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.
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.
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]
23 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
28 /* All Rights Reserved */
31 * University Copyright- Copyright (c) 1982, 1986, 1988
32 * The Regents of the University of California
35 * University Acknowledgment- Portions of this document are derived from
36 * software developed by the University of California, Berkeley, and its
40 #include <sys/types.h>
41 #include <sys/param.h>
42 #include <sys/t_lock.h>
43 #include <sys/errno.h>
48 #include <sys/pathname.h>
50 #include <sys/vfs_opreg.h>
51 #include <sys/vnode.h>
52 #include <sys/rwstlock.h>
57 #include <sys/sysmacros.h>
58 #include <sys/cmn_err.h>
59 #include <sys/systm.h>
61 #include <sys/debug.h>
64 #include <sys/nbmlock.h>
65 #include <sys/fcntl.h>
66 #include <fs/fs_subr.h>
67 #include <sys/taskq.h>
68 #include <fs/fs_reparse.h>
70 /* Determine if this vnode is a file that is read-only */
71 #define ISROFILE(vp) \
72 ((vp)->v_type != VCHR && (vp)->v_type != VBLK && \
73 (vp)->v_type != VFIFO && vn_is_readonly(vp))
75 /* Tunable via /etc/system; used only by admin/install */
76 int nfs_global_client_only
;
79 * Array of vopstats_t for per-FS-type vopstats. This array has the same
80 * number of entries as and parallel to the vfssw table. (Arguably, it could
81 * be part of the vfssw table.) Once it's initialized, it's accessed using
82 * the same fstype index that is used to index into the vfssw table.
84 vopstats_t
**vopstats_fstype
;
86 /* vopstats initialization template used for fast initialization via bcopy() */
87 static vopstats_t
*vs_templatep
;
89 /* Kmem cache handle for vsk_anchor_t allocations */
90 kmem_cache_t
*vsk_anchor_cache
;
92 /* file events cleanup routine */
93 extern void free_fopdata(vnode_t
*);
96 * Root of AVL tree for the kstats associated with vopstats. Lock protects
97 * updates to vsktat_tree.
99 avl_tree_t vskstat_tree
;
100 kmutex_t vskstat_tree_lock
;
102 /* Global variable which enables/disables the vopstats collection */
103 int vopstats_enabled
= 1;
106 * forward declarations for internal vnode specific data (vsd)
108 static void *vsd_realloc(void *, size_t, size_t);
111 * forward declarations for reparse point functions
113 static int fs_reparse_mark(char *target
, vattr_t
*vap
, xvattr_t
*xvattr
);
116 * VSD -- VNODE SPECIFIC DATA
117 * The v_data pointer is typically used by a file system to store a
118 * pointer to the file system's private node (e.g. ufs inode, nfs rnode).
119 * However, there are times when additional project private data needs
120 * to be stored separately from the data (node) pointed to by v_data.
121 * This additional data could be stored by the file system itself or
122 * by a completely different kernel entity. VSD provides a way for
123 * callers to obtain a key and store a pointer to private data associated
126 * Callers are responsible for protecting the vsd by holding v_vsd_lock
127 * for calls to vsd_set() and vsd_get().
132 * vsd_nkeys - creation and deletion of vsd keys
133 * vsd_list - insertion and deletion of vsd_node in the vsd_list
134 * vsd_destructor - adding and removing destructors to the list
136 static kmutex_t vsd_lock
;
137 static uint_t vsd_nkeys
; /* size of destructor array */
138 /* list of vsd_node's */
139 static list_t
*vsd_list
= NULL
;
140 /* per-key destructor funcs */
141 static void (**vsd_destructor
)(void *);
144 * The following is the common set of actions needed to update the
145 * vopstats structure from a vnode op. Both VOPSTATS_UPDATE() and
146 * VOPSTATS_UPDATE_IO() do almost the same thing, except for the
147 * recording of the bytes transferred. Since the code is similar
148 * but small, it is nearly a duplicate. Consequently any changes
149 * to one may need to be reflected in the other.
150 * Rundown of the variables:
151 * vp - Pointer to the vnode
152 * counter - Partial name structure member to update in vopstats for counts
153 * bytecounter - Partial name structure member to update in vopstats for bytes
154 * bytesval - Value to update in vopstats for bytes
155 * fstype - Index into vsanchor_fstype[], same as index into vfssw[]
156 * vsp - Pointer to vopstats structure (either in vfs or vsanchor_fstype[i])
159 #define VOPSTATS_UPDATE(vp, counter) { \
160 vfs_t *vfsp = (vp)->v_vfsp; \
161 if (vfsp && vfsp->vfs_implp && \
162 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \
163 vopstats_t *vsp = &vfsp->vfs_vopstats; \
164 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \
165 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
166 size_t, uint64_t *); \
167 __dtrace_probe___fsinfo_##counter(vp, 0, stataddr); \
169 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \
170 vsp->n##counter.value.ui64++; \
175 #define VOPSTATS_UPDATE_IO(vp, counter, bytecounter, bytesval) { \
176 vfs_t *vfsp = (vp)->v_vfsp; \
177 if (vfsp && vfsp->vfs_implp && \
178 (vfsp->vfs_flag & VFS_STATS) && (vp)->v_type != VBAD) { \
179 vopstats_t *vsp = &vfsp->vfs_vopstats; \
180 uint64_t *stataddr = &(vsp->n##counter.value.ui64); \
181 extern void __dtrace_probe___fsinfo_##counter(vnode_t *, \
182 size_t, uint64_t *); \
183 __dtrace_probe___fsinfo_##counter(vp, bytesval, stataddr); \
185 vsp->bytecounter.value.ui64 += bytesval; \
186 if ((vsp = vfsp->vfs_fstypevsp) != NULL) { \
187 vsp->n##counter.value.ui64++; \
188 vsp->bytecounter.value.ui64 += bytesval; \
194 * If the filesystem does not support XIDs map credential
195 * If the vfsp is NULL, perhaps we should also map?
197 #define VOPXID_MAP_CR(vp, cr) { \
198 vfs_t *vfsp = (vp)->v_vfsp; \
199 if (vfsp != NULL && (vfsp->vfs_flag & VFS_XID) == 0) \
200 cr = crgetmapped(cr); \
204 * Convert stat(2) formats to vnode types and vice versa. (Knows about
205 * numerical order of S_IFMT and vnode types.)
207 enum vtype iftovt_tab
[] = {
208 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
209 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VNON
212 ushort_t vttoif_tab
[] = {
213 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
, S_IFIFO
,
214 S_IFDOOR
, 0, S_IFSOCK
, S_IFPORT
, 0
218 * The system vnode cache.
221 kmem_cache_t
*vn_cache
;
225 * Vnode operations vector.
228 static const fs_operation_trans_def_t vn_ops_table
[] = {
229 VOPNAME_OPEN
, offsetof(struct vnodeops
, vop_open
),
232 VOPNAME_CLOSE
, offsetof(struct vnodeops
, vop_close
),
235 VOPNAME_READ
, offsetof(struct vnodeops
, vop_read
),
238 VOPNAME_WRITE
, offsetof(struct vnodeops
, vop_write
),
241 VOPNAME_IOCTL
, offsetof(struct vnodeops
, vop_ioctl
),
244 VOPNAME_SETFL
, offsetof(struct vnodeops
, vop_setfl
),
247 VOPNAME_GETATTR
, offsetof(struct vnodeops
, vop_getattr
),
250 VOPNAME_SETATTR
, offsetof(struct vnodeops
, vop_setattr
),
253 VOPNAME_ACCESS
, offsetof(struct vnodeops
, vop_access
),
256 VOPNAME_LOOKUP
, offsetof(struct vnodeops
, vop_lookup
),
259 VOPNAME_CREATE
, offsetof(struct vnodeops
, vop_create
),
262 VOPNAME_REMOVE
, offsetof(struct vnodeops
, vop_remove
),
265 VOPNAME_LINK
, offsetof(struct vnodeops
, vop_link
),
268 VOPNAME_RENAME
, offsetof(struct vnodeops
, vop_rename
),
271 VOPNAME_MKDIR
, offsetof(struct vnodeops
, vop_mkdir
),
274 VOPNAME_RMDIR
, offsetof(struct vnodeops
, vop_rmdir
),
277 VOPNAME_READDIR
, offsetof(struct vnodeops
, vop_readdir
),
280 VOPNAME_SYMLINK
, offsetof(struct vnodeops
, vop_symlink
),
283 VOPNAME_READLINK
, offsetof(struct vnodeops
, vop_readlink
),
286 VOPNAME_FSYNC
, offsetof(struct vnodeops
, vop_fsync
),
289 VOPNAME_INACTIVE
, offsetof(struct vnodeops
, vop_inactive
),
292 VOPNAME_FID
, offsetof(struct vnodeops
, vop_fid
),
295 VOPNAME_RWLOCK
, offsetof(struct vnodeops
, vop_rwlock
),
296 fs_rwlock
, fs_rwlock
,
298 VOPNAME_RWUNLOCK
, offsetof(struct vnodeops
, vop_rwunlock
),
299 (fs_generic_func_p
) fs_rwunlock
,
300 (fs_generic_func_p
) fs_rwunlock
, /* no errors allowed */
302 VOPNAME_SEEK
, offsetof(struct vnodeops
, vop_seek
),
305 VOPNAME_CMP
, offsetof(struct vnodeops
, vop_cmp
),
306 fs_cmp
, fs_cmp
, /* no errors allowed */
308 VOPNAME_FRLOCK
, offsetof(struct vnodeops
, vop_frlock
),
311 VOPNAME_SPACE
, offsetof(struct vnodeops
, vop_space
),
314 VOPNAME_REALVP
, offsetof(struct vnodeops
, vop_realvp
),
317 VOPNAME_GETPAGE
, offsetof(struct vnodeops
, vop_getpage
),
320 VOPNAME_PUTPAGE
, offsetof(struct vnodeops
, vop_putpage
),
323 VOPNAME_MAP
, offsetof(struct vnodeops
, vop_map
),
324 (fs_generic_func_p
) fs_nosys_map
,
325 (fs_generic_func_p
) fs_nosys_map
,
327 VOPNAME_ADDMAP
, offsetof(struct vnodeops
, vop_addmap
),
328 (fs_generic_func_p
) fs_nosys_addmap
,
329 (fs_generic_func_p
) fs_nosys_addmap
,
331 VOPNAME_DELMAP
, offsetof(struct vnodeops
, vop_delmap
),
334 VOPNAME_POLL
, offsetof(struct vnodeops
, vop_poll
),
335 (fs_generic_func_p
) fs_poll
, (fs_generic_func_p
) fs_nosys_poll
,
337 VOPNAME_DUMP
, offsetof(struct vnodeops
, vop_dump
),
340 VOPNAME_PATHCONF
, offsetof(struct vnodeops
, vop_pathconf
),
341 fs_pathconf
, fs_nosys
,
343 VOPNAME_PAGEIO
, offsetof(struct vnodeops
, vop_pageio
),
346 VOPNAME_DUMPCTL
, offsetof(struct vnodeops
, vop_dumpctl
),
349 VOPNAME_DISPOSE
, offsetof(struct vnodeops
, vop_dispose
),
350 (fs_generic_func_p
) fs_dispose
,
351 (fs_generic_func_p
) fs_nodispose
,
353 VOPNAME_SETSECATTR
, offsetof(struct vnodeops
, vop_setsecattr
),
356 VOPNAME_GETSECATTR
, offsetof(struct vnodeops
, vop_getsecattr
),
357 fs_fab_acl
, fs_nosys
,
359 VOPNAME_SHRLOCK
, offsetof(struct vnodeops
, vop_shrlock
),
360 fs_shrlock
, fs_nosys
,
362 VOPNAME_VNEVENT
, offsetof(struct vnodeops
, vop_vnevent
),
363 (fs_generic_func_p
) fs_vnevent_nosupport
,
364 (fs_generic_func_p
) fs_vnevent_nosupport
,
366 VOPNAME_REQZCBUF
, offsetof(struct vnodeops
, vop_reqzcbuf
),
369 VOPNAME_RETZCBUF
, offsetof(struct vnodeops
, vop_retzcbuf
),
375 /* Extensible attribute (xva) routines. */
378 * Zero out the structure, set the size of the requested/returned bitmaps,
379 * set AT_XVATTR in the embedded vattr_t's va_mask, and set up the pointer
380 * to the returned attributes array.
383 xva_init(xvattr_t
*xvap
)
385 bzero(xvap
, sizeof (xvattr_t
));
386 xvap
->xva_mapsize
= XVA_MAPSIZE
;
387 xvap
->xva_magic
= XVA_MAGIC
;
388 xvap
->xva_vattr
.va_mask
= AT_XVATTR
;
389 xvap
->xva_rtnattrmapp
= &(xvap
->xva_rtnattrmap
)[0];
393 * If AT_XVATTR is set, returns a pointer to the embedded xoptattr_t
394 * structure. Otherwise, returns NULL.
397 xva_getxoptattr(xvattr_t
*xvap
)
399 xoptattr_t
*xoap
= NULL
;
400 if (xvap
->xva_vattr
.va_mask
& AT_XVATTR
)
401 xoap
= &xvap
->xva_xoptattrs
;
406 * Used by the AVL routines to compare two vsk_anchor_t structures in the tree.
407 * We use the f_fsid reported by VFS_STATVFS() since we use that for the
411 vska_compar(const void *n1
, const void *n2
)
414 ulong_t p1
= ((vsk_anchor_t
*)n1
)->vsk_fsid
;
415 ulong_t p2
= ((vsk_anchor_t
*)n2
)->vsk_fsid
;
419 } else if (p1
> p2
) {
429 * Used to create a single template which will be bcopy()ed to a newly
430 * allocated vsanchor_combo_t structure in new_vsanchor(), below.
433 create_vopstats_template()
437 vsp
= kmem_alloc(sizeof (vopstats_t
), KM_SLEEP
);
438 bzero(vsp
, sizeof (*vsp
)); /* Start fresh */
441 kstat_named_init(&vsp
->nopen
, "nopen", KSTAT_DATA_UINT64
);
443 kstat_named_init(&vsp
->nclose
, "nclose", KSTAT_DATA_UINT64
);
445 kstat_named_init(&vsp
->nread
, "nread", KSTAT_DATA_UINT64
);
446 kstat_named_init(&vsp
->read_bytes
, "read_bytes", KSTAT_DATA_UINT64
);
448 kstat_named_init(&vsp
->nwrite
, "nwrite", KSTAT_DATA_UINT64
);
449 kstat_named_init(&vsp
->write_bytes
, "write_bytes", KSTAT_DATA_UINT64
);
451 kstat_named_init(&vsp
->nioctl
, "nioctl", KSTAT_DATA_UINT64
);
453 kstat_named_init(&vsp
->nsetfl
, "nsetfl", KSTAT_DATA_UINT64
);
455 kstat_named_init(&vsp
->ngetattr
, "ngetattr", KSTAT_DATA_UINT64
);
457 kstat_named_init(&vsp
->nsetattr
, "nsetattr", KSTAT_DATA_UINT64
);
459 kstat_named_init(&vsp
->naccess
, "naccess", KSTAT_DATA_UINT64
);
461 kstat_named_init(&vsp
->nlookup
, "nlookup", KSTAT_DATA_UINT64
);
463 kstat_named_init(&vsp
->ncreate
, "ncreate", KSTAT_DATA_UINT64
);
465 kstat_named_init(&vsp
->nremove
, "nremove", KSTAT_DATA_UINT64
);
467 kstat_named_init(&vsp
->nlink
, "nlink", KSTAT_DATA_UINT64
);
469 kstat_named_init(&vsp
->nrename
, "nrename", KSTAT_DATA_UINT64
);
471 kstat_named_init(&vsp
->nmkdir
, "nmkdir", KSTAT_DATA_UINT64
);
473 kstat_named_init(&vsp
->nrmdir
, "nrmdir", KSTAT_DATA_UINT64
);
474 /* VOP_READDIR I/O */
475 kstat_named_init(&vsp
->nreaddir
, "nreaddir", KSTAT_DATA_UINT64
);
476 kstat_named_init(&vsp
->readdir_bytes
, "readdir_bytes",
479 kstat_named_init(&vsp
->nsymlink
, "nsymlink", KSTAT_DATA_UINT64
);
481 kstat_named_init(&vsp
->nreadlink
, "nreadlink", KSTAT_DATA_UINT64
);
483 kstat_named_init(&vsp
->nfsync
, "nfsync", KSTAT_DATA_UINT64
);
485 kstat_named_init(&vsp
->ninactive
, "ninactive", KSTAT_DATA_UINT64
);
487 kstat_named_init(&vsp
->nfid
, "nfid", KSTAT_DATA_UINT64
);
489 kstat_named_init(&vsp
->nrwlock
, "nrwlock", KSTAT_DATA_UINT64
);
491 kstat_named_init(&vsp
->nrwunlock
, "nrwunlock", KSTAT_DATA_UINT64
);
493 kstat_named_init(&vsp
->nseek
, "nseek", KSTAT_DATA_UINT64
);
495 kstat_named_init(&vsp
->ncmp
, "ncmp", KSTAT_DATA_UINT64
);
497 kstat_named_init(&vsp
->nfrlock
, "nfrlock", KSTAT_DATA_UINT64
);
499 kstat_named_init(&vsp
->nspace
, "nspace", KSTAT_DATA_UINT64
);
501 kstat_named_init(&vsp
->nrealvp
, "nrealvp", KSTAT_DATA_UINT64
);
503 kstat_named_init(&vsp
->ngetpage
, "ngetpage", KSTAT_DATA_UINT64
);
505 kstat_named_init(&vsp
->nputpage
, "nputpage", KSTAT_DATA_UINT64
);
507 kstat_named_init(&vsp
->nmap
, "nmap", KSTAT_DATA_UINT64
);
509 kstat_named_init(&vsp
->naddmap
, "naddmap", KSTAT_DATA_UINT64
);
511 kstat_named_init(&vsp
->ndelmap
, "ndelmap", KSTAT_DATA_UINT64
);
513 kstat_named_init(&vsp
->npoll
, "npoll", KSTAT_DATA_UINT64
);
515 kstat_named_init(&vsp
->ndump
, "ndump", KSTAT_DATA_UINT64
);
517 kstat_named_init(&vsp
->npathconf
, "npathconf", KSTAT_DATA_UINT64
);
519 kstat_named_init(&vsp
->npageio
, "npageio", KSTAT_DATA_UINT64
);
521 kstat_named_init(&vsp
->ndumpctl
, "ndumpctl", KSTAT_DATA_UINT64
);
523 kstat_named_init(&vsp
->ndispose
, "ndispose", KSTAT_DATA_UINT64
);
525 kstat_named_init(&vsp
->nsetsecattr
, "nsetsecattr", KSTAT_DATA_UINT64
);
527 kstat_named_init(&vsp
->ngetsecattr
, "ngetsecattr", KSTAT_DATA_UINT64
);
529 kstat_named_init(&vsp
->nshrlock
, "nshrlock", KSTAT_DATA_UINT64
);
531 kstat_named_init(&vsp
->nvnevent
, "nvnevent", KSTAT_DATA_UINT64
);
533 kstat_named_init(&vsp
->nreqzcbuf
, "nreqzcbuf", KSTAT_DATA_UINT64
);
535 kstat_named_init(&vsp
->nretzcbuf
, "nretzcbuf", KSTAT_DATA_UINT64
);
541 * Creates a kstat structure associated with a vopstats structure.
544 new_vskstat(char *ksname
, vopstats_t
*vsp
)
548 if (!vopstats_enabled
) {
552 ksp
= kstat_create("unix", 0, ksname
, "misc", KSTAT_TYPE_NAMED
,
553 sizeof (vopstats_t
)/sizeof (kstat_named_t
),
554 KSTAT_FLAG_VIRTUAL
|KSTAT_FLAG_WRITABLE
);
564 * Called from vfsinit() to initialize the support mechanisms for vopstats
569 if (!vopstats_enabled
)
573 * Creates the AVL tree which holds per-vfs vopstat anchors. This
574 * is necessary since we need to check if a kstat exists before we
575 * attempt to create it. Also, initialize its lock.
577 avl_create(&vskstat_tree
, vska_compar
, sizeof (vsk_anchor_t
),
578 offsetof(vsk_anchor_t
, vsk_node
));
579 mutex_init(&vskstat_tree_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
581 vsk_anchor_cache
= kmem_cache_create("vsk_anchor_cache",
582 sizeof (vsk_anchor_t
), sizeof (uintptr_t), NULL
, NULL
, NULL
,
586 * Set up the array of pointers for the vopstats-by-FS-type.
587 * The entries will be allocated/initialized as each file system
588 * goes through modload/mod_installfs.
590 vopstats_fstype
= (vopstats_t
**)kmem_zalloc(
591 (sizeof (vopstats_t
*) * nfstype
), KM_SLEEP
);
593 /* Set up the global vopstats initialization template */
594 vs_templatep
= create_vopstats_template();
598 * We need to have the all of the counters zeroed.
599 * The initialization of the vopstats_t includes on the order of
600 * 50 calls to kstat_named_init(). Rather that do that on every call,
601 * we do it once in a template (vs_templatep) then bcopy it over.
604 initialize_vopstats(vopstats_t
*vsp
)
609 bcopy(vs_templatep
, vsp
, sizeof (vopstats_t
));
613 * If possible, determine which vopstats by fstype to use and
614 * return a pointer to the caller.
617 get_fstype_vopstats(vfs_t
*vfsp
, struct vfssw
*vswp
)
619 int fstype
= 0; /* Index into vfssw[] */
620 vopstats_t
*vsp
= NULL
;
622 if (vfsp
== NULL
|| (vfsp
->vfs_flag
& VFS_STATS
) == 0 ||
626 * Set up the fstype. We go to so much trouble because all versions
627 * of NFS use the same fstype in their vfs even though they have
628 * distinct entries in the vfssw[] table.
629 * NOTE: A special vfs (e.g., EIO_vfs) may not have an entry.
632 fstype
= vswp
- vfssw
; /* Gets us the index */
634 fstype
= vfsp
->vfs_fstype
;
638 * Point to the per-fstype vopstats. The only valid values are
639 * non-zero positive values less than the number of vfssw[] table
642 if (fstype
> 0 && fstype
< nfstype
) {
643 vsp
= vopstats_fstype
[fstype
];
650 * Generate a kstat name, create the kstat structure, and allocate a
651 * vsk_anchor_t to hold it together. Return the pointer to the vsk_anchor_t
652 * to the caller. This must only be called from a mount.
655 get_vskstat_anchor(vfs_t
*vfsp
)
657 char kstatstr
[KSTAT_STRLEN
]; /* kstat name for vopstats */
658 statvfs64_t statvfsbuf
; /* Needed to find f_fsid */
659 vsk_anchor_t
*vskp
= NULL
; /* vfs <--> kstat anchor */
660 kstat_t
*ksp
; /* Ptr to new kstat */
661 avl_index_t where
; /* Location in the AVL tree */
663 if (vfsp
== NULL
|| vfsp
->vfs_implp
== NULL
||
664 (vfsp
->vfs_flag
& VFS_STATS
) == 0 || !vopstats_enabled
)
667 /* Need to get the fsid to build a kstat name */
668 if (VFS_STATVFS(vfsp
, &statvfsbuf
) == 0) {
669 /* Create a name for our kstats based on fsid */
670 (void) snprintf(kstatstr
, KSTAT_STRLEN
, "%s%lx",
671 VOPSTATS_STR
, statvfsbuf
.f_fsid
);
673 /* Allocate and initialize the vsk_anchor_t */
674 vskp
= kmem_cache_alloc(vsk_anchor_cache
, KM_SLEEP
);
675 bzero(vskp
, sizeof (*vskp
));
676 vskp
->vsk_fsid
= statvfsbuf
.f_fsid
;
678 mutex_enter(&vskstat_tree_lock
);
679 if (avl_find(&vskstat_tree
, vskp
, &where
) == NULL
) {
680 avl_insert(&vskstat_tree
, vskp
, where
);
681 mutex_exit(&vskstat_tree_lock
);
684 * Now that we've got the anchor in the AVL
685 * tree, we can create the kstat.
687 ksp
= new_vskstat(kstatstr
, &vfsp
->vfs_vopstats
);
692 /* Oops, found one! Release memory and lock. */
693 mutex_exit(&vskstat_tree_lock
);
694 kmem_cache_free(vsk_anchor_cache
, vskp
);
702 * We're in the process of tearing down the vfs and need to cleanup
703 * the data structures associated with the vopstats. Must only be called
707 teardown_vopstats(vfs_t
*vfsp
)
712 if (vfsp
== NULL
|| vfsp
->vfs_implp
== NULL
||
713 (vfsp
->vfs_flag
& VFS_STATS
) == 0 || !vopstats_enabled
)
716 /* This is a safe check since VFS_STATS must be set (see above) */
717 if ((vskap
= vfsp
->vfs_vskap
) == NULL
)
720 /* Whack the pointer right away */
721 vfsp
->vfs_vskap
= NULL
;
723 /* Lock the tree, remove the node, and delete the kstat */
724 mutex_enter(&vskstat_tree_lock
);
725 if (avl_find(&vskstat_tree
, vskap
, &where
)) {
726 avl_remove(&vskstat_tree
, vskap
);
729 if (vskap
->vsk_ksp
) {
730 kstat_delete(vskap
->vsk_ksp
);
732 mutex_exit(&vskstat_tree_lock
);
734 kmem_cache_free(vsk_anchor_cache
, vskap
);
738 * Read or write a vnode. Called from kernel code.
749 rlim64_t ulimit
, /* meaningful only if rw is UIO_WRITE */
758 if (rw
== UIO_WRITE
&& ISROFILE(vp
))
764 VOPXID_MAP_CR(vp
, cr
);
770 uio
.uio_loffset
= offset
;
771 uio
.uio_segflg
= (short)seg
;
773 uio
.uio_llimit
= ulimit
;
776 * We have to enter the critical region before calling VOP_RWLOCK
777 * to avoid a deadlock with ufs.
779 if (nbl_need_check(vp
)) {
782 nbl_start_crit(vp
, RW_READER
);
784 error
= nbl_svmand(vp
, cr
, &svmand
);
787 if (nbl_conflict(vp
, rw
== UIO_WRITE
? NBL_WRITE
: NBL_READ
,
788 uio
.uio_offset
, uio
.uio_resid
, svmand
, NULL
)) {
794 (void) VOP_RWLOCK(vp
,
795 rw
== UIO_WRITE
? V_WRITELOCK_TRUE
: V_WRITELOCK_FALSE
, NULL
);
796 if (rw
== UIO_WRITE
) {
797 uio
.uio_fmode
= FWRITE
;
798 uio
.uio_extflg
= UIO_COPY_DEFAULT
;
799 error
= VOP_WRITE(vp
, &uio
, ioflag
, cr
, NULL
);
801 uio
.uio_fmode
= FREAD
;
802 uio
.uio_extflg
= UIO_COPY_CACHED
;
803 error
= VOP_READ(vp
, &uio
, ioflag
, cr
, NULL
);
806 rw
== UIO_WRITE
? V_WRITELOCK_TRUE
: V_WRITELOCK_FALSE
, NULL
);
808 *residp
= uio
.uio_resid
;
809 else if (uio
.uio_resid
)
819 * Release a vnode. Call VOP_INACTIVE on last reference or
820 * decrement reference count.
822 * To avoid race conditions, the v_count is left at 1 for
823 * the call to VOP_INACTIVE. This prevents another thread
824 * from reclaiming and releasing the vnode *before* the
825 * VOP_INACTIVE routine has a chance to destroy the vnode.
826 * We can't have more than 1 thread calling VOP_INACTIVE
832 VERIFY(vp
->v_count
> 0);
833 mutex_enter(&vp
->v_lock
);
834 if (vp
->v_count
== 1) {
835 mutex_exit(&vp
->v_lock
);
836 VOP_INACTIVE(vp
, CRED(), NULL
);
840 mutex_exit(&vp
->v_lock
);
844 * Release a vnode referenced by the DNLC. Multiple DNLC references are treated
845 * as a single reference, so v_count is not decremented until the last DNLC hold
846 * is released. This makes it possible to distinguish vnodes that are referenced
850 vn_rele_dnlc(vnode_t
*vp
)
852 VERIFY((vp
->v_count
> 0) && (vp
->v_count_dnlc
> 0));
853 mutex_enter(&vp
->v_lock
);
854 if (--vp
->v_count_dnlc
== 0) {
855 if (vp
->v_count
== 1) {
856 mutex_exit(&vp
->v_lock
);
857 VOP_INACTIVE(vp
, CRED(), NULL
);
862 mutex_exit(&vp
->v_lock
);
866 * Like vn_rele() except that it clears v_stream under v_lock.
867 * This is used by sockfs when it dismantels the association between
868 * the sockfs node and the vnode in the underlaying file system.
869 * v_lock has to be held to prevent a thread coming through the lookupname
870 * path from accessing a stream head that is going away.
873 vn_rele_stream(vnode_t
*vp
)
875 VERIFY(vp
->v_count
> 0);
876 mutex_enter(&vp
->v_lock
);
878 if (vp
->v_count
== 1) {
879 mutex_exit(&vp
->v_lock
);
880 VOP_INACTIVE(vp
, CRED(), NULL
);
884 mutex_exit(&vp
->v_lock
);
888 vn_rele_inactive(vnode_t
*vp
)
890 VOP_INACTIVE(vp
, CRED(), NULL
);
894 * Like vn_rele() except if we are going to call VOP_INACTIVE() then do it
895 * asynchronously using a taskq. This can avoid deadlocks caused by re-entering
896 * the file system as a result of releasing the vnode. Note, file systems
897 * already have to handle the race where the vnode is incremented before the
898 * inactive routine is called and does its locking.
900 * Warning: Excessive use of this routine can lead to performance problems.
901 * This is because taskqs throttle back allocation if too many are created.
904 vn_rele_async(vnode_t
*vp
, taskq_t
*taskq
)
906 VERIFY(vp
->v_count
> 0);
907 mutex_enter(&vp
->v_lock
);
908 if (vp
->v_count
== 1) {
909 mutex_exit(&vp
->v_lock
);
910 VERIFY(taskq_dispatch(taskq
, (task_func_t
*)vn_rele_inactive
,
911 vp
, TQ_SLEEP
) != NULL
);
915 mutex_exit(&vp
->v_lock
);
928 return (vn_openat(pnamep
, seg
, filemode
, createmode
, vpp
, crwhy
,
934 * Open/create a vnode.
935 * This may be callable by the kernel, the only known use
936 * of user context being that the current user credentials
937 * are used for permissions. crwhy is defined iff filemode & FCREAT.
948 struct vnode
*startvp
,
957 int shrlock_done
= 0;
959 enum symfollow follow
;
960 int estale_retry
= 0;
962 struct shr_locowner shr_own
;
966 if (filemode
& FREAD
)
968 if (filemode
& (FWRITE
|FTRUNC
))
970 if (filemode
& (FSEARCH
|FEXEC
|FXATTRDIROPEN
))
973 /* symlink interpretation */
974 if (filemode
& FNOFOLLOW
)
979 if (filemode
& FAPPEND
)
980 accessflags
|= V_APPEND
;
983 if (filemode
& FCREAT
) {
987 * Wish to create a file.
989 vattr
.va_type
= VREG
;
990 vattr
.va_mode
= createmode
;
991 vattr
.va_mask
= AT_TYPE
|AT_MODE
;
992 if (filemode
& FTRUNC
) {
994 vattr
.va_mask
|= AT_SIZE
;
996 if (filemode
& FEXCL
)
1002 vn_createat(pnamep
, seg
, &vattr
, excl
, mode
, &vp
, crwhy
,
1003 (filemode
& ~(FTRUNC
|FEXCL
)), umask
, startvp
))
1007 * Wish to open a file. Just look it up.
1009 if (error
= lookupnameat(pnamep
, seg
, follow
,
1010 NULLVPP
, &vp
, startvp
)) {
1011 if ((error
== ESTALE
) &&
1012 fs_need_estale_retry(estale_retry
++))
1018 * Get the attributes to check whether file is large.
1019 * We do this only if the FOFFMAX flag is not set and
1020 * only for regular files.
1023 if (!(filemode
& FOFFMAX
) && (vp
->v_type
== VREG
)) {
1024 vattr
.va_mask
= AT_SIZE
;
1025 if ((error
= VOP_GETATTR(vp
, &vattr
, 0,
1029 if (vattr
.va_size
> (u_offset_t
)MAXOFF32_T
) {
1031 * Large File API - regular open fails
1032 * if FOFFMAX flag is set in file mode
1039 * Can't write directories, active texts, or
1040 * read-only filesystems. Can't truncate files
1041 * on which mandatory locking is in effect.
1043 if (filemode
& (FWRITE
|FTRUNC
)) {
1045 * Allow writable directory if VDIROPEN flag is set.
1047 if (vp
->v_type
== VDIR
&& !(vp
->v_flag
& VDIROPEN
)) {
1056 * Can't truncate files on which
1057 * sysv mandatory locking is in effect.
1059 if (filemode
& FTRUNC
) {
1062 if (VOP_REALVP(vp
, &rvp
, NULL
) != 0)
1064 if (rvp
->v_filocks
!= NULL
) {
1065 vattr
.va_mask
= AT_MODE
;
1066 if ((error
= VOP_GETATTR(vp
,
1067 &vattr
, 0, CRED(), NULL
)) == 0 &&
1068 MANDLOCK(vp
, vattr
.va_mode
))
1076 * Check permissions.
1078 if (error
= VOP_ACCESS(vp
, mode
, accessflags
, CRED(), NULL
))
1081 * Require FSEARCH to return a directory.
1082 * Require FEXEC to return a regular file.
1084 if ((filemode
& FSEARCH
) && vp
->v_type
!= VDIR
) {
1088 if ((filemode
& FEXEC
) && vp
->v_type
!= VREG
) {
1089 error
= ENOEXEC
; /* XXX: error code? */
1095 * Do remaining checks for FNOFOLLOW and FNOLINKS.
1097 if ((filemode
& FNOFOLLOW
) && vp
->v_type
== VLNK
) {
1101 if (filemode
& FNOLINKS
) {
1102 vattr
.va_mask
= AT_NLINK
;
1103 if ((error
= VOP_GETATTR(vp
, &vattr
, 0, CRED(), NULL
))) {
1106 if (vattr
.va_nlink
!= 1) {
1113 * Opening a socket corresponding to the AF_UNIX pathname
1114 * in the filesystem name space is not supported.
1115 * However, VSOCK nodes in namefs are supported in order
1116 * to make fattach work for sockets.
1118 * XXX This uses VOP_REALVP to distinguish between
1119 * an unopened namefs node (where VOP_REALVP returns a
1120 * different VSOCK vnode) and a VSOCK created by vn_create
1121 * in some file system (where VOP_REALVP would never return
1122 * a different vnode).
1124 if (vp
->v_type
== VSOCK
) {
1127 error
= VOP_REALVP(vp
, &nvp
, NULL
);
1128 if (error
!= 0 || nvp
== NULL
|| nvp
== vp
||
1129 nvp
->v_type
!= VSOCK
) {
1135 if ((vp
->v_type
== VREG
) && nbl_need_check(vp
)) {
1136 /* get share reservation */
1138 if (filemode
& FWRITE
)
1139 shr
.s_access
|= F_WRACC
;
1140 if (filemode
& FREAD
)
1141 shr
.s_access
|= F_RDACC
;
1144 shr
.s_pid
= ttoproc(curthread
)->p_pid
;
1145 shr_own
.sl_pid
= shr
.s_pid
;
1147 shr
.s_own_len
= sizeof (shr_own
);
1148 shr
.s_owner
= (caddr_t
)&shr_own
;
1149 error
= VOP_SHRLOCK(vp
, F_SHARE_NBMAND
, &shr
, filemode
, CRED(),
1155 /* nbmand conflict check if truncating file */
1156 if ((filemode
& FTRUNC
) && !(filemode
& FCREAT
)) {
1157 nbl_start_crit(vp
, RW_READER
);
1160 vattr
.va_mask
= AT_SIZE
;
1161 if (error
= VOP_GETATTR(vp
, &vattr
, 0, CRED(), NULL
))
1163 if (nbl_conflict(vp
, NBL_WRITE
, 0, vattr
.va_size
, 0,
1172 * Do opening protocol.
1174 error
= VOP_OPEN(&vp
, filemode
, CRED(), NULL
);
1180 * Truncate if required.
1182 if ((filemode
& FTRUNC
) && !(filemode
& FCREAT
)) {
1184 vattr
.va_mask
= AT_SIZE
;
1185 if ((error
= VOP_SETATTR(vp
, &vattr
, 0, CRED(), NULL
)) != 0)
1189 ASSERT(vp
->v_count
> 0);
1197 (void) VOP_CLOSE(vp
, filemode
, 1, (offset_t
)0, CRED(),
1203 (void) VOP_SHRLOCK(vp
, F_UNSHARE
, &shr
, 0, CRED(),
1209 * The following clause was added to handle a problem
1210 * with NFS consistency. It is possible that a lookup
1211 * of the file to be opened succeeded, but the file
1212 * itself doesn't actually exist on the server. This
1213 * is chiefly due to the DNLC containing an entry for
1214 * the file which has been removed on the server. In
1215 * this case, we just start over. If there was some
1216 * other cause for the ESTALE error, then the lookup
1217 * of the file will fail and the error will be returned
1218 * above instead of looping around from here.
1221 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1229 * The following two accessor functions are for the NFSv4 server. Since there
1230 * is no VOP_OPEN_UP/DOWNGRADE we need a way for the NFS server to keep the
1231 * vnode open counts correct when a client "upgrades" an open or does an
1232 * open_downgrade. In NFS, an upgrade or downgrade can not only change the
1233 * open mode (add or subtract read or write), but also change the share/deny
1234 * modes. However, share reservations are not integrated with OPEN, yet, so
1235 * we need to handle each separately. These functions are cleaner than having
1236 * the NFS server manipulate the counts directly, however, nobody else should
1237 * use these functions.
1244 ASSERT(vp
->v_type
== VREG
);
1246 if (filemode
& FREAD
)
1247 atomic_add_32(&(vp
->v_rdcnt
), 1);
1248 if (filemode
& FWRITE
)
1249 atomic_add_32(&(vp
->v_wrcnt
), 1);
1258 ASSERT(vp
->v_type
== VREG
);
1260 if (filemode
& FREAD
) {
1261 ASSERT(vp
->v_rdcnt
> 0);
1262 atomic_add_32(&(vp
->v_rdcnt
), -1);
1264 if (filemode
& FWRITE
) {
1265 ASSERT(vp
->v_wrcnt
> 0);
1266 atomic_add_32(&(vp
->v_wrcnt
), -1);
1283 return (vn_createat(pnamep
, seg
, vap
, excl
, mode
, vpp
, why
, flag
,
1288 * Create a vnode (makenode).
1301 struct vnode
*startvp
)
1303 struct vnode
*dvp
; /* ptr to parent dir vnode */
1304 struct vnode
*vp
= NULL
;
1309 enum symfollow follow
;
1310 int estale_retry
= 0;
1311 uint32_t auditing
= AU_AUDITING();
1313 ASSERT((vap
->va_mask
& (AT_TYPE
|AT_MODE
)) == (AT_TYPE
|AT_MODE
));
1315 /* symlink interpretation */
1316 if ((flag
& FNOFOLLOW
) || excl
== EXCL
)
1320 flag
&= ~(FNOFOLLOW
|FNOLINKS
);
1325 * If new object is a file, call lower level to create it.
1326 * Note that it is up to the lower level to enforce exclusive
1327 * creation, if the file is already there.
1328 * This allows the lower level to do whatever
1329 * locking or protocol that is needed to prevent races.
1330 * If the new object is directory call lower level to make
1331 * the new directory, with "." and "..".
1333 if (error
= pn_get(pnamep
, seg
, &pn
))
1336 audit_vncreate_start();
1340 * lookup will find the parent directory for the vnode.
1341 * When it is done the pn holds the name of the entry
1343 * If this is a non-exclusive create we also find the node itself.
1345 error
= lookuppnat(&pn
, NULL
, follow
, &dvp
,
1346 (excl
== EXCL
) ? NULLVPP
: vpp
, startvp
);
1349 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1351 if (why
== CRMKDIR
&& error
== EINVAL
)
1352 error
= EEXIST
; /* SVID */
1357 vap
->va_mode
&= ~VSVTX
;
1360 * If default ACLs are defined for the directory don't apply the
1361 * umask if umask is passed.
1368 vsec
.vsa_aclcnt
= 0;
1369 vsec
.vsa_aclentp
= NULL
;
1370 vsec
.vsa_dfaclcnt
= 0;
1371 vsec
.vsa_dfaclentp
= NULL
;
1372 vsec
.vsa_mask
= VSA_DFACLCNT
;
1373 error
= VOP_GETSECATTR(dvp
, &vsec
, 0, CRED(), NULL
);
1375 * If error is ENOSYS then treat it as no error
1376 * Don't want to force all file systems to support
1377 * aclent_t style of ACL's.
1379 if (error
== ENOSYS
)
1387 * Apply the umask if no default ACLs.
1389 if (vsec
.vsa_dfaclcnt
== 0)
1390 vap
->va_mode
&= ~umask
;
1393 * VOP_GETSECATTR() may have allocated memory for
1394 * ACLs we didn't request, so double-check and
1395 * free it if necessary.
1397 if (vsec
.vsa_aclcnt
&& vsec
.vsa_aclentp
!= NULL
)
1398 kmem_free((caddr_t
)vsec
.vsa_aclentp
,
1399 vsec
.vsa_aclcnt
* sizeof (aclent_t
));
1400 if (vsec
.vsa_dfaclcnt
&& vsec
.vsa_dfaclentp
!= NULL
)
1401 kmem_free((caddr_t
)vsec
.vsa_dfaclentp
,
1402 vsec
.vsa_dfaclcnt
* sizeof (aclent_t
));
1407 * In general we want to generate EROFS if the file system is
1408 * readonly. However, POSIX (IEEE Std. 1003.1) section 5.3.1
1409 * documents the open system call, and it says that O_CREAT has no
1410 * effect if the file already exists. Bug 1119649 states
1411 * that open(path, O_CREAT, ...) fails when attempting to open an
1412 * existing file on a read only file system. Thus, the first part
1413 * of the following if statement has 3 checks:
1414 * if the file exists &&
1415 * it is being open with write access &&
1416 * the file system is read only
1417 * then generate EROFS
1419 if ((*vpp
!= NULL
&& (mode
& VWRITE
) && ISROFILE(*vpp
)) ||
1420 (*vpp
== NULL
&& dvp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)) {
1424 } else if (excl
== NONEXCL
&& *vpp
!= NULL
) {
1428 * File already exists. If a mandatory lock has been
1429 * applied, return error.
1432 if (VOP_REALVP(vp
, &rvp
, NULL
) != 0)
1434 if ((vap
->va_mask
& AT_SIZE
) && nbl_need_check(vp
)) {
1435 nbl_start_crit(vp
, RW_READER
);
1438 if (rvp
->v_filocks
!= NULL
|| rvp
->v_shrlocks
!= NULL
) {
1439 vattr
.va_mask
= AT_MODE
|AT_SIZE
;
1440 if (error
= VOP_GETATTR(vp
, &vattr
, 0, CRED(), NULL
)) {
1443 if (MANDLOCK(vp
, vattr
.va_mode
)) {
1448 * File cannot be truncated if non-blocking mandatory
1449 * locks are currently on the file.
1451 if ((vap
->va_mask
& AT_SIZE
) && in_crit
) {
1455 offset
= vap
->va_size
> vattr
.va_size
?
1456 vattr
.va_size
: vap
->va_size
;
1457 length
= vap
->va_size
> vattr
.va_size
?
1458 vap
->va_size
- vattr
.va_size
:
1459 vattr
.va_size
- vap
->va_size
;
1460 if (nbl_conflict(vp
, NBL_WRITE
, offset
,
1469 * If the file is the root of a VFS, we've crossed a
1470 * mount point and the "containing" directory that we
1471 * acquired above (dvp) is irrelevant because it's in
1472 * a different file system. We apply VOP_CREATE to the
1473 * target itself instead of to the containing directory
1474 * and supply a null path name to indicate (conventionally)
1475 * the node itself as the "component" of interest.
1477 * The intercession of the file system is necessary to
1478 * ensure that the appropriate permission checks are
1481 if (vp
->v_flag
& VROOT
) {
1482 ASSERT(why
!= CRMKDIR
);
1483 error
= VOP_CREATE(vp
, "", vap
, excl
, mode
, vpp
,
1484 CRED(), flag
, NULL
, NULL
);
1486 * If the create succeeded, it will have created
1487 * a new reference to the vnode. Give up the
1488 * original reference. The assertion should not
1489 * get triggered because NBMAND locks only apply to
1490 * VREG files. And if in_crit is non-zero for some
1491 * reason, detect that here, rather than when we
1492 * deference a null vp.
1494 ASSERT(in_crit
== 0);
1501 * Large File API - non-large open (FOFFMAX flag not set)
1502 * of regular file fails if the file size exceeds MAXOFF32_T.
1504 if (why
!= CRMKDIR
&&
1505 !(flag
& FOFFMAX
) &&
1506 (vp
->v_type
== VREG
)) {
1507 vattr
.va_mask
= AT_SIZE
;
1508 if ((error
= VOP_GETATTR(vp
, &vattr
, 0,
1512 if ((vattr
.va_size
> (u_offset_t
)MAXOFF32_T
)) {
1521 * Call mkdir() if specified, otherwise create().
1523 int must_be_dir
= pn_fixslash(&pn
); /* trailing '/'? */
1527 * N.B., if vn_createat() ever requests
1528 * case-insensitive behavior then it will need
1529 * to be passed to VOP_MKDIR(). VOP_CREATE()
1530 * will already get it via "flag"
1532 error
= VOP_MKDIR(dvp
, pn
.pn_path
, vap
, vpp
, CRED(),
1534 else if (!must_be_dir
)
1535 error
= VOP_CREATE(dvp
, pn
.pn_path
, vap
,
1536 excl
, mode
, vpp
, CRED(), flag
, NULL
, NULL
);
1544 audit_vncreate_finish(*vpp
, error
);
1556 * The following clause was added to handle a problem
1557 * with NFS consistency. It is possible that a lookup
1558 * of the file to be created succeeded, but the file
1559 * itself doesn't actually exist on the server. This
1560 * is chiefly due to the DNLC containing an entry for
1561 * the file which has been removed on the server. In
1562 * this case, we just start over. If there was some
1563 * other cause for the ESTALE error, then the lookup
1564 * of the file will fail and the error will be returned
1565 * above instead of looping around from here.
1567 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1573 vn_link(char *from
, char *to
, enum uio_seg seg
)
1575 return (vn_linkat(NULL
, from
, NO_FOLLOW
, NULL
, to
, seg
));
1579 vn_linkat(vnode_t
*fstartvp
, char *from
, enum symfollow follow
,
1580 vnode_t
*tstartvp
, char *to
, enum uio_seg seg
)
1582 struct vnode
*fvp
; /* from vnode ptr */
1583 struct vnode
*tdvp
; /* to directory vnode ptr */
1588 int estale_retry
= 0;
1589 uint32_t auditing
= AU_AUDITING();
1593 if (error
= pn_get(to
, seg
, &pn
))
1595 if (auditing
&& fstartvp
!= NULL
)
1596 audit_setfsat_path(1);
1597 if (error
= lookupnameat(from
, seg
, follow
, NULLVPP
, &fvp
, fstartvp
))
1599 if (auditing
&& tstartvp
!= NULL
)
1600 audit_setfsat_path(3);
1601 if (error
= lookuppnat(&pn
, NULL
, NO_FOLLOW
, &tdvp
, NULLVPP
, tstartvp
))
1604 * Make sure both source vnode and target directory vnode are
1605 * in the same vfs and that it is writeable.
1607 vattr
.va_mask
= AT_FSID
;
1608 if (error
= VOP_GETATTR(fvp
, &vattr
, 0, CRED(), NULL
))
1610 fsid
= vattr
.va_fsid
;
1611 vattr
.va_mask
= AT_FSID
;
1612 if (error
= VOP_GETATTR(tdvp
, &vattr
, 0, CRED(), NULL
))
1614 if (fsid
!= vattr
.va_fsid
) {
1618 if (tdvp
->v_vfsp
->vfs_flag
& VFS_RDONLY
) {
1625 (void) pn_fixslash(&pn
);
1626 error
= VOP_LINK(tdvp
, fvp
, pn
.pn_path
, CRED(), NULL
, 0);
1633 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1639 vn_rename(char *from
, char *to
, enum uio_seg seg
)
1641 return (vn_renameat(NULL
, from
, NULL
, to
, seg
));
1645 vn_renameat(vnode_t
*fdvp
, char *fname
, vnode_t
*tdvp
,
1646 char *tname
, enum uio_seg seg
)
1650 struct pathname fpn
; /* from pathname */
1651 struct pathname tpn
; /* to pathname */
1653 int in_crit_src
, in_crit_targ
;
1654 vnode_t
*fromvp
, *fvp
;
1655 vnode_t
*tovp
, *targvp
;
1656 int estale_retry
= 0;
1657 uint32_t auditing
= AU_AUDITING();
1660 fvp
= fromvp
= tovp
= targvp
= NULL
;
1661 in_crit_src
= in_crit_targ
= 0;
1663 * Get to and from pathnames.
1665 if (error
= pn_get(fname
, seg
, &fpn
))
1667 if (error
= pn_get(tname
, seg
, &tpn
)) {
1673 * First we need to resolve the correct directories
1674 * The passed in directories may only be a starting point,
1675 * but we need the real directories the file(s) live in.
1676 * For example the fname may be something like usr/lib/sparc
1677 * and we were passed in the / directory, but we need to
1678 * use the lib directory for the rename.
1681 if (auditing
&& fdvp
!= NULL
)
1682 audit_setfsat_path(1);
1684 * Lookup to and from directories.
1686 if (error
= lookuppnat(&fpn
, NULL
, NO_FOLLOW
, &fromvp
, &fvp
, fdvp
)) {
1691 * Make sure there is an entry.
1698 if (auditing
&& tdvp
!= NULL
)
1699 audit_setfsat_path(3);
1700 if (error
= lookuppnat(&tpn
, NULL
, NO_FOLLOW
, &tovp
, &targvp
, tdvp
)) {
1705 * Make sure both the from vnode directory and the to directory
1706 * are in the same vfs and the to directory is writable.
1707 * We check fsid's, not vfs pointers, so loopback fs works.
1709 if (fromvp
!= tovp
) {
1710 vattr
.va_mask
= AT_FSID
;
1711 if (error
= VOP_GETATTR(fromvp
, &vattr
, 0, CRED(), NULL
))
1713 fsid
= vattr
.va_fsid
;
1714 vattr
.va_mask
= AT_FSID
;
1715 if (error
= VOP_GETATTR(tovp
, &vattr
, 0, CRED(), NULL
))
1717 if (fsid
!= vattr
.va_fsid
) {
1723 if (tovp
->v_vfsp
->vfs_flag
& VFS_RDONLY
) {
1728 if (targvp
&& (fvp
!= targvp
)) {
1729 nbl_start_crit(targvp
, RW_READER
);
1731 if (nbl_conflict(targvp
, NBL_REMOVE
, 0, 0, 0, NULL
)) {
1737 if (nbl_need_check(fvp
)) {
1738 nbl_start_crit(fvp
, RW_READER
);
1740 if (nbl_conflict(fvp
, NBL_RENAME
, 0, 0, 0, NULL
)) {
1749 (void) pn_fixslash(&tpn
);
1750 error
= VOP_RENAME(fromvp
, fpn
.pn_path
, tovp
, tpn
.pn_path
, CRED(),
1759 nbl_end_crit(targvp
);
1768 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1774 * Remove a file or directory.
1777 vn_remove(char *fnamep
, enum uio_seg seg
, enum rm dirflag
)
1779 return (vn_removeat(NULL
, fnamep
, seg
, dirflag
));
1783 vn_removeat(vnode_t
*startvp
, char *fnamep
, enum uio_seg seg
, enum rm dirflag
)
1785 struct vnode
*vp
; /* entry vnode */
1786 struct vnode
*dvp
; /* ptr to parent dir vnode */
1787 struct vnode
*coveredvp
;
1788 struct pathname pn
; /* name of entry */
1792 struct vfs
*dvfsp
; /* ptr to parent dir vfs */
1794 int estale_retry
= 0;
1797 if (error
= pn_get(fnamep
, seg
, &pn
))
1800 if (error
= lookuppnat(&pn
, NULL
, NO_FOLLOW
, &dvp
, &vp
, startvp
)) {
1802 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1808 * Make sure there is an entry.
1816 dvfsp
= dvp
->v_vfsp
;
1819 * If the named file is the root of a mounted filesystem, fail,
1820 * unless it's marked unlinkable. In that case, unmount the
1821 * filesystem and proceed to unlink the covered vnode. (If the
1822 * covered vnode is a directory, use rmdir instead of unlink,
1823 * to avoid file system corruption.)
1825 if (vp
->v_flag
& VROOT
) {
1826 if ((vfsp
->vfs_flag
& VFS_UNLINKABLE
) == 0) {
1832 * Namefs specific code starts here.
1835 if (dirflag
== RMDIRECTORY
) {
1837 * User called rmdir(2) on a file that has
1838 * been namefs mounted on top of. Since
1839 * namefs doesn't allow directories to
1840 * be mounted on other files we know
1841 * vp is not of type VDIR so fail to operation.
1848 * If VROOT is still set after grabbing vp->v_lock,
1849 * noone has finished nm_unmount so far and coveredvp
1851 * If we manage to grab vn_vfswlock(coveredvp) before releasing
1852 * vp->v_lock, any race window is eliminated.
1855 mutex_enter(&vp
->v_lock
);
1856 if ((vp
->v_flag
& VROOT
) == 0) {
1857 /* Someone beat us to the unmount */
1858 mutex_exit(&vp
->v_lock
);
1863 coveredvp
= vfsp
->vfs_vnodecovered
;
1866 * Note: Implementation of vn_vfswlock shows that ordering of
1867 * v_lock / vn_vfswlock is not an issue here.
1869 error
= vn_vfswlock(coveredvp
);
1870 mutex_exit(&vp
->v_lock
);
1877 error
= dounmount(vfsp
, 0, CRED());
1880 * Unmounted the namefs file system; now get
1881 * the object it was mounted over.
1885 * If namefs was mounted over a directory, then
1886 * we want to use rmdir() instead of unlink().
1888 if (vp
->v_type
== VDIR
)
1889 dirflag
= RMDIRECTORY
;
1896 * Make sure filesystem is writeable.
1897 * We check the parent directory's vfs in case this is an lofs vnode.
1899 if (dvfsp
&& dvfsp
->vfs_flag
& VFS_RDONLY
) {
1907 * If there is the possibility of an nbmand share reservation, make
1908 * sure it's okay to remove the file. Keep a reference to the
1909 * vnode, so that we can exit the nbl critical region after
1910 * calling VOP_REMOVE.
1911 * If there is no possibility of an nbmand share reservation,
1912 * release the vnode reference now. Filesystems like NFS may
1913 * behave differently if there is an extra reference, so get rid of
1914 * this one. Fortunately, we can't have nbmand mounts on NFS
1917 if (nbl_need_check(vp
)) {
1918 nbl_start_crit(vp
, RW_READER
);
1920 if (nbl_conflict(vp
, NBL_REMOVE
, 0, 0, 0, NULL
)) {
1929 if (dirflag
== RMDIRECTORY
) {
1931 * Caller is using rmdir(2), which can only be applied to
1934 if (vtype
!= VDIR
) {
1938 proc_t
*pp
= curproc
;
1940 mutex_enter(&pp
->p_lock
);
1941 cwd
= PTOU(pp
)->u_cdir
;
1943 mutex_exit(&pp
->p_lock
);
1944 error
= VOP_RMDIR(dvp
, pn
.pn_path
, cwd
, CRED(),
1950 * Unlink(2) can be applied to anything.
1952 error
= VOP_REMOVE(dvp
, pn
.pn_path
, CRED(), NULL
, 0);
1965 if ((error
== ESTALE
) && fs_need_estale_retry(estale_retry
++))
1971 * Utility function to compare equality of vnodes.
1972 * Compare the underlying real vnodes, if there are underlying vnodes.
1973 * This is a more thorough comparison than the VN_CMP() macro provides.
1976 vn_compare(vnode_t
*vp1
, vnode_t
*vp2
)
1980 if (vp1
!= NULL
&& VOP_REALVP(vp1
, &realvp
, NULL
) == 0)
1982 if (vp2
!= NULL
&& VOP_REALVP(vp2
, &realvp
, NULL
) == 0)
1984 return (VN_CMP(vp1
, vp2
));
1988 * The number of locks to hash into. This value must be a power
1989 * of 2 minus 1 and should probably also be prime.
1991 #define NUM_BUCKETS 1023
1993 struct vn_vfslocks_bucket
{
1995 vn_vfslocks_entry_t
*vb_list
;
1996 char pad
[64 - sizeof (kmutex_t
) - sizeof (void *)];
2000 * Total number of buckets will be NUM_BUCKETS + 1 .
2003 #pragma align 64(vn_vfslocks_buckets)
2004 static struct vn_vfslocks_bucket vn_vfslocks_buckets
[NUM_BUCKETS
+ 1];
2006 #define VN_VFSLOCKS_SHIFT 9
2008 #define VN_VFSLOCKS_HASH(vfsvpptr) \
2009 ((((intptr_t)(vfsvpptr)) >> VN_VFSLOCKS_SHIFT) & NUM_BUCKETS)
2012 * vn_vfslocks_getlock() uses an HASH scheme to generate
2013 * rwstlock using vfs/vnode pointer passed to it.
2015 * vn_vfslocks_rele() releases a reference in the
2016 * HASH table which allows the entry allocated by
2017 * vn_vfslocks_getlock() to be freed at a later
2018 * stage when the refcount drops to zero.
2021 vn_vfslocks_entry_t
*
2022 vn_vfslocks_getlock(void *vfsvpptr
)
2024 struct vn_vfslocks_bucket
*bp
;
2025 vn_vfslocks_entry_t
*vep
;
2026 vn_vfslocks_entry_t
*tvep
;
2028 ASSERT(vfsvpptr
!= NULL
);
2029 bp
= &vn_vfslocks_buckets
[VN_VFSLOCKS_HASH(vfsvpptr
)];
2031 mutex_enter(&bp
->vb_lock
);
2032 for (vep
= bp
->vb_list
; vep
!= NULL
; vep
= vep
->ve_next
) {
2033 if (vep
->ve_vpvfs
== vfsvpptr
) {
2035 mutex_exit(&bp
->vb_lock
);
2039 mutex_exit(&bp
->vb_lock
);
2040 vep
= kmem_alloc(sizeof (*vep
), KM_SLEEP
);
2041 rwst_init(&vep
->ve_lock
, NULL
, RW_DEFAULT
, NULL
);
2042 vep
->ve_vpvfs
= (char *)vfsvpptr
;
2044 mutex_enter(&bp
->vb_lock
);
2045 for (tvep
= bp
->vb_list
; tvep
!= NULL
; tvep
= tvep
->ve_next
) {
2046 if (tvep
->ve_vpvfs
== vfsvpptr
) {
2048 mutex_exit(&bp
->vb_lock
);
2051 * There is already an entry in the hash
2052 * destroy what we just allocated.
2054 rwst_destroy(&vep
->ve_lock
);
2055 kmem_free(vep
, sizeof (*vep
));
2059 vep
->ve_next
= bp
->vb_list
;
2061 mutex_exit(&bp
->vb_lock
);
2066 vn_vfslocks_rele(vn_vfslocks_entry_t
*vepent
)
2068 struct vn_vfslocks_bucket
*bp
;
2069 vn_vfslocks_entry_t
*vep
;
2070 vn_vfslocks_entry_t
*pvep
;
2072 ASSERT(vepent
!= NULL
);
2073 ASSERT(vepent
->ve_vpvfs
!= NULL
);
2075 bp
= &vn_vfslocks_buckets
[VN_VFSLOCKS_HASH(vepent
->ve_vpvfs
)];
2077 mutex_enter(&bp
->vb_lock
);
2078 vepent
->ve_refcnt
--;
2080 if ((int32_t)vepent
->ve_refcnt
< 0)
2081 cmn_err(CE_PANIC
, "vn_vfslocks_rele: refcount negative");
2083 if (vepent
->ve_refcnt
== 0) {
2084 for (vep
= bp
->vb_list
; vep
!= NULL
; vep
= vep
->ve_next
) {
2085 if (vep
->ve_vpvfs
== vepent
->ve_vpvfs
) {
2086 if (bp
->vb_list
== vep
)
2087 bp
->vb_list
= vep
->ve_next
;
2090 pvep
->ve_next
= vep
->ve_next
;
2092 mutex_exit(&bp
->vb_lock
);
2093 rwst_destroy(&vep
->ve_lock
);
2094 kmem_free(vep
, sizeof (*vep
));
2099 cmn_err(CE_PANIC
, "vn_vfslocks_rele: vp/vfs not found");
2101 mutex_exit(&bp
->vb_lock
);
2105 * vn_vfswlock_wait is used to implement a lock which is logically a writers
2106 * lock protecting the v_vfsmountedhere field.
2107 * vn_vfswlock_wait has been modified to be similar to vn_vfswlock,
2108 * except that it blocks to acquire the lock VVFSLOCK.
2110 * traverse() and routines re-implementing part of traverse (e.g. autofs)
2111 * need to hold this lock. mount(), vn_rename(), vn_remove() and so on
2112 * need the non-blocking version of the writers lock i.e. vn_vfswlock
2115 vn_vfswlock_wait(vnode_t
*vp
)
2118 vn_vfslocks_entry_t
*vpvfsentry
;
2121 vpvfsentry
= vn_vfslocks_getlock(vp
);
2122 retval
= rwst_enter_sig(&vpvfsentry
->ve_lock
, RW_WRITER
);
2124 if (retval
== EINTR
) {
2125 vn_vfslocks_rele(vpvfsentry
);
2132 vn_vfsrlock_wait(vnode_t
*vp
)
2135 vn_vfslocks_entry_t
*vpvfsentry
;
2138 vpvfsentry
= vn_vfslocks_getlock(vp
);
2139 retval
= rwst_enter_sig(&vpvfsentry
->ve_lock
, RW_READER
);
2141 if (retval
== EINTR
) {
2142 vn_vfslocks_rele(vpvfsentry
);
2151 * vn_vfswlock is used to implement a lock which is logically a writers lock
2152 * protecting the v_vfsmountedhere field.
2155 vn_vfswlock(vnode_t
*vp
)
2157 vn_vfslocks_entry_t
*vpvfsentry
;
2160 * If vp is NULL then somebody is trying to lock the covered vnode
2161 * of /. (vfs_vnodecovered is NULL for /). This situation will
2162 * only happen when unmounting /. Since that operation will fail
2163 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2168 vpvfsentry
= vn_vfslocks_getlock(vp
);
2170 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_WRITER
))
2173 vn_vfslocks_rele(vpvfsentry
);
2178 vn_vfsrlock(vnode_t
*vp
)
2180 vn_vfslocks_entry_t
*vpvfsentry
;
2183 * If vp is NULL then somebody is trying to lock the covered vnode
2184 * of /. (vfs_vnodecovered is NULL for /). This situation will
2185 * only happen when unmounting /. Since that operation will fail
2186 * anyway, return EBUSY here instead of in VFS_UNMOUNT.
2191 vpvfsentry
= vn_vfslocks_getlock(vp
);
2193 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_READER
))
2196 vn_vfslocks_rele(vpvfsentry
);
2201 vn_vfsunlock(vnode_t
*vp
)
2203 vn_vfslocks_entry_t
*vpvfsentry
;
2206 * ve_refcnt needs to be decremented twice.
2207 * 1. To release refernce after a call to vn_vfslocks_getlock()
2208 * 2. To release the reference from the locking routines like
2209 * vn_vfsrlock/vn_vfswlock etc,.
2211 vpvfsentry
= vn_vfslocks_getlock(vp
);
2212 vn_vfslocks_rele(vpvfsentry
);
2214 rwst_exit(&vpvfsentry
->ve_lock
);
2215 vn_vfslocks_rele(vpvfsentry
);
2219 vn_vfswlock_held(vnode_t
*vp
)
2222 vn_vfslocks_entry_t
*vpvfsentry
;
2226 vpvfsentry
= vn_vfslocks_getlock(vp
);
2227 held
= rwst_lock_held(&vpvfsentry
->ve_lock
, RW_WRITER
);
2229 vn_vfslocks_rele(vpvfsentry
);
2236 const char *name
, /* Name of file system */
2237 const fs_operation_def_t
*templ
, /* Operation specification */
2238 vnodeops_t
**actual
) /* Return the vnodeops */
2243 *actual
= (vnodeops_t
*)kmem_alloc(sizeof (vnodeops_t
), KM_SLEEP
);
2245 (*actual
)->vnop_name
= name
;
2247 error
= fs_build_vector(*actual
, &unused_ops
, vn_ops_table
, templ
);
2249 kmem_free(*actual
, sizeof (vnodeops_t
));
2253 if (unused_ops
!= 0)
2254 cmn_err(CE_WARN
, "vn_make_ops: %s: %d operations supplied "
2255 "but not used", name
, unused_ops
);
2262 * Free the vnodeops created as a result of vn_make_ops()
2265 vn_freevnodeops(vnodeops_t
*vnops
)
2267 kmem_free(vnops
, sizeof (vnodeops_t
));
2276 vn_cache_constructor(void *buf
, void *cdrarg
, int kmflags
)
2282 mutex_init(&vp
->v_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2283 mutex_init(&vp
->v_vsd_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2284 cv_init(&vp
->v_cv
, NULL
, CV_DEFAULT
, NULL
);
2285 rw_init(&vp
->v_nbllock
, NULL
, RW_DEFAULT
, NULL
);
2286 vp
->v_femhead
= NULL
; /* Must be done before vn_reinit() */
2288 vp
->v_mpssdata
= NULL
;
2290 vp
->v_fopdata
= NULL
;
2297 vn_cache_destructor(void *buf
, void *cdrarg
)
2303 rw_destroy(&vp
->v_nbllock
);
2304 cv_destroy(&vp
->v_cv
);
2305 mutex_destroy(&vp
->v_vsd_lock
);
2306 mutex_destroy(&vp
->v_lock
);
2310 vn_create_cache(void)
2313 ASSERT((1 << VNODE_ALIGN_LOG2
) ==
2314 P2ROUNDUP(sizeof (struct vnode
), VNODE_ALIGN
));
2315 vn_cache
= kmem_cache_create("vn_cache", sizeof (struct vnode
),
2316 VNODE_ALIGN
, vn_cache_constructor
, vn_cache_destructor
, NULL
, NULL
,
2321 vn_destroy_cache(void)
2323 kmem_cache_destroy(vn_cache
);
2327 * Used by file systems when fs-specific nodes (e.g., ufs inodes) are
2328 * cached by the file system and vnodes remain associated.
2331 vn_recycle(vnode_t
*vp
)
2333 ASSERT(vp
->v_pages
== NULL
);
2336 * XXX - This really belongs in vn_reinit(), but we have some issues
2337 * with the counts. Best to have it here for clean initialization.
2341 vp
->v_mmap_read
= 0;
2342 vp
->v_mmap_write
= 0;
2345 * If FEM was in use, make sure everything gets cleaned up
2346 * NOTE: vp->v_femhead is initialized to NULL in the vnode
2349 if (vp
->v_femhead
) {
2350 /* XXX - There should be a free_femhead() that does all this */
2351 ASSERT(vp
->v_femhead
->femh_list
== NULL
);
2352 mutex_destroy(&vp
->v_femhead
->femh_lock
);
2353 kmem_free(vp
->v_femhead
, sizeof (*(vp
->v_femhead
)));
2354 vp
->v_femhead
= NULL
;
2357 kmem_free(vp
->v_path
, strlen(vp
->v_path
) + 1);
2361 if (vp
->v_fopdata
!= NULL
) {
2364 vp
->v_mpssdata
= NULL
;
2369 * Used to reset the vnode fields including those that are directly accessible
2370 * as well as those which require an accessor function.
2372 * Does not initialize:
2373 * synchronization objects: v_lock, v_vsd_lock, v_nbllock, v_cv
2374 * v_data (since FS-nodes and vnodes point to each other and should
2375 * be updated simultaneously)
2376 * v_op (in case someone needs to make a VOP call on this object)
2379 vn_reinit(vnode_t
*vp
)
2382 vp
->v_count_dnlc
= 0;
2384 vp
->v_stream
= NULL
;
2385 vp
->v_vfsmountedhere
= NULL
;
2390 vp
->v_filocks
= NULL
;
2391 vp
->v_shrlocks
= NULL
;
2394 vp
->v_locality
= NULL
;
2395 vp
->v_xattrdir
= NULL
;
2397 /* Handles v_femhead, v_path, and the r/w/map counts */
2402 vn_alloc(int kmflag
)
2406 vp
= kmem_cache_alloc(vn_cache
, kmflag
);
2409 vp
->v_femhead
= NULL
; /* Must be done before vn_reinit() */
2410 vp
->v_fopdata
= NULL
;
2418 vn_free(vnode_t
*vp
)
2420 ASSERT(vp
->v_shrlocks
== NULL
);
2421 ASSERT(vp
->v_filocks
== NULL
);
2424 * Some file systems call vn_free() with v_count of zero,
2425 * some with v_count of 1. In any case, the value should
2426 * never be anything else.
2428 ASSERT((vp
->v_count
== 0) || (vp
->v_count
== 1));
2429 ASSERT(vp
->v_count_dnlc
== 0);
2430 if (vp
->v_path
!= NULL
) {
2431 kmem_free(vp
->v_path
, strlen(vp
->v_path
) + 1);
2435 /* If FEM was in use, make sure everything gets cleaned up */
2436 if (vp
->v_femhead
) {
2437 /* XXX - There should be a free_femhead() that does all this */
2438 ASSERT(vp
->v_femhead
->femh_list
== NULL
);
2439 mutex_destroy(&vp
->v_femhead
->femh_lock
);
2440 kmem_free(vp
->v_femhead
, sizeof (*(vp
->v_femhead
)));
2441 vp
->v_femhead
= NULL
;
2444 if (vp
->v_fopdata
!= NULL
) {
2447 vp
->v_mpssdata
= NULL
;
2449 kmem_cache_free(vn_cache
, vp
);
2453 * vnode status changes, should define better states than 1, 0.
2456 vn_reclaim(vnode_t
*vp
)
2458 vfs_t
*vfsp
= vp
->v_vfsp
;
2461 vfsp
->vfs_implp
== NULL
|| vfsp
->vfs_femhead
== NULL
) {
2464 (void) VFS_VNSTATE(vfsp
, vp
, VNTRANS_RECLAIMED
);
2468 vn_idle(vnode_t
*vp
)
2470 vfs_t
*vfsp
= vp
->v_vfsp
;
2473 vfsp
->vfs_implp
== NULL
|| vfsp
->vfs_femhead
== NULL
) {
2476 (void) VFS_VNSTATE(vfsp
, vp
, VNTRANS_IDLED
);
2479 vn_exists(vnode_t
*vp
)
2481 vfs_t
*vfsp
= vp
->v_vfsp
;
2484 vfsp
->vfs_implp
== NULL
|| vfsp
->vfs_femhead
== NULL
) {
2487 (void) VFS_VNSTATE(vfsp
, vp
, VNTRANS_EXISTS
);
2491 vn_invalid(vnode_t
*vp
)
2493 vfs_t
*vfsp
= vp
->v_vfsp
;
2496 vfsp
->vfs_implp
== NULL
|| vfsp
->vfs_femhead
== NULL
) {
2499 (void) VFS_VNSTATE(vfsp
, vp
, VNTRANS_DESTROYED
);
2502 /* Vnode event notification */
2505 vnevent_support(vnode_t
*vp
, caller_context_t
*ct
)
2510 return (VOP_VNEVENT(vp
, VE_SUPPORT
, NULL
, NULL
, ct
));
2514 vnevent_rename_src(vnode_t
*vp
, vnode_t
*dvp
, char *name
, caller_context_t
*ct
)
2516 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2519 (void) VOP_VNEVENT(vp
, VE_RENAME_SRC
, dvp
, name
, ct
);
2523 vnevent_rename_dest(vnode_t
*vp
, vnode_t
*dvp
, char *name
,
2524 caller_context_t
*ct
)
2526 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2529 (void) VOP_VNEVENT(vp
, VE_RENAME_DEST
, dvp
, name
, ct
);
2533 vnevent_rename_dest_dir(vnode_t
*vp
, caller_context_t
*ct
)
2535 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2538 (void) VOP_VNEVENT(vp
, VE_RENAME_DEST_DIR
, NULL
, NULL
, ct
);
2542 vnevent_remove(vnode_t
*vp
, vnode_t
*dvp
, char *name
, caller_context_t
*ct
)
2544 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2547 (void) VOP_VNEVENT(vp
, VE_REMOVE
, dvp
, name
, ct
);
2551 vnevent_rmdir(vnode_t
*vp
, vnode_t
*dvp
, char *name
, caller_context_t
*ct
)
2553 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2556 (void) VOP_VNEVENT(vp
, VE_RMDIR
, dvp
, name
, ct
);
2560 vnevent_create(vnode_t
*vp
, caller_context_t
*ct
)
2562 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2565 (void) VOP_VNEVENT(vp
, VE_CREATE
, NULL
, NULL
, ct
);
2569 vnevent_link(vnode_t
*vp
, caller_context_t
*ct
)
2571 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2574 (void) VOP_VNEVENT(vp
, VE_LINK
, NULL
, NULL
, ct
);
2578 vnevent_mountedover(vnode_t
*vp
, caller_context_t
*ct
)
2580 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2583 (void) VOP_VNEVENT(vp
, VE_MOUNTEDOVER
, NULL
, NULL
, ct
);
2587 vnevent_truncate(vnode_t
*vp
, caller_context_t
*ct
)
2589 if (vp
== NULL
|| vp
->v_femhead
== NULL
) {
2592 (void) VOP_VNEVENT(vp
, VE_TRUNCATE
, NULL
, NULL
, ct
);
2600 vn_is_readonly(vnode_t
*vp
)
2602 return (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
);
2606 vn_has_flocks(vnode_t
*vp
)
2608 return (vp
->v_filocks
!= NULL
);
2612 vn_has_mandatory_locks(vnode_t
*vp
, int mode
)
2614 return ((vp
->v_filocks
!= NULL
) && (MANDLOCK(vp
, mode
)));
2618 vn_has_cached_data(vnode_t
*vp
)
2620 return (vp
->v_pages
!= NULL
);
2624 * Return 0 if the vnode in question shouldn't be permitted into a zone via
2628 vn_can_change_zones(vnode_t
*vp
)
2634 if (nfs_global_client_only
!= 0)
2638 * We always want to look at the underlying vnode if there is one.
2640 if (VOP_REALVP(vp
, &rvp
, NULL
) != 0)
2643 * Some pseudo filesystems (including doorfs) don't actually register
2644 * their vfsops_t, so the following may return NULL; we happily let
2645 * such vnodes switch zones.
2647 vswp
= vfs_getvfsswbyvfsops(vfs_getops(rvp
->v_vfsp
));
2649 if (vswp
->vsw_flag
& VSW_NOTZONESAFE
)
2651 vfs_unrefvfssw(vswp
);
2657 * Return nonzero if the vnode is a mount point, zero if not.
2660 vn_ismntpt(vnode_t
*vp
)
2662 return (vp
->v_vfsmountedhere
!= NULL
);
2665 /* Retrieve the vfs (if any) mounted on this vnode */
2667 vn_mountedvfs(vnode_t
*vp
)
2669 return (vp
->v_vfsmountedhere
);
2673 * Return nonzero if the vnode is referenced by the dnlc, zero if not.
2676 vn_in_dnlc(vnode_t
*vp
)
2678 return (vp
->v_count_dnlc
> 0);
2682 * vn_has_other_opens() checks whether a particular file is opened by more than
2683 * just the caller and whether the open is for read and/or write.
2684 * This routine is for calling after the caller has already called VOP_OPEN()
2685 * and the caller wishes to know if they are the only one with it open for
2686 * the mode(s) specified.
2688 * Vnode counts are only kept on regular files (v_type=VREG).
2700 if (vp
->v_wrcnt
> 1)
2704 if ((vp
->v_rdcnt
> 1) || (vp
->v_wrcnt
> 1))
2708 if ((vp
->v_rdcnt
> 1) && (vp
->v_wrcnt
> 1))
2712 if (vp
->v_rdcnt
> 1)
2721 * vn_is_opened() checks whether a particular file is opened and
2722 * whether the open is for read and/or write.
2724 * Vnode counts are only kept on regular files (v_type=VREG).
2740 if (vp
->v_rdcnt
&& vp
->v_wrcnt
)
2744 if (vp
->v_rdcnt
|| vp
->v_wrcnt
)
2757 * vn_is_mapped() checks whether a particular file is mapped and whether
2758 * the file is mapped read and/or write.
2771 * The atomic_add_64_nv functions force atomicity in the
2772 * case of 32 bit architectures. Otherwise the 64 bit values
2773 * require two fetches. The value of the fields may be
2774 * (potentially) changed between the first fetch and the
2778 if (atomic_add_64_nv((&(vp
->v_mmap_write
)), 0))
2782 if ((atomic_add_64_nv((&(vp
->v_mmap_read
)), 0)) &&
2783 (atomic_add_64_nv((&(vp
->v_mmap_write
)), 0)))
2787 if ((atomic_add_64_nv((&(vp
->v_mmap_read
)), 0)) ||
2788 (atomic_add_64_nv((&(vp
->v_mmap_write
)), 0)))
2792 if (atomic_add_64_nv((&(vp
->v_mmap_read
)), 0))
2799 if (vp
->v_mmap_write
)
2803 if (vp
->v_mmap_read
&& vp
->v_mmap_write
)
2807 if (vp
->v_mmap_read
|| vp
->v_mmap_write
)
2811 if (vp
->v_mmap_read
)
2821 * Set the operations vector for a vnode.
2823 * FEM ensures that the v_femhead pointer is filled in before the
2824 * v_op pointer is changed. This means that if the v_femhead pointer
2825 * is NULL, and the v_op field hasn't changed since before which checked
2826 * the v_femhead pointer; then our update is ok - we are not racing with
2830 vn_setops(vnode_t
*vp
, vnodeops_t
*vnodeops
)
2835 ASSERT(vnodeops
!= NULL
);
2840 * If vp->v_femhead == NULL, then we'll call casptr() to do the
2841 * compare-and-swap on vp->v_op. If either fails, then FEM is
2842 * in effect on the vnode and we need to have FEM deal with it.
2844 if (vp
->v_femhead
!= NULL
|| casptr(&vp
->v_op
, op
, vnodeops
) != op
) {
2845 fem_setvnops(vp
, vnodeops
);
2850 * Retrieve the operations vector for a vnode
2851 * As with vn_setops(above); make sure we aren't racing with FEM.
2852 * FEM sets the v_op to a special, internal, vnodeops that wouldn't
2853 * make sense to the callers of this routine.
2856 vn_getops(vnode_t
*vp
)
2864 if (vp
->v_femhead
== NULL
&& op
== vp
->v_op
) {
2867 return (fem_getvnops(vp
));
2872 * Returns non-zero (1) if the vnodeops matches that of the vnode.
2873 * Returns zero (0) if not.
2876 vn_matchops(vnode_t
*vp
, vnodeops_t
*vnodeops
)
2878 return (vn_getops(vp
) == vnodeops
);
2882 * Returns non-zero (1) if the specified operation matches the
2883 * corresponding operation for that the vnode.
2884 * Returns zero (0) if not.
2887 #define MATCHNAME(n1, n2) (((n1)[0] == (n2)[0]) && (strcmp((n1), (n2)) == 0))
2890 vn_matchopval(vnode_t
*vp
, char *vopname
, fs_generic_func_p funcp
)
2892 const fs_operation_trans_def_t
*otdp
;
2893 fs_generic_func_p
*loc
= NULL
;
2894 vnodeops_t
*vop
= vn_getops(vp
);
2896 ASSERT(vopname
!= NULL
);
2898 for (otdp
= vn_ops_table
; otdp
->name
!= NULL
; otdp
++) {
2899 if (MATCHNAME(otdp
->name
, vopname
)) {
2900 loc
= (fs_generic_func_p
*)
2901 ((char *)(vop
) + otdp
->offset
);
2906 return ((loc
!= NULL
) && (*loc
== funcp
));
2910 * fs_new_caller_id() needs to return a unique ID on a given local system.
2911 * The IDs do not need to survive across reboots. These are primarily
2912 * used so that (FEM) monitors can detect particular callers (such as
2913 * the NFS server) to a given vnode/vfs operation.
2918 static uint64_t next_caller_id
= 0LL; /* First call returns 1 */
2920 return ((u_longlong_t
)atomic_add_64_nv(&next_caller_id
, 1));
2924 * Given a starting vnode and a path, updates the path in the target vnode in
2925 * a safe manner. If the vnode already has path information embedded, then the
2926 * cached path is left untouched.
2929 size_t max_vnode_path
= 4 * MAXPATHLEN
;
2932 vn_setpath(vnode_t
*rootvp
, struct vnode
*startvp
, struct vnode
*vp
,
2933 const char *path
, size_t plen
)
2937 size_t rpathlen
, rpathalloc
;
2949 * We cannot grab base->v_lock while we hold vp->v_lock because of
2950 * the potential for deadlock.
2952 mutex_enter(&base
->v_lock
);
2953 if (base
->v_path
== NULL
) {
2954 mutex_exit(&base
->v_lock
);
2958 rpathlen
= strlen(base
->v_path
);
2959 rpathalloc
= rpathlen
+ plen
+ 1;
2960 /* Avoid adding a slash if there's already one there */
2961 if (base
->v_path
[rpathlen
-1] == '/')
2967 * We don't want to call kmem_alloc(KM_SLEEP) with kernel locks held,
2968 * so we must do this dance. If, by chance, something changes the path,
2969 * just give up since there is no real harm.
2971 mutex_exit(&base
->v_lock
);
2973 /* Paths should stay within reason */
2974 if (rpathalloc
> max_vnode_path
)
2977 rpath
= kmem_alloc(rpathalloc
, KM_SLEEP
);
2979 mutex_enter(&base
->v_lock
);
2980 if (base
->v_path
== NULL
|| strlen(base
->v_path
) != rpathlen
) {
2981 mutex_exit(&base
->v_lock
);
2982 kmem_free(rpath
, rpathalloc
);
2985 bcopy(base
->v_path
, rpath
, rpathlen
);
2986 mutex_exit(&base
->v_lock
);
2989 rpath
[rpathlen
++] = '/';
2990 bcopy(path
, rpath
+ rpathlen
, plen
);
2991 rpath
[rpathlen
+ plen
] = '\0';
2993 mutex_enter(&vp
->v_lock
);
2994 if (vp
->v_path
!= NULL
) {
2995 mutex_exit(&vp
->v_lock
);
2996 kmem_free(rpath
, rpathalloc
);
2999 mutex_exit(&vp
->v_lock
);
3004 * Sets the path to the vnode to be the given string, regardless of current
3005 * context. The string must be a complete path from rootdir. This is only used
3006 * by fsop_root() for setting the path based on the mountpoint.
3009 vn_setpath_str(struct vnode
*vp
, const char *str
, size_t len
)
3011 char *buf
= kmem_alloc(len
+ 1, KM_SLEEP
);
3013 mutex_enter(&vp
->v_lock
);
3014 if (vp
->v_path
!= NULL
) {
3015 mutex_exit(&vp
->v_lock
);
3016 kmem_free(buf
, len
+ 1);
3021 bcopy(str
, vp
->v_path
, len
);
3022 vp
->v_path
[len
] = '\0';
3024 mutex_exit(&vp
->v_lock
);
3028 * Called from within filesystem's vop_rename() to handle renames once the
3029 * target vnode is available.
3032 vn_renamepath(vnode_t
*dvp
, vnode_t
*vp
, const char *nm
, size_t len
)
3036 mutex_enter(&vp
->v_lock
);
3039 mutex_exit(&vp
->v_lock
);
3040 vn_setpath(rootdir
, dvp
, vp
, nm
, len
);
3042 kmem_free(tmp
, strlen(tmp
) + 1);
3046 * Similar to vn_setpath_str(), this function sets the path of the destination
3047 * vnode to the be the same as the source vnode.
3050 vn_copypath(struct vnode
*src
, struct vnode
*dst
)
3055 mutex_enter(&src
->v_lock
);
3056 if (src
->v_path
== NULL
) {
3057 mutex_exit(&src
->v_lock
);
3060 alloc
= strlen(src
->v_path
) + 1;
3062 /* avoid kmem_alloc() with lock held */
3063 mutex_exit(&src
->v_lock
);
3064 buf
= kmem_alloc(alloc
, KM_SLEEP
);
3065 mutex_enter(&src
->v_lock
);
3066 if (src
->v_path
== NULL
|| strlen(src
->v_path
) + 1 != alloc
) {
3067 mutex_exit(&src
->v_lock
);
3068 kmem_free(buf
, alloc
);
3071 bcopy(src
->v_path
, buf
, alloc
);
3072 mutex_exit(&src
->v_lock
);
3074 mutex_enter(&dst
->v_lock
);
3075 if (dst
->v_path
!= NULL
) {
3076 mutex_exit(&dst
->v_lock
);
3077 kmem_free(buf
, alloc
);
3081 mutex_exit(&dst
->v_lock
);
3085 * XXX Private interface for segvn routines that handle vnode
3086 * large page segments.
3088 * return 1 if vp's file system VOP_PAGEIO() implementation
3089 * can be safely used instead of VOP_GETPAGE() for handling
3090 * pagefaults against regular non swap files. VOP_PAGEIO()
3091 * interface is considered safe here if its implementation
3092 * is very close to VOP_GETPAGE() implementation.
3093 * e.g. It zero's out the part of the page beyond EOF. Doesn't
3094 * panic if there're file holes but instead returns an error.
3095 * Doesn't assume file won't be changed by user writes, etc.
3097 * return 0 otherwise.
3099 * For now allow segvn to only use VOP_PAGEIO() with ufs and nfs.
3102 vn_vmpss_usepageio(vnode_t
*vp
)
3104 vfs_t
*vfsp
= vp
->v_vfsp
;
3105 char *fsname
= vfssw
[vfsp
->vfs_fstype
].vsw_name
;
3106 char *pageio_ok_fss
[] = {"ufs", "nfs", NULL
};
3107 char **fsok
= pageio_ok_fss
;
3109 if (fsname
== NULL
) {
3113 for (; *fsok
; fsok
++) {
3114 if (strcmp(*fsok
, fsname
) == 0) {
3121 /* VOP_XXX() macros call the corresponding fop_xxx() function */
3128 caller_context_t
*ct
)
3135 * Adding to the vnode counts before calling open
3136 * avoids the need for a mutex. It circumvents a race
3137 * condition where a query made on the vnode counts results in a
3138 * false negative. The inquirer goes away believing the file is
3139 * not open when there is an open on the file already under way.
3141 * The counts are meant to prevent NFS from granting a delegation
3142 * when it would be dangerous to do so.
3144 * The vnode counts are only kept on regular files
3146 if ((*vpp
)->v_type
== VREG
) {
3148 atomic_add_32(&((*vpp
)->v_rdcnt
), 1);
3150 atomic_add_32(&((*vpp
)->v_wrcnt
), 1);
3153 VOPXID_MAP_CR(vp
, cr
);
3155 ret
= (*(*(vpp
))->v_op
->vop_open
)(vpp
, mode
, cr
, ct
);
3159 * Use the saved vp just in case the vnode ptr got trashed
3162 VOPSTATS_UPDATE(vp
, open
);
3163 if ((vp
->v_type
== VREG
) && (mode
& FREAD
))
3164 atomic_add_32(&(vp
->v_rdcnt
), -1);
3165 if ((vp
->v_type
== VREG
) && (mode
& FWRITE
))
3166 atomic_add_32(&(vp
->v_wrcnt
), -1);
3169 * Some filesystems will return a different vnode,
3170 * but the same path was still used to open it.
3171 * So if we do change the vnode and need to
3172 * copy over the path, do so here, rather than special
3173 * casing each filesystem. Adjust the vnode counts to
3174 * reflect the vnode switch.
3176 VOPSTATS_UPDATE(*vpp
, open
);
3177 if (*vpp
!= vp
&& *vpp
!= NULL
) {
3178 vn_copypath(vp
, *vpp
);
3179 if (((*vpp
)->v_type
== VREG
) && (mode
& FREAD
))
3180 atomic_add_32(&((*vpp
)->v_rdcnt
), 1);
3181 if ((vp
->v_type
== VREG
) && (mode
& FREAD
))
3182 atomic_add_32(&(vp
->v_rdcnt
), -1);
3183 if (((*vpp
)->v_type
== VREG
) && (mode
& FWRITE
))
3184 atomic_add_32(&((*vpp
)->v_wrcnt
), 1);
3185 if ((vp
->v_type
== VREG
) && (mode
& FWRITE
))
3186 atomic_add_32(&(vp
->v_wrcnt
), -1);
3200 caller_context_t
*ct
)
3204 VOPXID_MAP_CR(vp
, cr
);
3206 err
= (*(vp
)->v_op
->vop_close
)(vp
, flag
, count
, offset
, cr
, ct
);
3207 VOPSTATS_UPDATE(vp
, close
);
3209 * Check passed in count to handle possible dups. Vnode counts are only
3210 * kept on regular files
3212 if ((vp
->v_type
== VREG
) && (count
== 1)) {
3214 ASSERT(vp
->v_rdcnt
> 0);
3215 atomic_add_32(&(vp
->v_rdcnt
), -1);
3217 if (flag
& FWRITE
) {
3218 ASSERT(vp
->v_wrcnt
> 0);
3219 atomic_add_32(&(vp
->v_wrcnt
), -1);
3231 caller_context_t
*ct
)
3234 ssize_t resid_start
= uiop
->uio_resid
;
3236 VOPXID_MAP_CR(vp
, cr
);
3238 err
= (*(vp
)->v_op
->vop_read
)(vp
, uiop
, ioflag
, cr
, ct
);
3239 VOPSTATS_UPDATE_IO(vp
, read
,
3240 read_bytes
, (resid_start
- uiop
->uio_resid
));
3250 caller_context_t
*ct
)
3253 ssize_t resid_start
= uiop
->uio_resid
;
3255 VOPXID_MAP_CR(vp
, cr
);
3257 err
= (*(vp
)->v_op
->vop_write
)(vp
, uiop
, ioflag
, cr
, ct
);
3258 VOPSTATS_UPDATE_IO(vp
, write
,
3259 write_bytes
, (resid_start
- uiop
->uio_resid
));
3271 caller_context_t
*ct
)
3275 VOPXID_MAP_CR(vp
, cr
);
3277 err
= (*(vp
)->v_op
->vop_ioctl
)(vp
, cmd
, arg
, flag
, cr
, rvalp
, ct
);
3278 VOPSTATS_UPDATE(vp
, ioctl
);
3288 caller_context_t
*ct
)
3292 VOPXID_MAP_CR(vp
, cr
);
3294 err
= (*(vp
)->v_op
->vop_setfl
)(vp
, oflags
, nflags
, cr
, ct
);
3295 VOPSTATS_UPDATE(vp
, setfl
);
3305 caller_context_t
*ct
)
3309 VOPXID_MAP_CR(vp
, cr
);
3312 * If this file system doesn't understand the xvattr extensions
3313 * then turn off the xvattr bit.
3315 if (vfs_has_feature(vp
->v_vfsp
, VFSFT_XVATTR
) == 0) {
3316 vap
->va_mask
&= ~AT_XVATTR
;
3320 * We're only allowed to skip the ACL check iff we used a 32 bit
3321 * ACE mask with VOP_ACCESS() to determine permissions.
3323 if ((flags
& ATTR_NOACLCHECK
) &&
3324 vfs_has_feature(vp
->v_vfsp
, VFSFT_ACEMASKONACCESS
) == 0) {
3327 err
= (*(vp
)->v_op
->vop_getattr
)(vp
, vap
, flags
, cr
, ct
);
3328 VOPSTATS_UPDATE(vp
, getattr
);
3338 caller_context_t
*ct
)
3342 VOPXID_MAP_CR(vp
, cr
);
3345 * If this file system doesn't understand the xvattr extensions
3346 * then turn off the xvattr bit.
3348 if (vfs_has_feature(vp
->v_vfsp
, VFSFT_XVATTR
) == 0) {
3349 vap
->va_mask
&= ~AT_XVATTR
;
3353 * We're only allowed to skip the ACL check iff we used a 32 bit
3354 * ACE mask with VOP_ACCESS() to determine permissions.
3356 if ((flags
& ATTR_NOACLCHECK
) &&
3357 vfs_has_feature(vp
->v_vfsp
, VFSFT_ACEMASKONACCESS
) == 0) {
3360 err
= (*(vp
)->v_op
->vop_setattr
)(vp
, vap
, flags
, cr
, ct
);
3361 VOPSTATS_UPDATE(vp
, setattr
);
3371 caller_context_t
*ct
)
3375 if ((flags
& V_ACE_MASK
) &&
3376 vfs_has_feature(vp
->v_vfsp
, VFSFT_ACEMASKONACCESS
) == 0) {
3380 VOPXID_MAP_CR(vp
, cr
);
3382 err
= (*(vp
)->v_op
->vop_access
)(vp
, mode
, flags
, cr
, ct
);
3383 VOPSTATS_UPDATE(vp
, access
);
3396 caller_context_t
*ct
,
3397 int *deflags
, /* Returned per-dirent flags */
3398 pathname_t
*ppnp
) /* Returned case-preserved name in directory */
3403 * If this file system doesn't support case-insensitive access
3404 * and said access is requested, fail quickly. It is required
3405 * that if the vfs supports case-insensitive lookup, it also
3406 * supports extended dirent flags.
3408 if (flags
& FIGNORECASE
&&
3409 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3410 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3413 VOPXID_MAP_CR(dvp
, cr
);
3415 if ((flags
& LOOKUP_XATTR
) && (flags
& LOOKUP_HAVE_SYSATTR_DIR
) == 0) {
3416 ret
= xattr_dir_lookup(dvp
, vpp
, flags
, cr
);
3418 ret
= (*(dvp
)->v_op
->vop_lookup
)
3419 (dvp
, nm
, vpp
, pnp
, flags
, rdir
, cr
, ct
, deflags
, ppnp
);
3421 if (ret
== 0 && *vpp
) {
3422 VOPSTATS_UPDATE(*vpp
, lookup
);
3423 if ((*vpp
)->v_path
== NULL
) {
3424 vn_setpath(rootdir
, dvp
, *vpp
, nm
, strlen(nm
));
3441 caller_context_t
*ct
,
3442 vsecattr_t
*vsecp
) /* ACL to set during create */
3446 if (vsecp
!= NULL
&&
3447 vfs_has_feature(dvp
->v_vfsp
, VFSFT_ACLONCREATE
) == 0) {
3451 * If this file system doesn't support case-insensitive access
3452 * and said access is requested, fail quickly.
3454 if (flags
& FIGNORECASE
&&
3455 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3456 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3459 VOPXID_MAP_CR(dvp
, cr
);
3461 ret
= (*(dvp
)->v_op
->vop_create
)
3462 (dvp
, name
, vap
, excl
, mode
, vpp
, cr
, flags
, ct
, vsecp
);
3463 if (ret
== 0 && *vpp
) {
3464 VOPSTATS_UPDATE(*vpp
, create
);
3465 if ((*vpp
)->v_path
== NULL
) {
3466 vn_setpath(rootdir
, dvp
, *vpp
, name
, strlen(name
));
3478 caller_context_t
*ct
,
3484 * If this file system doesn't support case-insensitive access
3485 * and said access is requested, fail quickly.
3487 if (flags
& FIGNORECASE
&&
3488 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3489 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3492 VOPXID_MAP_CR(dvp
, cr
);
3494 err
= (*(dvp
)->v_op
->vop_remove
)(dvp
, nm
, cr
, ct
, flags
);
3495 VOPSTATS_UPDATE(dvp
, remove
);
3505 caller_context_t
*ct
,
3511 * If the target file system doesn't support case-insensitive access
3512 * and said access is requested, fail quickly.
3514 if (flags
& FIGNORECASE
&&
3515 (vfs_has_feature(tdvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3516 vfs_has_feature(tdvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3519 VOPXID_MAP_CR(tdvp
, cr
);
3521 err
= (*(tdvp
)->v_op
->vop_link
)(tdvp
, svp
, tnm
, cr
, ct
, flags
);
3522 VOPSTATS_UPDATE(tdvp
, link
);
3533 caller_context_t
*ct
,
3539 * If the file system involved does not support
3540 * case-insensitive access and said access is requested, fail
3543 if (flags
& FIGNORECASE
&&
3544 ((vfs_has_feature(sdvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3545 vfs_has_feature(sdvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0)))
3548 VOPXID_MAP_CR(tdvp
, cr
);
3550 err
= (*(sdvp
)->v_op
->vop_rename
)(sdvp
, snm
, tdvp
, tnm
, cr
, ct
, flags
);
3551 VOPSTATS_UPDATE(sdvp
, rename
);
3562 caller_context_t
*ct
,
3564 vsecattr_t
*vsecp
) /* ACL to set during create */
3568 if (vsecp
!= NULL
&&
3569 vfs_has_feature(dvp
->v_vfsp
, VFSFT_ACLONCREATE
) == 0) {
3573 * If this file system doesn't support case-insensitive access
3574 * and said access is requested, fail quickly.
3576 if (flags
& FIGNORECASE
&&
3577 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3578 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3581 VOPXID_MAP_CR(dvp
, cr
);
3583 ret
= (*(dvp
)->v_op
->vop_mkdir
)
3584 (dvp
, dirname
, vap
, vpp
, cr
, ct
, flags
, vsecp
);
3585 if (ret
== 0 && *vpp
) {
3586 VOPSTATS_UPDATE(*vpp
, mkdir
);
3587 if ((*vpp
)->v_path
== NULL
) {
3588 vn_setpath(rootdir
, dvp
, *vpp
, dirname
,
3602 caller_context_t
*ct
,
3608 * If this file system doesn't support case-insensitive access
3609 * and said access is requested, fail quickly.
3611 if (flags
& FIGNORECASE
&&
3612 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3613 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3616 VOPXID_MAP_CR(dvp
, cr
);
3618 err
= (*(dvp
)->v_op
->vop_rmdir
)(dvp
, nm
, cdir
, cr
, ct
, flags
);
3619 VOPSTATS_UPDATE(dvp
, rmdir
);
3629 caller_context_t
*ct
,
3633 ssize_t resid_start
= uiop
->uio_resid
;
3636 * If this file system doesn't support retrieving directory
3637 * entry flags and said access is requested, fail quickly.
3639 if (flags
& V_RDDIR_ENTFLAGS
&&
3640 vfs_has_feature(vp
->v_vfsp
, VFSFT_DIRENTFLAGS
) == 0)
3643 VOPXID_MAP_CR(vp
, cr
);
3645 err
= (*(vp
)->v_op
->vop_readdir
)(vp
, uiop
, cr
, eofp
, ct
, flags
);
3646 VOPSTATS_UPDATE_IO(vp
, readdir
,
3647 readdir_bytes
, (resid_start
- uiop
->uio_resid
));
3658 caller_context_t
*ct
,
3665 * If this file system doesn't support case-insensitive access
3666 * and said access is requested, fail quickly.
3668 if (flags
& FIGNORECASE
&&
3669 (vfs_has_feature(dvp
->v_vfsp
, VFSFT_CASEINSENSITIVE
) == 0 &&
3670 vfs_has_feature(dvp
->v_vfsp
, VFSFT_NOCASESENSITIVE
) == 0))
3673 VOPXID_MAP_CR(dvp
, cr
);
3675 /* check for reparse point */
3676 if ((vfs_has_feature(dvp
->v_vfsp
, VFSFT_REPARSE
)) &&
3677 (strncmp(target
, FS_REPARSE_TAG_STR
,
3678 strlen(FS_REPARSE_TAG_STR
)) == 0)) {
3679 if (!fs_reparse_mark(target
, vap
, &xvattr
))
3680 vap
= (vattr_t
*)&xvattr
;
3683 err
= (*(dvp
)->v_op
->vop_symlink
)
3684 (dvp
, linkname
, vap
, target
, cr
, ct
, flags
);
3685 VOPSTATS_UPDATE(dvp
, symlink
);
3694 caller_context_t
*ct
)
3698 VOPXID_MAP_CR(vp
, cr
);
3700 err
= (*(vp
)->v_op
->vop_readlink
)(vp
, uiop
, cr
, ct
);
3701 VOPSTATS_UPDATE(vp
, readlink
);
3710 caller_context_t
*ct
)
3714 VOPXID_MAP_CR(vp
, cr
);
3716 err
= (*(vp
)->v_op
->vop_fsync
)(vp
, syncflag
, cr
, ct
);
3717 VOPSTATS_UPDATE(vp
, fsync
);
3725 caller_context_t
*ct
)
3727 /* Need to update stats before vop call since we may lose the vnode */
3728 VOPSTATS_UPDATE(vp
, inactive
);
3730 VOPXID_MAP_CR(vp
, cr
);
3732 (*(vp
)->v_op
->vop_inactive
)(vp
, cr
, ct
);
3739 caller_context_t
*ct
)
3743 err
= (*(vp
)->v_op
->vop_fid
)(vp
, fidp
, ct
);
3744 VOPSTATS_UPDATE(vp
, fid
);
3752 caller_context_t
*ct
)
3756 ret
= ((*(vp
)->v_op
->vop_rwlock
)(vp
, write_lock
, ct
));
3757 VOPSTATS_UPDATE(vp
, rwlock
);
3765 caller_context_t
*ct
)
3767 (*(vp
)->v_op
->vop_rwunlock
)(vp
, write_lock
, ct
);
3768 VOPSTATS_UPDATE(vp
, rwunlock
);
3776 caller_context_t
*ct
)
3780 err
= (*(vp
)->v_op
->vop_seek
)(vp
, ooff
, noffp
, ct
);
3781 VOPSTATS_UPDATE(vp
, seek
);
3789 caller_context_t
*ct
)
3793 err
= (*(vp1
)->v_op
->vop_cmp
)(vp1
, vp2
, ct
);
3794 VOPSTATS_UPDATE(vp1
, cmp
);
3805 struct flk_callback
*flk_cbp
,
3807 caller_context_t
*ct
)
3811 VOPXID_MAP_CR(vp
, cr
);
3813 err
= (*(vp
)->v_op
->vop_frlock
)
3814 (vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
);
3815 VOPSTATS_UPDATE(vp
, frlock
);
3827 caller_context_t
*ct
)
3831 VOPXID_MAP_CR(vp
, cr
);
3833 err
= (*(vp
)->v_op
->vop_space
)(vp
, cmd
, bfp
, flag
, offset
, cr
, ct
);
3834 VOPSTATS_UPDATE(vp
, space
);
3842 caller_context_t
*ct
)
3846 err
= (*(vp
)->v_op
->vop_realvp
)(vp
, vpp
, ct
);
3847 VOPSTATS_UPDATE(vp
, realvp
);
3863 caller_context_t
*ct
)
3867 VOPXID_MAP_CR(vp
, cr
);
3869 err
= (*(vp
)->v_op
->vop_getpage
)
3870 (vp
, off
, len
, protp
, plarr
, plsz
, seg
, addr
, rw
, cr
, ct
);
3871 VOPSTATS_UPDATE(vp
, getpage
);
3882 caller_context_t
*ct
)
3886 VOPXID_MAP_CR(vp
, cr
);
3888 err
= (*(vp
)->v_op
->vop_putpage
)(vp
, off
, len
, flags
, cr
, ct
);
3889 VOPSTATS_UPDATE(vp
, putpage
);
3904 caller_context_t
*ct
)
3908 VOPXID_MAP_CR(vp
, cr
);
3910 err
= (*(vp
)->v_op
->vop_map
)
3911 (vp
, off
, as
, addrp
, len
, prot
, maxprot
, flags
, cr
, ct
);
3912 VOPSTATS_UPDATE(vp
, map
);
3927 caller_context_t
*ct
)
3932 VOPXID_MAP_CR(vp
, cr
);
3934 error
= (*(vp
)->v_op
->vop_addmap
)
3935 (vp
, off
, as
, addr
, len
, prot
, maxprot
, flags
, cr
, ct
);
3937 if ((!error
) && (vp
->v_type
== VREG
)) {
3938 delta
= (u_longlong_t
)btopr(len
);
3940 * If file is declared MAP_PRIVATE, it can't be written back
3941 * even if open for write. Handle as read.
3943 if (flags
& MAP_PRIVATE
) {
3944 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
3948 * atomic_add_64 forces the fetch of a 64 bit value to
3949 * be atomic on 32 bit machines
3951 if (maxprot
& PROT_WRITE
)
3952 atomic_add_64((uint64_t *)(&(vp
->v_mmap_write
)),
3954 if (maxprot
& PROT_READ
)
3955 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
3957 if (maxprot
& PROT_EXEC
)
3958 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
3962 VOPSTATS_UPDATE(vp
, addmap
);
3977 caller_context_t
*ct
)
3982 VOPXID_MAP_CR(vp
, cr
);
3984 error
= (*(vp
)->v_op
->vop_delmap
)
3985 (vp
, off
, as
, addr
, len
, prot
, maxprot
, flags
, cr
, ct
);
3988 * NFS calls into delmap twice, the first time
3989 * it simply establishes a callback mechanism and returns EAGAIN
3990 * while the real work is being done upon the second invocation.
3991 * We have to detect this here and only decrement the counts upon
3992 * the second delmap request.
3994 if ((error
!= EAGAIN
) && (vp
->v_type
== VREG
)) {
3996 delta
= (u_longlong_t
)btopr(len
);
3998 if (flags
& MAP_PRIVATE
) {
3999 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
4003 * atomic_add_64 forces the fetch of a 64 bit value
4004 * to be atomic on 32 bit machines
4006 if (maxprot
& PROT_WRITE
)
4007 atomic_add_64((uint64_t *)(&(vp
->v_mmap_write
)),
4009 if (maxprot
& PROT_READ
)
4010 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
4012 if (maxprot
& PROT_EXEC
)
4013 atomic_add_64((uint64_t *)(&(vp
->v_mmap_read
)),
4017 VOPSTATS_UPDATE(vp
, delmap
);
4028 struct pollhead
**phpp
,
4029 caller_context_t
*ct
)
4033 err
= (*(vp
)->v_op
->vop_poll
)(vp
, events
, anyyet
, reventsp
, phpp
, ct
);
4034 VOPSTATS_UPDATE(vp
, poll
);
4044 caller_context_t
*ct
)
4048 /* ensure lbdn and dblks can be passed safely to bdev_dump */
4049 if ((lbdn
!= (daddr_t
)lbdn
) || (dblks
!= (int)dblks
))
4052 err
= (*(vp
)->v_op
->vop_dump
)(vp
, addr
, lbdn
, dblks
, ct
);
4053 VOPSTATS_UPDATE(vp
, dump
);
4063 caller_context_t
*ct
)
4067 VOPXID_MAP_CR(vp
, cr
);
4069 err
= (*(vp
)->v_op
->vop_pathconf
)(vp
, cmd
, valp
, cr
, ct
);
4070 VOPSTATS_UPDATE(vp
, pathconf
);
4082 caller_context_t
*ct
)
4086 VOPXID_MAP_CR(vp
, cr
);
4088 err
= (*(vp
)->v_op
->vop_pageio
)(vp
, pp
, io_off
, io_len
, flags
, cr
, ct
);
4089 VOPSTATS_UPDATE(vp
, pageio
);
4098 caller_context_t
*ct
)
4101 err
= (*(vp
)->v_op
->vop_dumpctl
)(vp
, action
, blkp
, ct
);
4102 VOPSTATS_UPDATE(vp
, dumpctl
);
4113 caller_context_t
*ct
)
4115 /* Must do stats first since it's possible to lose the vnode */
4116 VOPSTATS_UPDATE(vp
, dispose
);
4118 VOPXID_MAP_CR(vp
, cr
);
4120 (*(vp
)->v_op
->vop_dispose
)(vp
, pp
, flag
, dn
, cr
, ct
);
4129 caller_context_t
*ct
)
4133 VOPXID_MAP_CR(vp
, cr
);
4136 * We're only allowed to skip the ACL check iff we used a 32 bit
4137 * ACE mask with VOP_ACCESS() to determine permissions.
4139 if ((flag
& ATTR_NOACLCHECK
) &&
4140 vfs_has_feature(vp
->v_vfsp
, VFSFT_ACEMASKONACCESS
) == 0) {
4143 err
= (*(vp
)->v_op
->vop_setsecattr
) (vp
, vsap
, flag
, cr
, ct
);
4144 VOPSTATS_UPDATE(vp
, setsecattr
);
4154 caller_context_t
*ct
)
4159 * We're only allowed to skip the ACL check iff we used a 32 bit
4160 * ACE mask with VOP_ACCESS() to determine permissions.
4162 if ((flag
& ATTR_NOACLCHECK
) &&
4163 vfs_has_feature(vp
->v_vfsp
, VFSFT_ACEMASKONACCESS
) == 0) {
4167 VOPXID_MAP_CR(vp
, cr
);
4169 err
= (*(vp
)->v_op
->vop_getsecattr
) (vp
, vsap
, flag
, cr
, ct
);
4170 VOPSTATS_UPDATE(vp
, getsecattr
);
4178 struct shrlock
*shr
,
4181 caller_context_t
*ct
)
4185 VOPXID_MAP_CR(vp
, cr
);
4187 err
= (*(vp
)->v_op
->vop_shrlock
)(vp
, cmd
, shr
, flag
, cr
, ct
);
4188 VOPSTATS_UPDATE(vp
, shrlock
);
4193 fop_vnevent(vnode_t
*vp
, vnevent_t vnevent
, vnode_t
*dvp
, char *fnm
,
4194 caller_context_t
*ct
)
4198 err
= (*(vp
)->v_op
->vop_vnevent
)(vp
, vnevent
, dvp
, fnm
, ct
);
4199 VOPSTATS_UPDATE(vp
, vnevent
);
4204 fop_reqzcbuf(vnode_t
*vp
, enum uio_rw ioflag
, xuio_t
*uiop
, cred_t
*cr
,
4205 caller_context_t
*ct
)
4209 if (vfs_has_feature(vp
->v_vfsp
, VFSFT_ZEROCOPY_SUPPORTED
) == 0)
4211 err
= (*(vp
)->v_op
->vop_reqzcbuf
)(vp
, ioflag
, uiop
, cr
, ct
);
4212 VOPSTATS_UPDATE(vp
, reqzcbuf
);
4217 fop_retzcbuf(vnode_t
*vp
, xuio_t
*uiop
, cred_t
*cr
, caller_context_t
*ct
)
4221 if (vfs_has_feature(vp
->v_vfsp
, VFSFT_ZEROCOPY_SUPPORTED
) == 0)
4223 err
= (*(vp
)->v_op
->vop_retzcbuf
)(vp
, uiop
, cr
, ct
);
4224 VOPSTATS_UPDATE(vp
, retzcbuf
);
4229 * Default destructor
4230 * Needed because NULL destructor means that the key is unused
4234 vsd_defaultdestructor(void *value
)
4238 * Create a key (index into per vnode array)
4239 * Locks out vsd_create, vsd_destroy, and vsd_free
4240 * May allocate memory with lock held
4243 vsd_create(uint_t
*keyp
, void (*destructor
)(void *))
4249 * if key is allocated, do nothing
4251 mutex_enter(&vsd_lock
);
4253 mutex_exit(&vsd_lock
);
4257 * find an unused key
4259 if (destructor
== NULL
)
4260 destructor
= vsd_defaultdestructor
;
4262 for (i
= 0; i
< vsd_nkeys
; ++i
)
4263 if (vsd_destructor
[i
] == NULL
)
4267 * if no unused keys, increase the size of the destructor array
4269 if (i
== vsd_nkeys
) {
4270 if ((nkeys
= (vsd_nkeys
<< 1)) == 0)
4273 (void (**)(void *))vsd_realloc((void *)vsd_destructor
,
4274 (size_t)(vsd_nkeys
* sizeof (void (*)(void *))),
4275 (size_t)(nkeys
* sizeof (void (*)(void *))));
4280 * allocate the next available unused key
4282 vsd_destructor
[i
] = destructor
;
4285 /* create vsd_list, if it doesn't exist */
4286 if (vsd_list
== NULL
) {
4287 vsd_list
= kmem_alloc(sizeof (list_t
), KM_SLEEP
);
4288 list_create(vsd_list
, sizeof (struct vsd_node
),
4289 offsetof(struct vsd_node
, vs_nodes
));
4292 mutex_exit(&vsd_lock
);
4298 * Assumes that the caller is preventing vsd_set and vsd_get
4299 * Locks out vsd_create, vsd_destroy, and vsd_free
4300 * May free memory with lock held
4303 vsd_destroy(uint_t
*keyp
)
4306 struct vsd_node
*vsd
;
4309 * protect the key namespace and our destructor lists
4311 mutex_enter(&vsd_lock
);
4315 ASSERT(key
<= vsd_nkeys
);
4318 * if the key is valid
4323 * for every vnode with VSD, call key's destructor
4325 for (vsd
= list_head(vsd_list
); vsd
!= NULL
;
4326 vsd
= list_next(vsd_list
, vsd
)) {
4328 * no VSD for key in this vnode
4330 if (key
> vsd
->vs_nkeys
)
4333 * call destructor for key
4335 if (vsd
->vs_value
[k
] && vsd_destructor
[k
])
4336 (*vsd_destructor
[k
])(vsd
->vs_value
[k
]);
4338 * reset value for key
4340 vsd
->vs_value
[k
] = NULL
;
4343 * actually free the key (NULL destructor == unused)
4345 vsd_destructor
[k
] = NULL
;
4348 mutex_exit(&vsd_lock
);
4352 * Quickly return the per vnode value that was stored with the specified key
4353 * Assumes the caller is protecting key from vsd_create and vsd_destroy
4354 * Assumes the caller is holding v_vsd_lock to protect the vsd.
4357 vsd_get(vnode_t
*vp
, uint_t key
)
4359 struct vsd_node
*vsd
;
4362 ASSERT(mutex_owned(&vp
->v_vsd_lock
));
4366 if (key
&& vsd
!= NULL
&& key
<= vsd
->vs_nkeys
)
4367 return (vsd
->vs_value
[key
- 1]);
4372 * Set a per vnode value indexed with the specified key
4373 * Assumes the caller is holding v_vsd_lock to protect the vsd.
4376 vsd_set(vnode_t
*vp
, uint_t key
, void *value
)
4378 struct vsd_node
*vsd
;
4381 ASSERT(mutex_owned(&vp
->v_vsd_lock
));
4388 vsd
= vp
->v_vsd
= kmem_zalloc(sizeof (*vsd
), KM_SLEEP
);
4391 * If the vsd was just allocated, vs_nkeys will be 0, so the following
4392 * code won't happen and we will continue down and allocate space for
4393 * the vs_value array.
4394 * If the caller is replacing one value with another, then it is up
4395 * to the caller to free/rele/destroy the previous value (if needed).
4397 if (key
<= vsd
->vs_nkeys
) {
4398 vsd
->vs_value
[key
- 1] = value
;
4402 ASSERT(key
<= vsd_nkeys
);
4404 if (vsd
->vs_nkeys
== 0) {
4405 mutex_enter(&vsd_lock
); /* lock out vsd_destroy() */
4407 * Link onto list of all VSD nodes.
4409 list_insert_head(vsd_list
, vsd
);
4410 mutex_exit(&vsd_lock
);
4414 * Allocate vnode local storage and set the value for key
4416 vsd
->vs_value
= vsd_realloc(vsd
->vs_value
,
4417 vsd
->vs_nkeys
* sizeof (void *),
4418 key
* sizeof (void *));
4419 vsd
->vs_nkeys
= key
;
4420 vsd
->vs_value
[key
- 1] = value
;
4426 * Called from vn_free() to run the destructor function for each vsd
4427 * Locks out vsd_create and vsd_destroy
4428 * Assumes that the destructor *DOES NOT* use vsd
4431 vsd_free(vnode_t
*vp
)
4434 struct vsd_node
*vsd
= vp
->v_vsd
;
4439 if (vsd
->vs_nkeys
== 0) {
4440 kmem_free(vsd
, sizeof (*vsd
));
4446 * lock out vsd_create and vsd_destroy, call
4447 * the destructor, and mark the value as destroyed.
4449 mutex_enter(&vsd_lock
);
4451 for (i
= 0; i
< vsd
->vs_nkeys
; i
++) {
4452 if (vsd
->vs_value
[i
] && vsd_destructor
[i
])
4453 (*vsd_destructor
[i
])(vsd
->vs_value
[i
]);
4454 vsd
->vs_value
[i
] = NULL
;
4458 * remove from linked list of VSD nodes
4460 list_remove(vsd_list
, vsd
);
4462 mutex_exit(&vsd_lock
);
4467 kmem_free(vsd
->vs_value
, vsd
->vs_nkeys
* sizeof (void *));
4468 kmem_free(vsd
, sizeof (struct vsd_node
));
4476 vsd_realloc(void *old
, size_t osize
, size_t nsize
)
4480 new = kmem_zalloc(nsize
, KM_SLEEP
);
4482 bcopy(old
, new, osize
);
4483 kmem_free(old
, osize
);
4489 * Setup the extensible system attribute for creating a reparse point.
4490 * The symlink data 'target' is validated for proper format of a reparse
4491 * string and a check also made to make sure the symlink data does not
4492 * point to an existing file.
4494 * return 0 if ok else -1.
4497 fs_reparse_mark(char *target
, vattr_t
*vap
, xvattr_t
*xvattr
)
4501 if ((!target
) || (!vap
) || (!xvattr
))
4504 /* validate reparse string */
4505 if (reparse_validate((const char *)target
))
4509 xvattr
->xva_vattr
= *vap
;
4510 xvattr
->xva_vattr
.va_mask
|= AT_XVATTR
;
4511 xoap
= xva_getxoptattr(xvattr
);
4513 XVA_SET_REQ(xvattr
, XAT_REPARSE
);
4514 xoap
->xoa_reparse
= 1;
4520 * Function to check whether a symlink is a reparse point.
4521 * Return B_TRUE if it is a reparse point, else return B_FALSE
4524 vn_is_reparse(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4529 if ((vp
->v_type
!= VLNK
) ||
4530 !(vfs_has_feature(vp
->v_vfsp
, VFSFT_XVATTR
)))
4534 xoap
= xva_getxoptattr(&xvattr
);
4536 XVA_SET_REQ(&xvattr
, XAT_REPARSE
);
4538 if (VOP_GETATTR(vp
, &xvattr
.xva_vattr
, 0, cr
, ct
))
4541 if ((!(xvattr
.xva_vattr
.va_mask
& AT_XVATTR
)) ||
4542 (!(XVA_ISSET_RTN(&xvattr
, XAT_REPARSE
))))
4545 return (xoap
->xoa_reparse
? B_TRUE
: B_FALSE
);