tvep = bp->vb_list; tvep != NULL; tvep = tvep->ve_next) {
2163                 if (tvep->ve_vpvfs == vfsvpptr) {
2164                         tvep->ve_refcnt++;
2165                         mutex_exit(&bp->vb_lock);
2166 
2167                         /*
2168                          * There is already an entry in the hash
2169                          * destroy what we just allocated.
2170                          */
2171                         rwst_destroy(&vep->ve_lock);
2172                         kmem_free(vep, sizeof (*vep));
2173                         return (tvep);
2174                 }
2175         }
2176         vep->ve_next = bp->vb_list;
2177         bp->vb_list = vep;
2178         mutex_exit(&bp->vb_lock);
2179         return (vep);
2180 }
2181 
2182 void
2183 vn_vfslocks_rele(vn_vfslocks_entry_t *vepent)
2184 {
2185         struct vn_vfslocks_bucket *bp;
2186         vn_vfslocks_entry_t *vep;
2187         vn_vfslocks_entry_t *pvep;
2188 
2189         ASSERT(vepent != NULL);
2190         ASSERT(vepent->ve_vpvfs != NULL);
2191 
2192         bp = &vn_vfslocks_buckets[VN_VFSLOCKS_HASH(vepent->ve_vpvfs)];
2193 
2194         mutex_enter(&bp->vb_lock);
2195         vepent->ve_refcnt--;
2196 
2197         if ((int32_t)vepent->ve_refcnt < 0)
2198                 cmn_err(CE_PANIC, "vn_vfslocks_rele: refcount negative");
2199 
2200         pvep = NULL;
2201         if (vepent->ve_refcnt == 0) {
2202                 for (vep = bp->vb_list; vep != NULL; vep = vep->ve_next) {
2203                         if (vep->ve_vpvfs == vepent->ve_vpvfs) {
2204                                 if (pvep == NULL)
2205                                         bp->vb_list = vep->ve_next;
2206                                 else {
2207                                         pvep->ve_next = vep->ve_next;
2208                                 }
2209                                 mutex_exit(&bp->vb_lock);
2210                                 rwst_destroy(&vep->ve_lock);
2211                                 kmem_free(vep, sizeof (*vep));
2212                                 return;
2213                         }
2214                         pvep = vep;
2215                 }
2216                 cmn_err(CE_PANIC, "vn_vfslocks_rele: vp/vfs not found");
2217         }
2218         mutex_exit(&bp->vb_lock);
2219 }
2220 
2221 /*
2222  * vn_vfswlock_wait is used to implement a lock which is logically a writers
2223  * lock protecting the v_vfsmountedhere field.
2224  * vn_vfswlock_wait has been modified to be similar to vn_vfswlock,
2225  * except that it blocks to acquire the lock VVFSLOCK.
2226  *
2227  * traverse() and routines re-implementing part of traverse (e.g. autofs)
2228  * need to hold this lock. mount(), vn_rename(), vn_remove() and so on
2229  * need the non-blocking version of the writers lock i.e. vn_vfswlock
2230  */
2231 int
2232 vn_vfswlock_wait(vnode_t *vp)
2233 {
2234         int retval;
2235         vn_vfslocks_entry_t *vpvfsentry;
2236         ASSERT(vp != NULL);
2237 
2238         vpvfsentry = vn_vfslocks_getlock(vp);
2239         retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_WRITER);
2240 
2241         if (retval == EINTR) {
2242                 vn_vfslocks_rele(vpvfsentry);
2243                 return (EINTR);
2244         }
2245         return (retval);
2246 }
2247 
2248 int
2249 vn_vfsrlock_wait(vnode_t *vp)
2250 {
2251         int retval;
2252         vn_vfslocks_entry_t *vpvfsentry;
2253         ASSERT(vp != NULL);
2254 
2255         vpvfsentry = vn_vfslocks_getlock(vp);
2256         retval = rwst_enter_sig(&vpvfsentry->ve_lock, RW_READER);
2257 
2258         if (retval == EINTR) {
2259                 vn_vfslocks_rele(vpvfsentry);
2260                 return (EINTR);
2261         }
2262 
2263         return (retval);
2264 }
2265 
2266 
2267 /*
2268  * vn_vfswlock is used to implement a lock which is logically a writers lock
2269  * protecting the v_vfsmountedhere field.
2270  */
2271 int
2272 vn_vfswlock(vnode_t *vp)
2273 {
2274         vn_vfslocks_entry_t *vpvfsentry;
2275 
2276         /*
2277          * If vp is NULL then somebody is trying to lock the covered vnode
2278          * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2279          * only happen when unmounting /.  Since that operation will fail
2280          * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2281          */
2282         if (vp == NULL)
2283                 return (EBUSY);
2284 
2285         vpvfsentry = vn_vfslocks_getlock(vp);
2286 
2287         if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER))
2288                 return (0);
2289 
2290         vn_vfslocks_rele(vpvfsentry);
2291         return (EBUSY);
2292 }
2293 
2294 int
2295 vn_vfsrlock(vnode_t *vp)
2296 {
2297         vn_vfslocks_entry_t *vpvfsentry;
2298 
2299         /*
2300          * If vp is NULL then somebody is trying to lock the covered vnode
2301          * of /.  (vfs_vnodecovered is NULL for /).  This situation will
2302          * only happen when unmounting /.  Since that operation will fail
2303          * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2304          */
2305         if (vp == NULL)
2306                 return (EBUSY);
2307 
2308         vpvfsentry = vn_vfslocks_getlock(vp);
2309 
2310         if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER))
2311                 return (0);
2312 
2313         vn_vfslocks_rele(vpvfsentry);
2314         return (EBUSY);
2315 }
2316 
2317 void
2318 vn_vfsunlock(vnode_t *vp)
2319 {
2320         vn_vfslocks_entry_t *vpvfsentry;
2321 
2322         /*
2323          * ve_refcnt needs to be decremented twice.
2324          * 1. To release refernce after a call to vn_vfslocks_getlock()
2325          * 2. To release the reference from the locking routines like
2326          *    vn_vfsrlock/vn_vfswlock etc,.
2327          */
2328         vpvfsentry = vn_vfslocks_getlock(vp);
2329         vn_vfslocks_rele(vpvfsentry);
2330 
2331         rwst_exit(&vpvfsentry->ve_lock);
2332         vn_vfslocks_rele(vpvfsentry);
2333 }
2334 
2335 int
2336 vn_vfswlock_held(vnode_t *vp)
2337 {
2338         int held;
2339         vn_vfslocks_entry_t *vpvfsentry;
2340 
2341         ASSERT(vp != NULL);
2342 
2343         vpvfsentry = vn_vfslocks_getlock(vp);
2344         held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER);
2345 
2346         vn_vfslocks_rele(vpvfsentry);
2347         return (held);
2348 }
2349 
2350 
2351 int
2352 vn_make_ops(
2353         const char *name,                       /* Name of file system */
2354         const fs_operation_def_t *templ,        /* Operation specification */
2355         vnodeops_t **actual)                    /* Return the vnodeops */
2356 {
2357         int unused_ops;
2358         int error;
2359 
2360         *actual = (vnodeops_t *)kmem_alloc(sizeof (vnodeops_t), KM_SLEEP);
2361 
2362         (*actual)->vnop_name = name;
2363 
2364         error = fs_build_vector(*actual, &unused_ops, vn_ops_table, templ);
2365         if (error) {
2366                 kmem_free(*actual, sizeof (vnodeops_t));
2367         }
2368 
2369 #if DEBUG
2370         if (unused_ops != 0)
2371                 cmn_err(CE_WARN, "vn_make_ops: %s: %d operations supplied "
2372                     "but not used", name, unused_ops);
2373 #endif
2374 
2375         return (error);
2376 }
2377 
2378 /*
2379  * Free the vnodeops created as a result of vn_make_ops()
2380  */
2381 void
2382 vn_freevnodeops(vnodeops_t *vnops)
2383 {
2384         kmem_free(vnops, sizeof (vnodeops_t));
2385 }
2386 
2387 /*
2388  * Vnode cache.
2389  */
2390 
2391 /* ARGSUSED */
2392 static int
2393 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
2394 {
2395         struct vnode *vp;
2396 
2397         vp = buf;
2398 
2399         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
2400         mutex_init(&vp->v_vsd_lock, NULL, MUTEX_DEFAULT, NULL);
2401         cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
2402         rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
2403         vp->v_femhead = NULL;        /* Must be done before vn_reinit() */
2404         vp->v_path = vn_vpath_empty;
2405         vp->v_path_stamp = 0;
2406         vp->v_mpssdata = NULL;
2407         vp->v_vsd = NULL;
2408         vp->v_fopdata = NULL;
2409 
2410         return (0);
2411 }
2412 
2413 /* ARGSUSED */
2414 static void
2415 vn_cache_destructor(void *buf, void *cdrarg)
2416 {
2417         struct vnode *vp;
2418 
2419         vp = buf;
2420 
2421         rw_destroy(&vp->v_nbllock);
2422         cv_destroy(&vp->v_cv);
2423         mutex_destroy(&vp->v_vsd_lock);
2424         mutex_destroy(&vp->v_lock);
2425 }
2426 
2427 void
2428 vn_create_cache(void)
2429 {
2430         /* LINTED */
2431         ASSERT((1 << VNODE_ALIGN_LOG2) ==
2432             P2ROUNDUP(sizeof (struct vnode), VNODE_ALIGN));
2433         vn_cache = kmem_cache_create("vn_cache", sizeof (struct vnode),
2434             VNODE_ALIGN, vn_cache_constructor, vn_cache_destructor, NULL, NULL,
2435             NULL, 0);
2436 }
2437 
2438 void
2439 vn_destroy_cache(void)
2440 {
2441         kmem_cache_destroy(vn_cache);
2442 }
2443 
2444 /*
2445  * Used by file systems when fs-specific nodes (e.g., ufs inodes) are
2446  * cached by the file system and vnodes remain associated.
2447  */
2448 void
2449 vn_recycle(vnode_t *vp)
2450 {
2451         ASSERT(vp->v_pages == NULL);
2452         VERIFY(vp->v_path != NULL);
2453 
2454         /*
2455          * XXX - This really belongs in vn_reinit(), but we have some issues
2456          * with the counts.  Best to have it here for clean initialization.
2457          */
2458         vp->v_rdcnt = 0;
2459         vp->v_wrcnt = 0;
2460         vp->v_mmap_read = 0;
2461         vp->v_mmap_write = 0;
2462 
2463         /*
2464          * If FEM was in use, make sure everything gets cleaned up
2465          * NOTE: vp->v_femhead is initialized to NULL in the vnode
2466          * constructor.
2467          */
2468         if (vp->v_femhead) {
2469                 /* XXX - There should be a free_femhead() that does all this */
2470                 ASSERT(vp->v_femhead->femh_list == NULL);
2471                 mutex_destroy(&vp->v_femhead->femh_lock);
2472                 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2473                 vp->v_femhead = NULL;
2474         }
2475         if (vp->v_path != vn_vpath_empty) {
2476                 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2477                 vp->v_path = vn_vpath_empty;
2478         }
2479         vp->v_path_stamp = 0;
2480 
2481         if (vp->v_fopdata != NULL) {
2482                 free_fopdata(vp);
2483         }
2484         vp->v_mpssdata = NULL;
2485         vsd_free(vp);
2486 }
2487 
2488 /*
2489  * Used to reset the vnode fields including those that are directly accessible
2490  * as well as those which require an accessor function.
2491  *
2492  * Does not initialize:
2493  *      synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv
2494  *      v_data (since FS-nodes and vnodes point to each other and should
2495  *              be updated simultaneously)
2496  *      v_op (in case someone needs to make a VOP call on this object)
2497  */
2498 void
2499 vn_reinit(vnode_t *vp)
2500 {
2501         vp->v_count = 1;
2502         vp->v_count_dnlc = 0;
2503         vp->v_phantom_count = 0;
2504         vp->v_vfsp = NULL;
2505         vp->v_stream = NULL;
2506         vp->v_vfsmountedhere = NULL;
2507         vp->v_flag = 0;
2508         vp->v_type = VNON;
2509         vp->v_rdev = NODEV;
2510 
2511         vp->v_filocks = NULL;
2512         vp->v_shrlocks = NULL;
2513         vp->v_pages = NULL;
2514 
2515         vp->v_locality = NULL;
2516         vp->v_xattrdir = NULL;
2517 
2518         /*
2519          * In a few specific instances, vn_reinit() is used to initialize
2520          * locally defined vnode_t instances.  Lacking the construction offered
2521          * by vn_alloc(), these vnodes require v_path initialization.
2522          */
2523         if (vp->v_path == NULL) {
2524                 vp->v_path = vn_vpath_empty;
2525         }
2526 
2527         /* Handles v_femhead, v_path, and the r/w/map counts */
2528         vn_recycle(vp);
2529 }
2530 
2531 vnode_t *
2532 vn_alloc(int kmflag)
2533 {
2534         vnode_t *vp;
2535 
2536         vp = kmem_cache_alloc(vn_cache, kmflag);
2537 
2538         if (vp != NULL) {
2539                 vp->v_femhead = NULL;        /* Must be done before vn_reinit() */
2540                 vp->v_fopdata = NULL;
2541                 vn_reinit(vp);
2542         }
2543 
2544         return (vp);
2545 }
2546 
2547 void
2548 vn_free(vnode_t *vp)
2549 {
2550         ASSERT(vp->v_shrlocks == NULL);
2551         ASSERT(vp->v_filocks == NULL);
2552 
2553         /*
2554          * Some file systems call vn_free() with v_count of zero,
2555          * some with v_count of 1.  In any case, the value should
2556          * never be anything else.
2557          */
2558         ASSERT((vp->v_count == 0) || (vp->v_count == 1));
2559         ASSERT(vp->v_count_dnlc == 0);
2560         ASSERT0(vp->v_phantom_count);
2561         VERIFY(vp->v_path != NULL);
2562         if (vp->v_path != vn_vpath_empty) {
2563                 kmem_free(vp->v_path, strlen(vp->v_path) + 1);
2564                 vp->v_path = vn_vpath_empty;
2565         }
2566 
2567         /* If FEM was in use, make sure everything gets cleaned up */
2568         if (vp->v_femhead) {
2569                 /* XXX - There should be a free_femhead() that does all this */
2570                 ASSERT(vp->v_femhead->femh_list == NULL);
2571                 mutex_destroy(&vp->v_femhead->femh_lock);
2572                 kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
2573                 vp->v_femhead = NULL;
2574         }
2575 
2576         if (vp->v_fopdata != NULL) {
2577                 free_fopdata(vp);
2578         }
2579         vp->v_mpssdata = NULL;
2580         vsd_free(vp);
2581         kmem_cache_free(vn_cache, vp);
2582 }
2583 
2584 /*
2585  * vnode status changes, should define better states than 1, 0.
2586  */
2587 void
2588 vn_reclaim(vnode_t *vp)
2589 {
2590         vfs_t   *vfsp = vp->v_vfsp;
2591 
2592         if (vfsp == NULL ||
2593             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2594                 return;
2595         }
2596         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_RECLAIMED);
2597 }
2598 
2599 void
2600 vn_idle(vnode_t *vp)
2601 {
2602         vfs_t   *vfsp = vp->v_vfsp;
2603 
2604         if (vfsp == NULL ||
2605             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2606                 return;
2607         }
2608         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_IDLED);
2609 }
2610 void
2611 vn_exists(vnode_t *vp)
2612 {
2613         vfs_t   *vfsp = vp->v_vfsp;
2614 
2615         if (vfsp == NULL ||
2616             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2617                 return;
2618         }
2619         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_EXISTS);
2620 }
2621 
2622 void
2623 vn_invalid(vnode_t *vp)
2624 {
2625         vfs_t   *vfsp = vp->v_vfsp;
2626 
2627         if (vfsp == NULL ||
2628             vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) {
2629                 return;
2630         }
2631         (void) VFS_VNSTATE(vfsp, vp, VNTRANS_DESTROYED);
2632 }
2633 
2634 /* Vnode event notification */
2635 
2636 int
2637 vnevent_support(vnode_t *vp, caller_context_t *ct)
2638 {
2639         if (vp == NULL)
2640                 return (EINVAL);
2641 
2642         return (VOP_VNEVENT(vp, VE_SUPPORT, NULL, NULL, ct));
2643 }
2644 
2645 void
2646 vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2647 {
2648         if (vp == NULL || vp->v_femhead == NULL) {
2649                 return;
2650         }
2651         (void) VOP_VNEVENT(dvp, VE_RENAME_SRC_DIR, vp, name, ct);
2652         (void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
2653 }
2654 
2655 void
2656 vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
2657     caller_context_t *ct)
2658 {
2659         if (vp == NULL || vp->v_femhead == NULL) {
2660                 return;
2661         }
2662         (void) VOP_VNEVENT(vp, VE_RENAME_DEST, dvp, name, ct);
2663 }
2664 
2665 void
2666 vnevent_rename_dest_dir(vnode_t *vp, vnode_t *nvp, char *name,
2667     caller_context_t *ct)
2668 {
2669         if (vp == NULL || vp->v_femhead == NULL) {
2670                 return;
2671         }
2672         (void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, nvp, name, ct);
2673 }
2674 
2675 void
2676 vnevent_remove(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2677 {
2678         if (vp == NULL || vp->v_femhead == NULL) {
2679                 return;
2680         }
2681         (void) VOP_VNEVENT(vp, VE_REMOVE, dvp, name, ct);
2682 }
2683 
2684 void
2685 vnevent_rmdir(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
2686 {
2687         if (vp == NULL || vp->v_femhead == NULL) {
2688                 return;
2689         }
2690         (void) VOP_VNEVENT(vp, VE_RMDIR, dvp, name, ct);
2691 }
2692 
2693 void
2694 vnevent_pre_rename_src(vnode_t *vp, vnode_t *dvp, char *name,
2695     caller_context_t *ct)
2696 {
2697         if (vp == NULL || vp->v_femhead == NULL) {
2698                 return;
2699         }
2700         (void) VOP_VNEVENT(vp, VE_PRE_RENAME_SRC, dvp, name, ct);
2701 }
2702 
2703 void
2704 vnevent_pre_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
2705     caller_context_t *ct)
2706 {
2707         if (vp == NULL || vp->v_femhead == NULL) {
2708                 return;
2709         }
2710         (void) VOP_VNEVENT(vp, VE_PRE_RENAME_DEST, dvp, name, ct);
2711 }
2712 
2713 void
2714 vnevent_pre_rename_dest_dir(vnode_t *vp, vnode_t *nvp, char *name,
2715     caller_context_t *ct)
2716 {
2717         if (vp == NULL || vp->v_femhead == NULL) {
2718                 return;
2719         }
2720         (void) VOP_VNEVENT(vp, VE_PRE_RENAME_DEST_DIR, nvp, name, ct);
2721 }
2722 
2723 void
2724 vnevent_create(vnode_t *vp, caller_context_t *ct)
2725 {
2726         if (vp == NULL || vp->v_femhead == NULL) {
2727                 return;
2728         }
2729         (void) VOP_VNEVENT(vp, VE_CREATE, NULL, NULL, ct);
2730 }
2731 
2732 void
2733 vnevent_link(vnode_t *vp, caller_context_t *ct)
2734 {
2735         if (vp == NULL || vp->v_femhead == NULL) {
2736                 return;
2737         }
2738         (void) VOP_VNEVENT(vp, VE_LINK, NULL, NULL, ct);
2739 }
2740 
2741 void
2742 vnevent_mountedover(vnode_t *vp, caller_context_t *ct)
2743 {
2744         if (vp == NULL || vp->v_femhead == NULL) {
2745                 return;
2746         }
2747         (void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct);
2748 }
2749 
2750 void
2751 vnevent_truncate(vnode_t *vp, caller_context_t *ct)
2752 {
2753         if (vp == NULL || vp->v_femhead == NULL) {
2754                 return;
2755         }
2756         (void) VOP_VNEVENT(vp, VE_TRUNCATE, NULL, NULL, ct);
2757 }
2758 
2759 void
2760 vnevent_resize(vnode_t *vp, caller_context_t *ct)
2761 {
2762         if (vp == NULL || vp->v_femhead == NULL) {
2763                 return;
2764         }
2765         (void) VOP_VNEVENT(vp, VE_RESIZE, NULL, NULL, ct);
2766 }
2767 
2768 /*
2769  * Vnode accessors.
2770  */
2771 
2772 int
2773 vn_is_readonly(vnode_t *vp)
2774 {
2775         return (vp->v_vfsp->vfs_flag & VFS_RDONLY);
2776 }
2777 
2778 int
2779 vn_has_flocks(vnode_t *vp)
2780 {
2781         return (vp->v_filocks != NULL);
2782 }
2783 
2784 int
2785 vn_has_mandatory_locks(vnode_t *vp, int mode)
2786 {
2787         return ((vp->v_filocks != NULL) && (MANDLOCK(vp, mode)));
2788 }
2789 
2790 int
2791 vn_has_cached_data(vnode_t *vp)
2792 {
2793         return (vp->v_pages != NULL);
2794 }
2795 
2796 /*
2797  * Return 0 if the vnode in question shouldn't be permitted into a zone via
2798  * zone_enter(2).
2799  */
2800 int
2801 vn_can_change_zones(vnode_t *vp)
2802 {
2803         struct vfssw *vswp;
2804         int allow = 1;
2805         vnode_t *rvp;
2806 
2807         if (nfs_global_client_only != 0)
2808                 return (1);
2809 
2810         /*
2811          * We always want to look at the underlying vnode if there is one.
2812          */
2813         if (VOP_REALVP(vp, &rvp, NULL) != 0)
2814                 rvp = vp;
2815         /*
2816          * Some pseudo filesystems (including doorfs) don't actually register
2817          * their vfsops_t, so the following may return NULL; we happily let
2818          * such vnodes switch zones.
2819          */
2820         vswp = vfs_getvfsswbyvfsops(vfs_getops(rvp->v_vfsp));
2821         if (vswp != NULL) {
2822                 if (vswp->vsw_flag & VSW_NOTZONESAFE)
2823                         allow = 0;
2824                 vfs_unrefvfssw(vswp);
2825         }
2826         return (allow);
2827 }
2828 
2829 /*
2830  * Return nonzero if the vnode is a mount point, zero if not.
2831  */
2832 int
2833 vn_ismntpt(vnode_t *vp)
2834 {
2835         return (vp->v_vfsmountedhere != NULL);
2836 }
2837 
2838 /* Retrieve the vfs (if any) mounted on this vnode */
2839 vfs_t *
2840 vn_mountedvfs(vnode_t *vp)
2841 {
2842         return (vp->v_vfsmountedhere);
2843 }
2844 
2845 /*
2846  * Return nonzero if the vnode is referenced by the dnlc, zero if not.
2847  */
2848 int
2849 vn_in_dnlc(vnode_t *vp)
2850 {
2851         return (vp->v_count_dnlc > 0);
2852 }
2853 
2854 /*
2855  * vn_has_other_opens() checks whether a particular file is opened by more than
2856  * just the caller and whether the open is for read and/or write.
2857  * This routine is for calling after the caller has already called VOP_OPEN()
2858  * and the caller wishes to know if they are the only one with it open for
2859  * the mode(s) specified.
2860  *
2861  * Vnode counts are only kept on regular files (v_type=VREG).
2862  */
2863 int
2864 vn_has_other_opens(
2865         vnode_t *vp,
2866         v_mode_t mode)
2867 {
2868 
2869         ASSERT(vp != NULL);
2870 
2871         switch (mode) {
2872         case V_WRITE:
2873                 if (vp->v_wrcnt > 1)
2874                         return (V_TRUE);
2875                 break;
2876         case V_RDORWR:
2877                 if ((vp->v_rdcnt > 1) || (vp->v_wrcnt > 1))
2878                         return (V_TRUE);
2879                 break;
2880         case V_RDANDWR:
2881                 if ((vp->v_rdcnt > 1) && (vp->v_wrcnt > 1))
2882                         return (V_TRUE);
2883                 break;
2884         case V_READ:
2885                 if (vp->v_rdcnt > 1)
2886                         return (V_TRUE);
2887                 break;
2888         }
2889 
2890         return (V_FALSE);
2891 }
2892 
2893 /*
2894  * vn_is_opened() checks whether a particular file is opened and
2895  * whether the open is for read and/or write.
2896  *
2897  * Vnode counts are only kept on regular files (v_type=VREG).
2898  */
2899 int
2900 vn_is_opened(
2901         vnode_t *vp,
2902         v_mode_t mode)
2903 {
2904 
2905         ASSERT(vp != NULL);
2906 
2907         switch (mode) {
2908         case V_WRITE:
2909                 if (vp->v_wrcnt)
2910                         return (V_TRUE);
2911                 break;
2912         case V_RDANDWR:
2913                 if (vp->v_rdcnt && vp->v_wrcnt)
2914                         return (V_TRUE);
2915                 break;
2916         case V_RDORWR:
2917                 if (vp->v_rdcnt || vp->v_wrcnt)
2918                         return (V_TRUE);
2919                 break;
2920         case V_READ:
2921                 if (vp->v_rdcnt)
2922                         return (V_TRUE);
2923                 break;
2924         }
2925 
2926         return (V_FALSE);
2927 }
2928 
2929 /*
2930  * vn_is_mapped() checks whether a particular file is mapped and whether
2931  * the file is mapped read and/or write.
2932  */
2933 int
2934 vn_is_mapped(
2935         vnode_t *vp,
2936         v_mode_t mode)
2937 {
2938 
2939         ASSERT(vp != NULL);
2940 
2941 #if !defined(_LP64)
2942         switch (mode) {
2943         /*
2944          * The atomic_add_64_nv functions force atomicity in the
2945          * case of 32 bit architectures. Otherwise the 64 bit values
2946          * require two fetches. The value of the fields may be
2947          * (potentially) changed between the first fetch and the
2948          * second
2949          */
2950         case V_WRITE:
2951                 if (atomic_add_64_nv((&(vp->v_mmap_write)), 0))
2952                         return (V_TRUE);
2953                 break;
2954         case V_RDANDWR:
2955                 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) &&
2956                     (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2957                         return (V_TRUE);
2958                 break;
2959         case V_RDORWR:
2960                 if ((atomic_add_64_nv((&(vp->v_mmap_read)), 0)) ||
2961                     (atomic_add_64_nv((&(vp->v_mmap_write)), 0)))
2962                         return (V_TRUE);
2963                 break;
2964         case V_READ:
2965                 if (atomic_add_64_nv((&(vp->v_mmap_read)), 0))
2966                         return (V_TRUE);
2967                 break;
2968         }
2969 #else
2970         switch (mode) {
2971         case V_WRITE:
2972                 if (vp->v_mmap_write)
2973                         return (V_TRUE);
2974                 break;
2975         case V_RDANDWR:
2976                 if (vp->v_mmap_read && vp->v_mmap_write)
2977                         return (V_TRUE);
2978                 break;
2979         case V_RDORWR:
2980                 if (vp->v_mmap_read || vp->v_mmap_write)
2981                         return (V_TRUE);
2982                 break;
2983         case V_READ:
2984                 if (vp->v_mmap_read)
2985                         return (V_TRUE);
2986                 break;
2987         }
2988 #endif
2989 
2990         return (V_FALSE);
2991 }
2992 
2993 /*
2994  * Set the operations vector for a vnode.
2995  *
2996  * FEM ensures that the v_femhead pointer is filled in before the
2997  * v_op pointer is changed.  This means that if the v_femhead pointer
2998  * is NULL, and the v_op field hasn't changed since before which checked
2999  * the v_femhead pointer; then our update is ok - we are not racing with
3000  * FEM.
3001  */
3002 void
3003 vn_setops(vnode_t *vp, vnodeops_t *vnodeops)
3004 {
3005         vnodeops_t      *op;
3006 
3007         ASSERT(vp != NULL);
3008         ASSERT(vnodeops != NULL);
3009 
3010         op = vp->v_op;
3011         membar_consumer();
3012         /*
3013          * If vp->v_femhead == NULL, then we'll call atomic_cas_ptr() to do
3014          * the compare-and-swap on vp->v_op.  If either fails, then FEM is
3015          * in effect on the vnode and we need to have FEM deal with it.
3016          */
3017         if (vp->v_femhead != NULL || atomic_cas_ptr(&vp->v_op, op, vnodeops) !=
3018             op) {
3019                 fem_setvnops(vp, vnodeops);
3020         }
3021 }
3022 
3023 /*
3024  * Retrieve the operations vector for a vnode
3025  * As with vn_setops(above); make sure we aren't racing with FEM.
3026  * FEM sets the v_op to a special, internal, vnodeops that wouldn't
3027  * make sense to the callers of this routine.
3028  */
3029 vnodeops_t *
3030 vn_getops(vnode_t *vp)
3031 {
3032         vnodeops_t      *op;
3033 
3034         ASSERT(vp != NULL);
3035 
3036         op = vp->v_op;
3037         membar_consumer();
3038         if (vp->v_femhead == NULL && op == vp->v_op) {
3039                 return (op);
3040         } else {
3041                 return (fem_getvnops(vp));
3042         }
3043 }
3044 
3045 /*
3046  * Returns non-zero (1) if the vnodeops matches that of the vnode.
3047  * Returns zero (0) if not.
3048  */
3049 int
3050 vn_matchops(vnode_t *vp, vnodeops_t *vnodeops)
3051 {
3052         return (vn_getops(vp) == vnodeops);
3053 }
3054 
3055 /*
3056  * Returns non-zero (1) if the specified operation matches the
3057  * corresponding operation for that the vnode.
3058  * Returns zero (0) if not.
3059  */
3060 
3061 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0))
3062 
3063 int
3064 vn_matchopval(vnode_t *vp, char *vopname, fs_generic_func_p funcp)
3065 {
3066         const fs_operation_trans_def_t *otdp;
3067         fs_generic_func_p *loc = NULL;
3068         vnodeops_t      *vop = vn_getops(vp);
3069 
3070         ASSERT(vopname != NULL);
3071 
3072         for (otdp = vn_ops_table; otdp->name != NULL; otdp++) {
3073                 if (MATCHNAME(otdp->name, vopname)) {
3074                         loc = (fs_generic_func_p *)
3075                             ((char *)(vop) + otdp->offset);
3076                         break;
3077                 }
3078         }
3079 
3080         return ((loc != NULL) && (*loc == funcp));
3081 }
3082 
3083 /*
3084  * fs_new_caller_id() needs to return a unique ID on a given local system.
3085  * The IDs do not need to survive across reboots.  These are primarily
3086  * used so that (FEM) monitors can detect particular callers (such as
3087  * the NFS server) to a given vnode/vfs operation.
3088  */
3089 u_longlong_t
3090 fs_new_caller_id()
3091 {
3092         static uint64_t next_caller_id = 0LL; /* First call returns 1 */
3093 
3094         return ((u_longlong_t)atomic_inc_64_nv(&next_caller_id));
3095 }
3096 
3097 /*
3098  * The value stored in v_path is relative to rootdir, located in the global
3099  * zone.  Zones or chroot environments which reside deeper inside the VFS
3100  * hierarchy will have a relative view of MAXPATHLEN since they are unaware of
3101  * what lies below their perceived root.  In order to keep v_path usable for
3102  * these child environments, its allocations are allowed to exceed MAXPATHLEN.
3103  *
3104  * An upper bound of max_vnode_path is placed upon v_path allocations to
3105  * prevent the system from going too wild at the behest of pathological
3106  * behavior from the operator.
3107  */
3108 size_t max_vnode_path = 4 * MAXPATHLEN;
3109 
3110 
3111 void
3112 vn_clearpath(vnode_t *vp, hrtime_t compare_stamp)
3113 {
3114         char *buf;
3115 
3116         mutex_enter(&vp->v_lock);
3117         /*
3118          * If the snapshot of v_path_stamp passed in via compare_stamp does not
3119          * match the present value on the vnode, it indicates that subsequent
3120          * changes have occurred.  The v_path value is not cleared in this case
3121          * since the new value may be valid.
3122          */
3123         if (compare_stamp != 0 && vp->v_path_stamp != compare_stamp) {
3124                 mutex_exit(&vp->v_lock);
3125                 return;
3126         }
3127         buf = vp->v_path;
3128         vp->v_path = vn_vpath_empty;
3129         vp->v_path_stamp = 0;
3130         mutex_exit(&vp->v_lock);
3131         if (buf != vn_vpath_empty) {
3132                 kmem_free(buf, strlen(buf) + 1);
3133         }
3134 }
3135 
3136 static void
3137 vn_setpath_common(vnode_t *pvp, vnode_t *vp, const char *name, size_t len,
3138     boolean_t is_rename)
3139 {
3140         char *buf, *oldbuf;
3141         hrtime_t pstamp;
3142         size_t baselen, buflen = 0;
3143 
3144         /* Handle the vn_setpath_str case. */
3145         if (pvp == NULL) {
3146                 if (len + 1 > max_vnode_path) {
3147                         DTRACE_PROBE4(vn__setpath__too__long, vnode_t *, pvp,
3148                             vnode_t *, vp, char *, name, size_t, len + 1);
3149                         return;
3150                 }
3151                 buf = kmem_alloc(len + 1, KM_SLEEP);
3152                 bcopy(name, buf, len);
3153                 buf[len] = '\0';
3154 
3155                 mutex_enter(&vp->v_lock);
3156                 oldbuf = vp->v_path;
3157                 vp->v_path = buf;
3158                 vp->v_path_stamp = gethrtime();
3159                 mutex_exit(&vp->v_lock);
3160                 if (oldbuf != vn_vpath_empty) {
3161                         kmem_free(oldbuf, strlen(oldbuf) + 1);
3162                 }
3163                 return;
3164         }
3165 
3166         /* Take snapshot of parent dir */
3167         mutex_enter(&pvp->v_lock);
3168 
3169         if ((pvp->v_flag & VTRAVERSE) != 0) {
3170                 /*
3171                  * When the parent vnode has VTRAVERSE set in its flags, normal
3172                  * assumptions about v_path calculation no longer apply.  The
3173                  * primary situation where this occurs is via the VFS tricks
3174                  * which procfs plays in order to allow /proc/PID/(root|cwd) to
3175                  * yield meaningful results.
3176                  *
3177                  * When this flag is set, v_path on the child must not be
3178                  * updated since the calculated value is likely to be
3179                  * incorrect, given the current context.
3180                  */
3181                 mutex_exit(&pvp->v_lock);
3182                 return;
3183         }
3184 
3185 retrybuf:
3186         if (pvp->v_path == vn_vpath_empty) {
3187                 /*
3188                  * Without v_path from the parent directory, generating a child
3189                  * path from the name is impossible.
3190                  */
3191                 if (len > 0) {
3192                         pstamp = pvp->v_path_stamp;
3193                         mutex_exit(&pvp->v_lock);
3194                         vn_clearpath(vp, pstamp);
3195                         return;
3196                 }
3197 
3198                 /*
3199                  * The only feasible case here is where a NUL lookup is being
3200                  * performed on rootdir prior to its v_path being populated.
3201                  */
3202                 ASSERT(pvp->v_path_stamp == 0);
3203                 baselen = 0;
3204                 pstamp = 0;
3205         } else {
3206                 pstamp = pvp->v_path_stamp;
3207                 baselen = strlen(pvp->v_path);
3208                 /* ignore a trailing slash if present */
3209                 if (pvp->v_path[baselen - 1] == '/') {
3210                         /* This should only the be case for rootdir */
3211                         ASSERT(baselen == 1 && pvp == rootdir);
3212                         baselen--;
3213                 }
3214         }
3215         mutex_exit(&pvp->v_lock);
3216 
3217         if (buflen != 0) {
3218                 /* Free the existing (mis-sized) buffer in case of retry */
3219                 kmem_free(buf, buflen);
3220         }
3221         /* base, '/', name and trailing NUL */
3222         buflen = baselen + len + 2;
3223         if (buflen > max_vnode_path) {
3224                 DTRACE_PROBE4(vn__setpath_too__long, vnode_t *, pvp,
3225                     vnode_t *, vp, char *, name, size_t, buflen);
3226                 return;
3227         }
3228         buf = kmem_alloc(buflen, KM_SLEEP);
3229 
3230         mutex_enter(&pvp->v_lock);
3231         if (pvp->v_path_stamp != pstamp) {
3232                 size_t vlen;
3233 
3234                 /*
3235                  * Since v_path_stamp changed on the parent, it is likely that
3236                  * v_path has been altered as well.  If the length does not
3237                  * exactly match what was previously measured, the buffer
3238                  * allocation must be repeated for proper sizing.
3239                  */
3240                 if (pvp->v_path == vn_vpath_empty) {
3241                         /* Give up if parent lack v_path */
3242                         mutex_exit(&pvp->v_lock);
3243                         kmem_free(buf, buflen);
3244                         return;
3245                 }
3246                 vlen = strlen(pvp->v_path);
3247                 if (pvp->v_path[vlen - 1] == '/') {
3248                         vlen--;
3249                 }
3250                 if (vlen != baselen) {
3251                         goto retrybuf;
3252                 }
3253         }
3254         bcopy(pvp->v_path, buf, baselen);
3255         mutex_exit(&pvp->v_lock);
3256 
3257         buf[baselen] = '/';
3258         baselen++;
3259         bcopy(name, &buf[baselen], len + 1);
3260 
3261         mutex_enter(&vp->v_lock);
3262         if (vp->v_path_stamp == 0) {
3263                 /* never-visited vnode can inherit stamp from parent */
3264                 ASSERT(vp->v_path == vn_vpath_empty);
3265                 vp->v_path_stamp = pstamp;
3266                 vp->v_path = buf;
3267                 mutex_exit(&vp->v_lock);
3268         } else if (vp->v_path_stamp < pstamp || is_rename) {
3269                 /*
3270                  * Install the updated path and stamp, ensuring that the v_path
3271                  * pointer is valid at all times for dtrace.
3272                  */
3273                 oldbuf = vp->v_path;
3274                 vp->v_path = buf;
3275                 vp->v_path_stamp = gethrtime();
3276                 mutex_exit(&vp->v_lock);
3277                 kmem_free(oldbuf, strlen(oldbuf) + 1);
3278         } else {
3279                 /*
3280                  * If the timestamp matches or is greater, it means another
3281                  * thread performed the update first while locks were dropped
3282                  * here to make the allocation.  We defer to the newer value.
3283                  */
3284                 mutex_exit(&vp->v_lock);
3285                 kmem_free(buf, buflen);
3286         }
3287         ASSERT(MUTEX_NOT_HELD(&vp->v_lock));
3288 }
3289 
3290 void
3291 vn_updatepath(vnode_t *pvp, vnode_t *vp, const char *name)
3292 {
3293         size_t len;
3294 
3295         /*
3296          * If the parent is older or empty, there's nothing further to do.
3297          */
3298         if (pvp->v_path == vn_vpath_empty ||
3299             pvp->v_path_stamp <= vp->v_path_stamp) {
3300                 return;
3301         }
3302 
3303         /*
3304          * Given the lack of appropriate context, meaningful updates to v_path
3305          * cannot be made for during lookups for the '.' or '..' entries.
3306          */
3307         len = strlen(name);
3308         if (len == 0 || (len == 1 && name[0] == '.') ||
3309             (len == 2 && name[0] == '.' && name[1] == '.')) {
3310                 return;
3311         }
3312 
3313         vn_setpath_common(pvp, vp, name, len, B_FALSE);
3314 }
3315 
3316 /*
3317  * Given a starting vnode and a path, updates the path in the target vnode in
3318  * a safe manner.  If the vnode already has path information embedded, then the
3319  * cached path is left untouched.
3320  */
3321 /* ARGSUSED */
3322 void
3323 vn_setpath(vnode_t *rootvp, vnode_t *pvp, vnode_t *vp, const char *name,
3324     size_t len)
3325 {
3326         vn_setpath_common(pvp, vp, name, len, B_FALSE);
3327 }
3328 
3329 /*
3330  * Sets the path to the vnode to be the given string, regardless of current
3331  * context.  The string must be a complete path from rootdir.  This is only used
3332  * by fsop_root() for setting the path based on the mountpoint.
3333  */
3334 void
3335 vn_setpath_str(vnode_t *vp, const char *str, size_t len)
3336 {
3337         vn_setpath_common(NULL, vp, str, len, B_FALSE);
3338 }
3339 
3340 /*
3341  * Called from within filesystem's vop_rename() to handle renames once the
3342  * target vnode is available.
3343  */
3344 void
3345 vn_renamepath(vnode_t *pvp, vnode_t *vp, const char *name, size_t len)
3346 {
3347         vn_setpath_common(pvp, vp, name, len, B_TRUE);
3348 }
3349 
3350 /*
3351  * Similar to vn_setpath_str(), this function sets the path of the destination
3352  * vnode to the be the same as the source vnode.
3353  */
3354 void
3355 vn_copypath(struct vnode *src, struct vnode *dst)
3356 {
3357         char *buf;
3358         hrtime_t stamp;
3359         size_t buflen;
3360 
3361         mutex_enter(&src->v_lock);
3362         if (src->v_path == vn_vpath_empty) {
3363                 mutex_exit(&src->v_lock);
3364                 return;
3365         }
3366         buflen = strlen(src->v_path) + 1;
3367         mutex_exit(&src->v_lock);
3368 
3369         buf = kmem_alloc(buflen, KM_SLEEP);
3370 
3371         mutex_enter(&src->v_lock);
3372         if (src->v_path == vn_vpath_empty ||
3373             strlen(src->v_path) + 1 != buflen) {
3374                 mutex_exit(&src->v_lock);
3375                 kmem_free(buf, buflen);
3376                 return;
3377         }
3378         bcopy(src->v_path, buf, buflen);
3379         stamp = src->v_path_stamp;
3380         mutex_exit(&src->v_lock);
3381 
3382         mutex_enter(&dst->v_lock);
3383         if (dst->v_path != vn_vpath_empty) {
3384                 mutex_exit(&dst->v_lock);
3385                 kmem_free(buf, buflen);
3386                 return;
3387         }
3388         dst->v_path = buf;
3389         dst->v_path_stamp = stamp;
3390         mutex_exit(&dst->v_lock);
3391 }
3392 
3393 
3394 /*
3395  * XXX Private interface for segvn routines that handle vnode
3396  * large page segments.
3397  *
3398  * return 1 if vp's file system VOP_PAGEIO() implementation
3399  * can be safely used instead of VOP_GETPAGE() for handling
3400  * pagefaults against regular non swap files. VOP_PAGEIO()
3401  * interface is considered safe here if its implementation
3402  * is very close to VOP_GETPAGE() implementation.
3403  * e.g. It zero's out the part of the page beyond EOF. Doesn't
3404  * panic if there're file holes but instead returns an error.
3405  * Doesn't assume file won't be changed by user writes, etc.
3406  *
3407  * return 0 otherwise.
3408  *
3409  * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs.
3410  */
3411 int
3412 vn_vmpss_usepageio(vnode_t *vp)
3413 {
3414         vfs_t   *vfsp = vp->v_vfsp;
3415         char *fsname = vfssw[vfsp->vfs_fstype].vsw_name;
3416         char *pageio_ok_fss[] = {"ufs", "nfs", NULL};
3417         char **fsok = pageio_ok_fss;
3418 
3419         if (fsname == NULL) {
3420                 return (0);
3421         }
3422 
3423         for (; *fsok; fsok++) {
3424                 if (strcmp(*fsok, fsname) == 0) {
3425                         return (1);
3426                 }
3427         }
3428         return (0);
3429 }
3430 
3431 /* VOP_XXX() macros call the corresponding fop_xxx() function */
3432 
3433 int
3434 fop_open(
3435         vnode_t **vpp,
3436         int mode,
3437         cred_t *cr,
3438         caller_context_t *ct)
3439 {
3440         int ret;
3441         vnode_t *vp = *vpp;
3442 
3443         VN_HOLD(vp);
3444         /*
3445          * Adding to the vnode counts before calling open
3446          * avoids the need for a mutex. It circumvents a race
3447          * condition where a query made on the vnode counts results in a
3448          * false negative. The inquirer goes away believing the file is
3449          * not open when there is an open on the file already under way.
3450          *
3451          * The counts are meant to prevent NFS from granting a delegation
3452          * when it would be dangerous to do so.
3453          *
3454          * The vnode counts are only kept on regular files
3455          */
3456         if ((*vpp)->v_type == VREG) {
3457                 if (mode & FREAD)
3458                         atomic_inc_32(&(*vpp)->v_rdcnt);
3459                 if (mode & FWRITE)
3460                         atomic_inc_32(&(*vpp)->v_wrcnt);
3461         }
3462 
3463         VOPXID_MAP_CR(vp, cr);
3464 
3465         ret = (*(*(vpp))->v_op->vop_open)(vpp, mode, cr, ct);
3466 
3467         if (ret) {
3468                 /*
3469                  * Use the saved vp just in case the vnode ptr got trashed
3470                  * by the error.
3471                  */
3472                 VOPSTATS_UPDATE(vp, open);
3473                 if ((vp->v_type == VREG) && (mode & FREAD))
3474                         atomic_dec_32(&vp->v_rdcnt);
3475                 if ((vp->v_type == VREG) && (mode & FWRITE))
3476                         atomic_dec_32(&vp->v_wrcnt);
3477         } else {
3478                 /*
3479                  * Some filesystems will return a different vnode,
3480                  * but the same path was still used to open it.
3481                  * So if we do change the vnode and need to
3482                  * copy over the path, do so here, rather than special
3483                  * casing each filesystem. Adjust the vnode counts to
3484                  * reflect the vnode switch.
3485                  */
3486                 VOPSTATS_UPDATE(*vpp, open);
3487                 if (*vpp != vp) {
3488                         vn_copypath(vp, *vpp);
3489                         if (((*vpp)->v_type == VREG) && (mode & FREAD))
3490                                 atomic_inc_32(&(*vpp)->v_rdcnt);
3491                         if ((vp->v_type == VREG) && (mode & FREAD))
3492                                 atomic_dec_32(&vp->v_rdcnt);
3493                         if (((*vpp)->v_type == VREG) && (mode & FWRITE))
3494                                 atomic_inc_32(&(*vpp)->v_wrcnt);
3495                         if ((vp->v_type == VREG) && (mode & FWRITE))
3496                                 atomic_dec_32(&vp->v_wrcnt);
3497                 }
3498         }
3499         VN_RELE(vp);
3500         return (ret);
3501 }
3502 
3503 int
3504 fop_close(
3505         vnode_t *vp,
3506         int flag,
3507         int count,
3508         offset_t offset,
3509         cred_t *cr,
3510         caller_context_t *ct)
3511 {
3512         int err;
3513 
3514         VOPXID_MAP_CR(vp, cr);
3515 
3516         err = (*(vp)->v_op->vop_close)(vp, flag, count, offset, cr, ct);
3517         VOPSTATS_UPDATE(vp, close);
3518         /*
3519          * Check passed in count to handle possible dups. Vnode counts are only
3520          * kept on regular files
3521          */
3522         if ((vp->v_type == VREG) && (count == 1))  {
3523                 if (flag & FREAD) {
3524                         ASSERT(vp->v_rdcnt > 0);
3525                         atomic_dec_32(&vp->v_rdcnt);
3526                 }
3527                 if (flag & FWRITE) {
3528                         ASSERT(vp->v_wrcnt > 0);
3529                         atomic_dec_32(&vp->v_wrcnt);
3530                 }
3531         }
3532         return (err);
3533 }
3534 
3535 int
3536 fop_read(
3537         vnode_t *vp,
3538         uio_t *uiop,
3539         int ioflag,
3540         cred_t *cr,
3541         caller_context_t *ct)
3542 {
3543         ssize_t resid_start = uiop->uio_resid;
3544         zone_t  *zonep = curzone;
3545         zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
3546 
3547         hrtime_t start = 0, lat;
3548         ssize_t len;
3549         int err;
3550 
3551         if ((vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) &&
3552             vp->v_vfsp != NULL && (vp->v_vfsp->vfs_flag & VFS_STATS)) {
3553                 start = gethrtime();
3554 
3555                 mutex_enter(&zonep->zone_vfs_lock);
3556                 kstat_runq_enter(&zonep->zone_vfs_rwstats);
3557                 mutex_exit(&zonep->zone_vfs_lock);
3558         }
3559 
3560         VOPXID_MAP_CR(vp, cr);
3561 
3562         err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
3563         len = resid_start - uiop->uio_resid;
3564 
3565         VOPSTATS_UPDATE_IO(vp, read, read_bytes, len);
3566 
3567         if (start != 0) {
3568                 mutex_enter(&zonep->zone_vfs_lock);
3569                 zonep->zone_vfs_rwstats.reads++;
3570                 zonep->zone_vfs_rwstats.nread += len;
3571                 kstat_runq_exit(&zonep->zone_vfs_rwstats);
3572                 mutex_exit(&zonep->zone_vfs_lock);
3573 
3574                 lat = gethrtime() - start;
3575 
3576                 if (lat >= VOP_LATENCY_10MS) {
3577                         if (lat < VOP_LATENCY_100MS)
3578                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3579                         else if (lat < VOP_LATENCY_1S) {
3580                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3581                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3582                         } else if (lat < VOP_LATENCY_10S) {
3583                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3584                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3585                                 atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
3586                         } else {
3587                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3588                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3589                                 atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
3590                                 atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
3591                         }
3592                 }
3593         }
3594 
3595         return (err);
3596 }
3597 
3598 int
3599 fop_write(
3600         vnode_t *vp,
3601         uio_t *uiop,
3602         int ioflag,
3603         cred_t *cr,
3604         caller_context_t *ct)
3605 {
3606         ssize_t resid_start = uiop->uio_resid;
3607         zone_t  *zonep = curzone;
3608         zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
3609 
3610         hrtime_t start = 0, lat;
3611         ssize_t len;
3612         int     err;
3613 
3614         /*
3615          * For the purposes of VFS kstat consumers, the "waitq" calculation is
3616          * repurposed as the active queue for VFS write operations.  There's no
3617          * actual wait queue for VFS operations.
3618          */
3619         if ((vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) &&
3620             vp->v_vfsp != NULL && (vp->v_vfsp->vfs_flag & VFS_STATS)) {
3621                 start = gethrtime();
3622 
3623                 mutex_enter(&zonep->zone_vfs_lock);
3624                 kstat_waitq_enter(&zonep->zone_vfs_rwstats);
3625                 mutex_exit(&zonep->zone_vfs_lock);
3626         }
3627 
3628         VOPXID_MAP_CR(vp, cr);
3629 
3630         err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
3631         len = resid_start - uiop->uio_resid;
3632 
3633         VOPSTATS_UPDATE_IO(vp, write, write_bytes, len);
3634 
3635         if (start != 0) {
3636                 mutex_enter(&zonep->zone_vfs_lock);
3637                 zonep->zone_vfs_rwstats.writes++;
3638                 zonep->zone_vfs_rwstats.nwritten += len;
3639                 kstat_waitq_exit(&zonep->zone_vfs_rwstats);
3640                 mutex_exit(&zonep->zone_vfs_lock);
3641 
3642                 lat = gethrtime() - start;
3643 
3644                 if (lat >= VOP_LATENCY_10MS) {
3645                         if (lat < VOP_LATENCY_100MS)
3646                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3647                         else if (lat < VOP_LATENCY_1S) {
3648                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3649                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3650                         } else if (lat < VOP_LATENCY_10S) {
3651                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3652                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3653                                 atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
3654                         } else {
3655                                 atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
3656                                 atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
3657                                 atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
3658                                 atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
3659                         }
3660                 }
3661         }
3662 
3663         return (err);
3664 }
3665 
3666 int
3667 fop_ioctl(
3668         vnode_t *vp,
3669         int cmd,
3670         intptr_t arg,
3671         int flag,
3672         cred_t *cr,
3673         int *rvalp,
3674         caller_context_t *ct)
3675 {
3676         int     err;
3677 
3678         VOPXID_MAP_CR(vp, cr);
3679 
3680         err = (*(vp)->v_op->vop_ioctl)(vp, cmd, arg, flag, cr, rvalp, ct);
3681         VOPSTATS_UPDATE(vp, ioctl);
3682         return (err);
3683 }
3684 
3685 int
3686 fop_setfl(
3687         vnode_t *vp,
3688         int oflags,
3689         int nflags,
3690         cred_t *cr,
3691         caller_context_t *ct)
3692 {
3693         int     err;
3694 
3695         VOPXID_MAP_CR(vp, cr);
3696 
3697         err = (*(vp)->v_op->vop_setfl)(vp, oflags, nflags, cr, ct);
3698         VOPSTATS_UPDATE(vp, setfl);
3699         return (err);
3700 }
3701 
3702 int
3703 fop_getattr(
3704         vnode_t *vp,
3705         vattr_t *vap,
3706         int flags,
3707         cred_t *cr,
3708         caller_context_t *ct)
3709 {
3710         int     err;
3711 
3712         VOPXID_MAP_CR(vp, cr);
3713 
3714         /*
3715          * If this file system doesn't understand the xvattr extensions
3716          * then turn off the xvattr bit.
3717          */
3718         if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3719                 vap->va_mask &= ~AT_XVATTR;
3720         }
3721 
3722         /*
3723          * We're only allowed to skip the ACL check iff we used a 32 bit
3724          * ACE mask with VOP_ACCESS() to determine permissions.
3725          */
3726         if ((flags & ATTR_NOACLCHECK) &&
3727             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3728                 return (EINVAL);
3729         }
3730         err = (*(vp)->v_op->vop_getattr)(vp, vap, flags, cr, ct);
3731         VOPSTATS_UPDATE(vp, getattr);
3732         return (err);
3733 }
3734 
3735 int
3736 fop_setattr(
3737         vnode_t *vp,
3738         vattr_t *vap,
3739         int flags,
3740         cred_t *cr,
3741         caller_context_t *ct)
3742 {
3743         int     err;
3744 
3745         VOPXID_MAP_CR(vp, cr);
3746 
3747         /*
3748          * If this file system doesn't understand the xvattr extensions
3749          * then turn off the xvattr bit.
3750          */
3751         if (vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) == 0) {
3752                 vap->va_mask &= ~AT_XVATTR;
3753         }
3754 
3755         /*
3756          * We're only allowed to skip the ACL check iff we used a 32 bit
3757          * ACE mask with VOP_ACCESS() to determine permissions.
3758          */
3759         if ((flags & ATTR_NOACLCHECK) &&
3760             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3761                 return (EINVAL);
3762         }
3763         err = (*(vp)->v_op->vop_setattr)(vp, vap, flags, cr, ct);
3764         VOPSTATS_UPDATE(vp, setattr);
3765         return (err);
3766 }
3767 
3768 int
3769 fop_access(
3770         vnode_t *vp,
3771         int mode,
3772         int flags,
3773         cred_t *cr,
3774         caller_context_t *ct)
3775 {
3776         int     err;
3777 
3778         if ((flags & V_ACE_MASK) &&
3779             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
3780                 return (EINVAL);
3781         }
3782 
3783         VOPXID_MAP_CR(vp, cr);
3784 
3785         err = (*(vp)->v_op->vop_access)(vp, mode, flags, cr, ct);
3786         VOPSTATS_UPDATE(vp, access);
3787         return (err);
3788 }
3789 
3790 int
3791 fop_lookup(
3792         vnode_t *dvp,
3793         char *nm,
3794         vnode_t **vpp,
3795         pathname_t *pnp,
3796         int flags,
3797         vnode_t *rdir,
3798         cred_t *cr,
3799         caller_context_t *ct,
3800         int *deflags,           /* Returned per-dirent flags */
3801         pathname_t *ppnp)       /* Returned case-preserved name in directory */
3802 {
3803         int ret;
3804 
3805         /*
3806          * If this file system doesn't support case-insensitive access
3807          * and said access is requested, fail quickly.  It is required
3808          * that if the vfs supports case-insensitive lookup, it also
3809          * supports extended dirent flags.
3810          */
3811         if (flags & FIGNORECASE &&
3812             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3813             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3814                 return (EINVAL);
3815 
3816         VOPXID_MAP_CR(dvp, cr);
3817 
3818         if ((flags & LOOKUP_XATTR) && (flags & LOOKUP_HAVE_SYSATTR_DIR) == 0) {
3819                 ret = xattr_dir_lookup(dvp, vpp, flags, cr);
3820         } else {
3821                 ret = (*(dvp)->v_op->vop_lookup)
3822                     (dvp, nm, vpp, pnp, flags, rdir, cr, ct, deflags, ppnp);
3823         }
3824         if (ret == 0 && *vpp) {
3825                 VOPSTATS_UPDATE(*vpp, lookup);
3826                 vn_updatepath(dvp, *vpp, nm);
3827         }
3828 
3829         return (ret);
3830 }
3831 
3832 int
3833 fop_create(
3834         vnode_t *dvp,
3835         char *name,
3836         vattr_t *vap,
3837         vcexcl_t excl,
3838         int mode,
3839         vnode_t **vpp,
3840         cred_t *cr,
3841         int flags,
3842         caller_context_t *ct,
3843         vsecattr_t *vsecp)      /* ACL to set during create */
3844 {
3845         int ret;
3846 
3847         if (vsecp != NULL &&
3848             vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3849                 return (EINVAL);
3850         }
3851         /*
3852          * If this file system doesn't support case-insensitive access
3853          * and said access is requested, fail quickly.
3854          */
3855         if (flags & FIGNORECASE &&
3856             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3857             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3858                 return (EINVAL);
3859 
3860         VOPXID_MAP_CR(dvp, cr);
3861 
3862         ret = (*(dvp)->v_op->vop_create)
3863             (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
3864         if (ret == 0 && *vpp) {
3865                 VOPSTATS_UPDATE(*vpp, create);
3866                 vn_updatepath(dvp, *vpp, name);
3867         }
3868 
3869         return (ret);
3870 }
3871 
3872 int
3873 fop_remove(
3874         vnode_t *dvp,
3875         char *nm,
3876         cred_t *cr,
3877         caller_context_t *ct,
3878         int flags)
3879 {
3880         int     err;
3881 
3882         /*
3883          * If this file system doesn't support case-insensitive access
3884          * and said access is requested, fail quickly.
3885          */
3886         if (flags & FIGNORECASE &&
3887             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3888             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3889                 return (EINVAL);
3890 
3891         VOPXID_MAP_CR(dvp, cr);
3892 
3893         err = (*(dvp)->v_op->vop_remove)(dvp, nm, cr, ct, flags);
3894         VOPSTATS_UPDATE(dvp, remove);
3895         return (err);
3896 }
3897 
3898 int
3899 fop_link(
3900         vnode_t *tdvp,
3901         vnode_t *svp,
3902         char *tnm,
3903         cred_t *cr,
3904         caller_context_t *ct,
3905         int flags)
3906 {
3907         int     err;
3908 
3909         /*
3910          * If the target file system doesn't support case-insensitive access
3911          * and said access is requested, fail quickly.
3912          */
3913         if (flags & FIGNORECASE &&
3914             (vfs_has_feature(tdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3915             vfs_has_feature(tdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3916                 return (EINVAL);
3917 
3918         VOPXID_MAP_CR(tdvp, cr);
3919 
3920         err = (*(tdvp)->v_op->vop_link)(tdvp, svp, tnm, cr, ct, flags);
3921         VOPSTATS_UPDATE(tdvp, link);
3922         return (err);
3923 }
3924 
3925 int
3926 fop_rename(
3927         vnode_t *sdvp,
3928         char *snm,
3929         vnode_t *tdvp,
3930         char *tnm,
3931         cred_t *cr,
3932         caller_context_t *ct,
3933         int flags)
3934 {
3935         int     err;
3936 
3937         /*
3938          * If the file system involved does not support
3939          * case-insensitive access and said access is requested, fail
3940          * quickly.
3941          */
3942         if (flags & FIGNORECASE &&
3943             ((vfs_has_feature(sdvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3944             vfs_has_feature(sdvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0)))
3945                 return (EINVAL);
3946 
3947         VOPXID_MAP_CR(tdvp, cr);
3948 
3949         err = (*(sdvp)->v_op->vop_rename)(sdvp, snm, tdvp, tnm, cr, ct, flags);
3950         VOPSTATS_UPDATE(sdvp, rename);
3951         return (err);
3952 }
3953 
3954 int
3955 fop_mkdir(
3956         vnode_t *dvp,
3957         char *dirname,
3958         vattr_t *vap,
3959         vnode_t **vpp,
3960         cred_t *cr,
3961         caller_context_t *ct,
3962         int flags,
3963         vsecattr_t *vsecp)      /* ACL to set during create */
3964 {
3965         int ret;
3966 
3967         if (vsecp != NULL &&
3968             vfs_has_feature(dvp->v_vfsp, VFSFT_ACLONCREATE) == 0) {
3969                 return (EINVAL);
3970         }
3971         /*
3972          * If this file system doesn't support case-insensitive access
3973          * and said access is requested, fail quickly.
3974          */
3975         if (flags & FIGNORECASE &&
3976             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
3977             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
3978                 return (EINVAL);
3979 
3980         VOPXID_MAP_CR(dvp, cr);
3981 
3982         ret = (*(dvp)->v_op->vop_mkdir)
3983             (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
3984         if (ret == 0 && *vpp) {
3985                 VOPSTATS_UPDATE(*vpp, mkdir);
3986                 vn_updatepath(dvp, *vpp, dirname);
3987         }
3988 
3989         return (ret);
3990 }
3991 
3992 int
3993 fop_rmdir(
3994         vnode_t *dvp,
3995         char *nm,
3996         vnode_t *cdir,
3997         cred_t *cr,
3998         caller_context_t *ct,
3999         int flags)
4000 {
4001         int     err;
4002 
4003         /*
4004          * If this file system doesn't support case-insensitive access
4005          * and said access is requested, fail quickly.
4006          */
4007         if (flags & FIGNORECASE &&
4008             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
4009             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
4010                 return (EINVAL);
4011 
4012         VOPXID_MAP_CR(dvp, cr);
4013 
4014         err = (*(dvp)->v_op->vop_rmdir)(dvp, nm, cdir, cr, ct, flags);
4015         VOPSTATS_UPDATE(dvp, rmdir);
4016         return (err);
4017 }
4018 
4019 int
4020 fop_readdir(
4021         vnode_t *vp,
4022         uio_t *uiop,
4023         cred_t *cr,
4024         int *eofp,
4025         caller_context_t *ct,
4026         int flags)
4027 {
4028         int     err;
4029         ssize_t resid_start = uiop->uio_resid;
4030 
4031         /*
4032          * If this file system doesn't support retrieving directory
4033          * entry flags and said access is requested, fail quickly.
4034          */
4035         if (flags & V_RDDIR_ENTFLAGS &&
4036             vfs_has_feature(vp->v_vfsp, VFSFT_DIRENTFLAGS) == 0)
4037                 return (EINVAL);
4038 
4039         VOPXID_MAP_CR(vp, cr);
4040 
4041         err = (*(vp)->v_op->vop_readdir)(vp, uiop, cr, eofp, ct, flags);
4042         VOPSTATS_UPDATE_IO(vp, readdir,
4043             readdir_bytes, (resid_start - uiop->uio_resid));
4044         return (err);
4045 }
4046 
4047 int
4048 fop_symlink(
4049         vnode_t *dvp,
4050         char *linkname,
4051         vattr_t *vap,
4052         char *target,
4053         cred_t *cr,
4054         caller_context_t *ct,
4055         int flags)
4056 {
4057         int     err;
4058         xvattr_t xvattr;
4059 
4060         /*
4061          * If this file system doesn't support case-insensitive access
4062          * and said access is requested, fail quickly.
4063          */
4064         if (flags & FIGNORECASE &&
4065             (vfs_has_feature(dvp->v_vfsp, VFSFT_CASEINSENSITIVE) == 0 &&
4066             vfs_has_feature(dvp->v_vfsp, VFSFT_NOCASESENSITIVE) == 0))
4067                 return (EINVAL);
4068 
4069         VOPXID_MAP_CR(dvp, cr);
4070 
4071         /* check for reparse point */
4072         if ((vfs_has_feature(dvp->v_vfsp, VFSFT_REPARSE)) &&
4073             (strncmp(target, FS_REPARSE_TAG_STR,
4074             strlen(FS_REPARSE_TAG_STR)) == 0)) {
4075                 if (!fs_reparse_mark(target, vap, &xvattr))
4076                         vap = (vattr_t *)&xvattr;
4077         }
4078 
4079         err = (*(dvp)->v_op->vop_symlink)
4080             (dvp, linkname, vap, target, cr, ct, flags);
4081         VOPSTATS_UPDATE(dvp, symlink);
4082         return (err);
4083 }
4084 
4085 int
4086 fop_readlink(
4087         vnode_t *vp,
4088         uio_t *uiop,
4089         cred_t *cr,
4090         caller_context_t *ct)
4091 {
4092         int     err;
4093 
4094         VOPXID_MAP_CR(vp, cr);
4095 
4096         err = (*(vp)->v_op->vop_readlink)(vp, uiop, cr, ct);
4097         VOPSTATS_UPDATE(vp, readlink);
4098         return (err);
4099 }
4100 
4101 int
4102 fop_fsync(
4103         vnode_t *vp,
4104         int syncflag,
4105         cred_t *cr,
4106         caller_context_t *ct)
4107 {
4108         int     err;
4109 
4110         VOPXID_MAP_CR(vp, cr);
4111 
4112         err = (*(vp)->v_op->vop_fsync)(vp, syncflag, cr, ct);
4113         VOPSTATS_UPDATE(vp, fsync);
4114         return (err);
4115 }
4116 
4117 void
4118 fop_inactive(
4119         vnode_t *vp,
4120         cred_t *cr,
4121         caller_context_t *ct)
4122 {
4123         /* Need to update stats before vop call since we may lose the vnode */
4124         VOPSTATS_UPDATE(vp, inactive);
4125 
4126         VOPXID_MAP_CR(vp, cr);
4127 
4128         (*(vp)->v_op->vop_inactive)(vp, cr, ct);
4129 }
4130 
4131 int
4132 fop_fid(
4133         vnode_t *vp,
4134         fid_t *fidp,
4135         caller_context_t *ct)
4136 {
4137         int     err;
4138 
4139         err = (*(vp)->v_op->vop_fid)(vp, fidp, ct);
4140         VOPSTATS_UPDATE(vp, fid);
4141         return (err);
4142 }
4143 
4144 int
4145 fop_rwlock(
4146         vnode_t *vp,
4147         int write_lock,
4148         caller_context_t *ct)
4149 {
4150         int     ret;
4151 
4152         ret = ((*(vp)->v_op->vop_rwlock)(vp, write_lock, ct));
4153         VOPSTATS_UPDATE(vp, rwlock);
4154         return (ret);
4155 }
4156 
4157 void
4158 fop_rwunlock(
4159         vnode_t *vp,
4160         int write_lock,
4161         caller_context_t *ct)
4162 {
4163         (*(vp)->v_op->vop_rwunlock)(vp, write_lock, ct);
4164         VOPSTATS_UPDATE(vp, rwunlock);
4165 }
4166 
4167 int
4168 fop_seek(
4169         vnode_t *vp,
4170         offset_t ooff,
4171         offset_t *noffp,
4172         caller_context_t *ct)
4173 {
4174         int     err;
4175 
4176         err = (*(vp)->v_op->vop_seek)(vp, ooff, noffp, ct);
4177         VOPSTATS_UPDATE(vp, seek);
4178         return (err);
4179 }
4180 
4181 int
4182 fop_cmp(
4183         vnode_t *vp1,
4184         vnode_t *vp2,
4185         caller_context_t *ct)
4186 {
4187         int     err;
4188 
4189         err = (*(vp1)->v_op->vop_cmp)(vp1, vp2, ct);
4190         VOPSTATS_UPDATE(vp1, cmp);
4191         return (err);
4192 }
4193 
4194 int
4195 fop_frlock(
4196         vnode_t *vp,
4197         int cmd,
4198         flock64_t *bfp,
4199         int flag,
4200         offset_t offset,
4201         struct flk_callback *flk_cbp,
4202         cred_t *cr,
4203         caller_context_t *ct)
4204 {
4205         int     err;
4206 
4207         VOPXID_MAP_CR(vp, cr);
4208 
4209         err = (*(vp)->v_op->vop_frlock)
4210             (vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
4211         VOPSTATS_UPDATE(vp, frlock);
4212         return (err);
4213 }
4214 
4215 int
4216 fop_space(
4217         vnode_t *vp,
4218         int cmd,
4219         flock64_t *bfp,
4220         int flag,
4221         offset_t offset,
4222         cred_t *cr,
4223         caller_context_t *ct)
4224 {
4225         int     err;
4226 
4227         VOPXID_MAP_CR(vp, cr);
4228 
4229         err = (*(vp)->v_op->vop_space)(vp, cmd, bfp, flag, offset, cr, ct);
4230         VOPSTATS_UPDATE(vp, space);
4231         return (err);
4232 }
4233 
4234 int
4235 fop_realvp(
4236         vnode_t *vp,
4237         vnode_t **vpp,
4238         caller_context_t *ct)
4239 {
4240         int     err;
4241 
4242         err = (*(vp)->v_op->vop_realvp)(vp, vpp, ct);
4243         VOPSTATS_UPDATE(vp, realvp);
4244         return (err);
4245 }
4246 
4247 int
4248 fop_getpage(
4249         vnode_t *vp,
4250         offset_t off,
4251         size_t len,
4252         uint_t *protp,
4253         page_t **plarr,
4254         size_t plsz,
4255         struct seg *seg,
4256         caddr_t addr,
4257         enum seg_rw rw,
4258         cred_t *cr,
4259         caller_context_t *ct)
4260 {
4261         int     err;
4262 
4263         VOPXID_MAP_CR(vp, cr);
4264 
4265         err = (*(vp)->v_op->vop_getpage)
4266             (vp, off, len, protp, plarr, plsz, seg, addr, rw, cr, ct);
4267         VOPSTATS_UPDATE(vp, getpage);
4268         return (err);
4269 }
4270 
4271 int
4272 fop_putpage(
4273         vnode_t *vp,
4274         offset_t off,
4275         size_t len,
4276         int flags,
4277         cred_t *cr,
4278         caller_context_t *ct)
4279 {
4280         int     err;
4281 
4282         VOPXID_MAP_CR(vp, cr);
4283 
4284         err = (*(vp)->v_op->vop_putpage)(vp, off, len, flags, cr, ct);
4285         VOPSTATS_UPDATE(vp, putpage);
4286         return (err);
4287 }
4288 
4289 int
4290 fop_map(
4291         vnode_t *vp,
4292         offset_t off,
4293         struct as *as,
4294         caddr_t *addrp,
4295         size_t len,
4296         uchar_t prot,
4297         uchar_t maxprot,
4298         uint_t flags,
4299         cred_t *cr,
4300         caller_context_t *ct)
4301 {
4302         int     err;
4303 
4304         VOPXID_MAP_CR(vp, cr);
4305 
4306         err = (*(vp)->v_op->vop_map)
4307             (vp, off, as, addrp, len, prot, maxprot, flags, cr, ct);
4308         VOPSTATS_UPDATE(vp, map);
4309         return (err);
4310 }
4311 
4312 int
4313 fop_addmap(
4314         vnode_t *vp,
4315         offset_t off,
4316         struct as *as,
4317         caddr_t addr,
4318         size_t len,
4319         uchar_t prot,
4320         uchar_t maxprot,
4321         uint_t flags,
4322         cred_t *cr,
4323         caller_context_t *ct)
4324 {
4325         int error;
4326         u_longlong_t delta;
4327 
4328         VOPXID_MAP_CR(vp, cr);
4329 
4330         error = (*(vp)->v_op->vop_addmap)
4331             (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
4332 
4333         if ((!error) && (vp->v_type == VREG)) {
4334                 delta = (u_longlong_t)btopr(len);
4335                 /*
4336                  * If file is declared MAP_PRIVATE, it can't be written back
4337                  * even if open for write. Handle as read.
4338                  */
4339                 if (flags & MAP_PRIVATE) {
4340                         atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4341                             (int64_t)delta);
4342                 } else {
4343                         /*
4344                          * atomic_add_64 forces the fetch of a 64 bit value to
4345                          * be atomic on 32 bit machines
4346                          */
4347                         if (maxprot & PROT_WRITE)
4348                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
4349                                     (int64_t)delta);
4350                         if (maxprot & PROT_READ)
4351                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4352                                     (int64_t)delta);
4353                         if (maxprot & PROT_EXEC)
4354                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4355                                     (int64_t)delta);
4356                 }
4357         }
4358         VOPSTATS_UPDATE(vp, addmap);
4359         return (error);
4360 }
4361 
4362 int
4363 fop_delmap(
4364         vnode_t *vp,
4365         offset_t off,
4366         struct as *as,
4367         caddr_t addr,
4368         size_t len,
4369         uint_t prot,
4370         uint_t maxprot,
4371         uint_t flags,
4372         cred_t *cr,
4373         caller_context_t *ct)
4374 {
4375         int error;
4376         u_longlong_t delta;
4377 
4378         VOPXID_MAP_CR(vp, cr);
4379 
4380         error = (*(vp)->v_op->vop_delmap)
4381             (vp, off, as, addr, len, prot, maxprot, flags, cr, ct);
4382 
4383         /*
4384          * NFS calls into delmap twice, the first time
4385          * it simply establishes a callback mechanism and returns EAGAIN
4386          * while the real work is being done upon the second invocation.
4387          * We have to detect this here and only decrement the counts upon
4388          * the second delmap request.
4389          */
4390         if ((error != EAGAIN) && (vp->v_type == VREG)) {
4391 
4392                 delta = (u_longlong_t)btopr(len);
4393 
4394                 if (flags & MAP_PRIVATE) {
4395                         atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4396                             (int64_t)(-delta));
4397                 } else {
4398                         /*
4399                          * atomic_add_64 forces the fetch of a 64 bit value
4400                          * to be atomic on 32 bit machines
4401                          */
4402                         if (maxprot & PROT_WRITE)
4403                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_write)),
4404                                     (int64_t)(-delta));
4405                         if (maxprot & PROT_READ)
4406                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4407                                     (int64_t)(-delta));
4408                         if (maxprot & PROT_EXEC)
4409                                 atomic_add_64((uint64_t *)(&(vp->v_mmap_read)),
4410                                     (int64_t)(-delta));
4411                 }
4412         }
4413         VOPSTATS_UPDATE(vp, delmap);
4414         return (error);
4415 }
4416 
4417 
4418 int
4419 fop_poll(
4420         vnode_t *vp,
4421         short events,
4422         int anyyet,
4423         short *reventsp,
4424         struct pollhead **phpp,
4425         caller_context_t *ct)
4426 {
4427         int     err;
4428 
4429         err = (*(vp)->v_op->vop_poll)(vp, events, anyyet, reventsp, phpp, ct);
4430         VOPSTATS_UPDATE(vp, poll);
4431         return (err);
4432 }
4433 
4434 int
4435 fop_dump(
4436         vnode_t *vp,
4437         caddr_t addr,
4438         offset_t lbdn,
4439         offset_t dblks,
4440         caller_context_t *ct)
4441 {
4442         int     err;
4443 
4444         /* ensure lbdn and dblks can be passed safely to bdev_dump */
4445         if ((lbdn != (daddr_t)lbdn) || (dblks != (int)dblks))
4446                 return (EIO);
4447 
4448         err = (*(vp)->v_op->vop_dump)(vp, addr, lbdn, dblks, ct);
4449         VOPSTATS_UPDATE(vp, dump);
4450         return (err);
4451 }
4452 
4453 int
4454 fop_pathconf(
4455         vnode_t *vp,
4456         int cmd,
4457         ulong_t *valp,
4458         cred_t *cr,
4459         caller_context_t *ct)
4460 {
4461         int     err;
4462 
4463         VOPXID_MAP_CR(vp, cr);
4464 
4465         err = (*(vp)->v_op->vop_pathconf)(vp, cmd, valp, cr, ct);
4466         VOPSTATS_UPDATE(vp, pathconf);
4467         return (err);
4468 }
4469 
4470 int
4471 fop_pageio(
4472         vnode_t *vp,
4473         struct page *pp,
4474         u_offset_t io_off,
4475         size_t io_len,
4476         int flags,
4477         cred_t *cr,
4478         caller_context_t *ct)
4479 {
4480         int     err;
4481 
4482         VOPXID_MAP_CR(vp, cr);
4483 
4484         err = (*(vp)->v_op->vop_pageio)(vp, pp, io_off, io_len, flags, cr, ct);
4485         VOPSTATS_UPDATE(vp, pageio);
4486         return (err);
4487 }
4488 
4489 int
4490 fop_dumpctl(
4491         vnode_t *vp,
4492         int action,
4493         offset_t *blkp,
4494         caller_context_t *ct)
4495 {
4496         int     err;
4497         err = (*(vp)->v_op->vop_dumpctl)(vp, action, blkp, ct);
4498         VOPSTATS_UPDATE(vp, dumpctl);
4499         return (err);
4500 }
4501 
4502 void
4503 fop_dispose(
4504         vnode_t *vp,
4505         page_t *pp,
4506         int flag,
4507         int dn,
4508         cred_t *cr,
4509         caller_context_t *ct)
4510 {
4511         /* Must do stats first since it's possible to lose the vnode */
4512         VOPSTATS_UPDATE(vp, dispose);
4513 
4514         VOPXID_MAP_CR(vp, cr);
4515 
4516         (*(vp)->v_op->vop_dispose)(vp, pp, flag, dn, cr, ct);
4517 }
4518 
4519 int
4520 fop_setsecattr(
4521         vnode_t *vp,
4522         vsecattr_t *vsap,
4523         int flag,
4524         cred_t *cr,
4525         caller_context_t *ct)
4526 {
4527         int     err;
4528 
4529         VOPXID_MAP_CR(vp, cr);
4530 
4531         /*
4532          * We're only allowed to skip the ACL check iff we used a 32 bit
4533          * ACE mask with VOP_ACCESS() to determine permissions.
4534          */
4535         if ((flag & ATTR_NOACLCHECK) &&
4536             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4537                 return (EINVAL);
4538         }
4539         err = (*(vp)->v_op->vop_setsecattr) (vp, vsap, flag, cr, ct);
4540         VOPSTATS_UPDATE(vp, setsecattr);
4541         return (err);
4542 }
4543 
4544 int
4545 fop_getsecattr(
4546         vnode_t *vp,
4547         vsecattr_t *vsap,
4548         int flag,
4549         cred_t *cr,
4550         caller_context_t *ct)
4551 {
4552         int     err;
4553 
4554         /*
4555          * We're only allowed to skip the ACL check iff we used a 32 bit
4556          * ACE mask with VOP_ACCESS() to determine permissions.
4557          */
4558         if ((flag & ATTR_NOACLCHECK) &&
4559             vfs_has_feature(vp->v_vfsp, VFSFT_ACEMASKONACCESS) == 0) {
4560                 return (EINVAL);
4561         }
4562 
4563         VOPXID_MAP_CR(vp, cr);
4564 
4565         err = (*(vp)->v_op->vop_getsecattr) (vp, vsap, flag, cr, ct);
4566         VOPSTATS_UPDATE(vp, getsecattr);
4567         return (err);
4568 }
4569 
4570 int
4571 fop_shrlock(
4572         vnode_t *vp,
4573         int cmd,
4574         struct shrlock *shr,
4575         int flag,
4576         cred_t *cr,
4577         caller_context_t *ct)
4578 {
4579         int     err;
4580 
4581         VOPXID_MAP_CR(vp, cr);
4582 
4583         err = (*(vp)->v_op->vop_shrlock)(vp, cmd, shr, flag, cr, ct);
4584         VOPSTATS_UPDATE(vp, shrlock);
4585         return (err);
4586 }
4587 
4588 int
4589 fop_vnevent(vnode_t *vp, vnevent_t vnevent, vnode_t *dvp, char *fnm,
4590     caller_context_t *ct)
4591 {
4592         int     err;
4593 
4594         err = (*(vp)->v_op->vop_vnevent)(vp, vnevent, dvp, fnm, ct);
4595         VOPSTATS_UPDATE(vp, vnevent);
4596         return (err);
4597 }
4598 
4599 int
4600 fop_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *uiop, cred_t *cr,
4601     caller_context_t *ct)
4602 {
4603         int err;
4604 
4605         if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4606                 return (ENOTSUP);
4607         err = (*(vp)->v_op->vop_reqzcbuf)(vp, ioflag, uiop, cr, ct);
4608         VOPSTATS_UPDATE(vp, reqzcbuf);
4609         return (err);
4610 }
4611 
4612 int
4613 fop_retzcbuf(vnode_t *vp, xuio_t *uiop, cred_t *cr, caller_context_t *ct)
4614 {
4615         int err;
4616 
4617         if (vfs_has_feature(vp->v_vfsp, VFSFT_ZEROCOPY_SUPPORTED) == 0)
4618                 return (ENOTSUP);
4619         err = (*(vp)->v_op->vop_retzcbuf)(vp, uiop, cr, ct);
4620         VOPSTATS_UPDATE(vp, retzcbuf);
4621         return (err);
4622 }
4623 
4624 /*
4625  * Default destructor
4626  *      Needed because NULL destructor means that the key is unused
4627  */
4628 /* ARGSUSED */
4629 void
4630 vsd_defaultdestructor(void *value)
4631 {}
4632 
4633 /*
4634  * Create a key (index into per vnode array)
4635  *      Locks out vsd_create, vsd_destroy, and vsd_free
4636  *      May allocate memory with lock held
4637  */
4638 void
4639 vsd_create(uint_t *keyp, void (*destructor)(void *))
4640 {
4641         int     i;
4642         uint_t  nkeys;
4643 
4644         /*
4645          * if key is allocated, do nothing
4646          */
4647         mutex_enter(&vsd_lock);
4648         if (*keyp) {
4649                 mutex_exit(&vsd_lock);
4650                 return;
4651         }
4652         /*
4653          * find an unused key
4654          */
4655         if (destructor == NULL)
4656                 destructor = vsd_defaultdestructor;
4657 
4658         for (i = 0; i < vsd_nkeys; ++i)
4659                 if (vsd_destructor[i] == NULL)
4660                         break;
4661 
4662         /*
4663          * if no unused keys, increase the size of the destructor array
4664          */
4665         if (i == vsd_nkeys) {
4666                 if ((nkeys = (vsd_nkeys << 1)) == 0)
4667                         nkeys = 1;
4668                 vsd_destructor =
4669                     (void (**)(void *))vsd_realloc((void *)vsd_destructor,
4670                     (size_t)(vsd_nkeys * sizeof (void (*)(void *))),
4671                     (size_t)(nkeys * sizeof (void (*)(void *))));
4672                 vsd_nkeys = nkeys;
4673         }
4674 
4675         /*
4676          * allocate the next available unused key
4677          */
4678         vsd_destructor[i] = destructor;
4679         *keyp = i + 1;
4680 
4681         /* create vsd_list, if it doesn't exist */
4682         if (vsd_list == NULL) {
4683                 vsd_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
4684                 list_create(vsd_list, sizeof (struct vsd_node),
4685                     offsetof(struct vsd_node, vs_nodes));
4686         }
4687 
4688         mutex_exit(&vsd_lock);
4689 }
4690 
4691 /*
4692  * Destroy a key
4693  *
4694  * Assumes that the caller is preventing vsd_set and vsd_get
4695  * Locks out vsd_create, vsd_destroy, and vsd_free
4696  * May free memory with lock held
4697  */
4698 void
4699 vsd_destroy(uint_t *keyp)
4700 {
4701         uint_t key;
4702         struct vsd_node *vsd;
4703 
4704         /*
4705          * protect the key namespace and our destructor lists
4706          */
4707         mutex_enter(&vsd_lock);
4708         key = *keyp;
4709         *keyp = 0;
4710 
4711         ASSERT(key <= vsd_nkeys);
4712 
4713         /*
4714          * if the key is valid
4715          */
4716         if (key != 0) {
4717                 uint_t k = key - 1;
4718                 /*
4719                  * for every vnode with VSD, call key's destructor
4720                  */
4721                 for (vsd = list_head(vsd_list); vsd != NULL;
4722                     vsd = list_next(vsd_list, vsd)) {
4723                         /*
4724                          * no VSD for key in this vnode
4725                          */
4726                         if (key > vsd->vs_nkeys)
4727                                 continue;
4728                         /*
4729                          * call destructor for key
4730                          */
4731                         if (vsd->vs_value[k] && vsd_destructor[k])
4732                                 (*vsd_destructor[k])(vsd->vs_value[k]);
4733                         /*
4734                          * reset value for key
4735                          */
4736                         vsd->vs_value[k] = NULL;
4737                 }
4738                 /*
4739                  * actually free the key (NULL destructor == unused)
4740                  */
4741                 vsd_destructor[k] = NULL;
4742         }
4743 
4744         mutex_exit(&vsd_lock);
4745 }
4746 
4747 /*
4748  * Quickly return the per vnode value that was stored with the specified key
4749  * Assumes the caller is protecting key from vsd_create and vsd_destroy
4750  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4751  */
4752 void *
4753 vsd_get(vnode_t *vp, uint_t key)
4754 {
4755         struct vsd_node *vsd;
4756 
4757         ASSERT(vp != NULL);
4758         ASSERT(mutex_owned(&vp->v_vsd_lock));
4759 
4760         vsd = vp->v_vsd;
4761 
4762         if (key && vsd != NULL && key <= vsd->vs_nkeys)
4763                 return (vsd->vs_value[key - 1]);
4764         return (NULL);
4765 }
4766 
4767 /*
4768  * Set a per vnode value indexed with the specified key
4769  * Assumes the caller is holding v_vsd_lock to protect the vsd.
4770  */
4771 int
4772 vsd_set(vnode_t *vp, uint_t key, void *value)
4773 {
4774         struct vsd_node *vsd;
4775 
4776         ASSERT(vp != NULL);
4777         ASSERT(mutex_owned(&vp->v_vsd_lock));
4778 
4779         if (key == 0)
4780                 return (EINVAL);
4781 
4782         vsd = vp->v_vsd;
4783         if (vsd == NULL)
4784                 vsd = vp->v_vsd = kmem_zalloc(sizeof (*vsd), KM_SLEEP);
4785 
4786         /*
4787          * If the vsd was just allocated, vs_nkeys will be 0, so the following
4788          * code won't happen and we will continue down and allocate space for
4789          * the vs_value array.
4790          * If the caller is replacing one value with another, then it is up
4791          * to the caller to free/rele/destroy the previous value (if needed).
4792          */
4793         if (key <= vsd->vs_nkeys) {
4794                 vsd->vs_value[key - 1] = value;
4795                 return (0);
4796         }
4797 
4798         ASSERT(key <= vsd_nkeys);
4799 
4800         if (vsd->vs_nkeys == 0) {
4801                 mutex_enter(&vsd_lock);     /* lock out vsd_destroy() */
4802                 /*
4803                  * Link onto list of all VSD nodes.
4804                  */
4805                 list_insert_head(vsd_list, vsd);
4806                 mutex_exit(&vsd_lock);
4807         }
4808 
4809         /*
4810          * Allocate vnode local storage and set the value for key
4811          */
4812         vsd->vs_value = vsd_realloc(vsd->vs_value,
4813             vsd->vs_nkeys * sizeof (void *),
4814             key * sizeof (void *));
4815         vsd->vs_nkeys = key;
4816         vsd->vs_value[key - 1] = value;
4817 
4818         return (0);
4819 }
4820 
4821 /*
4822  * Called from vn_free() to run the destructor function for each vsd
4823  *      Locks out vsd_create and vsd_destroy
4824  *      Assumes that the destructor *DOES NOT* use vsd
4825  */
4826 void
4827 vsd_free(vnode_t *vp)
4828 {
4829         int i;
4830         struct vsd_node *vsd = vp->v_vsd;
4831 
4832         if (vsd == NULL)
4833                 return;
4834 
4835         if (vsd->vs_nkeys == 0) {
4836                 kmem_free(vsd, sizeof (*vsd));
4837                 vp->v_vsd = NULL;
4838                 return;
4839         }
4840 
4841         /*
4842          * lock out vsd_create and vsd_destroy, call
4843          * the destructor, and mark the value as destroyed.
4844          */
4845         mutex_enter(&vsd_lock);
4846 
4847         for (i = 0; i < vsd->vs_nkeys; i++) {
4848                 if (vsd->vs_value[i] && vsd_destructor[i])
4849                         (*vsd_destructor[i])(vsd->vs_value[i]);
4850                 vsd->vs_value[i] = NULL;
4851         }
4852 
4853         /*
4854          * remove from linked list of VSD nodes
4855          */
4856         list_remove(vsd_list, vsd);
4857 
4858         mutex_exit(&vsd_lock);
4859 
4860         /*
4861          * free up the VSD
4862          */
4863         kmem_free(vsd->vs_value, vsd->vs_nkeys * sizeof (void *));
4864         kmem_free(vsd, sizeof (struct vsd_node));
4865         vp->v_vsd = NULL;
4866 }
4867 
4868 /*
4869  * realloc
4870  */
4871 static void *
4872 vsd_realloc(void *old, size_t osize, size_t nsize)
4873 {
4874         void *new;
4875 
4876         new = kmem_zalloc(nsize, KM_SLEEP);
4877         if (old) {
4878                 bcopy(old, new, osize);
4879                 kmem_free(old, osize);
4880         }
4881         return (new);
4882 }
4883 
4884 /*
4885  * Setup the extensible system attribute for creating a reparse point.
4886  * The symlink data 'target' is validated for proper format of a reparse
4887  * string and a check also made to make sure the symlink data does not
4888  * point to an existing file.
4889  *
4890  * return 0 if ok else -1.
4891  */
4892 static int
4893 fs_reparse_mark(char *target, vattr_t *vap, xvattr_t *xvattr)
4894 {
4895         xoptattr_t *xoap;
4896 
4897         if ((!target) || (!vap) || (!xvattr))
4898                 return (-1);
4899 
4900         /* validate reparse string */
4901         if (reparse_validate((const char *)target))
4902                 return (-1);
4903 
4904         xva_init(xvattr);
4905         xvattr->xva_vattr = *vap;
4906         xvattr->xva_vattr.va_mask |= AT_XVATTR;
4907         xoap = xva_getxoptattr(xvattr);
4908         ASSERT(xoap);
4909         XVA_SET_REQ(xvattr, XAT_REPARSE);
4910         xoap->xoa_reparse = 1;
4911 
4912         return (0);
4913 }
4914 
4915 /*
4916  * Function to check whether a symlink is a reparse point.
4917  * Return B_TRUE if it is a reparse point, else return B_FALSE
4918  */
4919 boolean_t
4920 vn_is_reparse(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4921 {
4922         xvattr_t xvattr;
4923         xoptattr_t *xoap;
4924 
4925         if ((vp->v_type != VLNK) ||
4926             !(vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR)))
4927                 return (B_FALSE);
4928 
4929         xva_init(&xvattr);
4930         xoap = xva_getxoptattr(&xvattr);
4931         ASSERT(xoap);
4932         XVA_SET_REQ(&xvattr, XAT_REPARSE);
4933 
4934         if (VOP_GETATTR(vp, &xvattr.xva_vattr, 0, cr, ct))
4935                 return (B_FALSE);
4936 
4937         if ((!(xvattr.xva_vattr.va_mask & AT_XVATTR)) ||
4938             (!(XVA_ISSET_RTN(&xvattr, XAT_REPARSE))))
4939                 return (B_FALSE);
4940 
4941         return (xoap->xoa_reparse ? B_TRUE : B_FALSE);
4942 }