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 2015 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright 2016 Joyent, Inc.
28 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
29 /* All Rights Reserved */
32 * University Copyright- Copyright (c) 1982, 1986, 1988
33 * The Regents of the University of California
36 * University Acknowledgment- Portions of this document are derived from
37 * software developed by the University of California, Berkeley, and its
41 #include <sys/types.h>
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/cpuvar.h>
45 #include <sys/errno.h>
50 #include <sys/vnode.h>
51 #include <sys/pathname.h>
53 #include <sys/vtrace.h>
54 #include <sys/sysmacros.h>
55 #include <sys/debug.h>
56 #include <sys/dirent.h>
60 #include <sys/fs/snode.h>
62 /* Controls whether paths are stored with vnodes. */
63 int vfs_vnode_path
= 1;
73 return (lookupnameatcred(fnamep
, seg
, followlink
, dirvpp
, compvpp
, NULL
,
78 * Lookup the user file name,
79 * Handle allocation and freeing of pathname buffer, return error.
83 char *fnamep
, /* user pathname */
84 enum uio_seg seg
, /* addr space that name is in */
85 int followlink
, /* follow sym links */
86 vnode_t
**dirvpp
, /* ret for ptr to parent dir vnode */
87 vnode_t
**compvpp
, /* ret for ptr to component vnode */
88 vnode_t
*startvp
, /* start path search from vp */
89 cred_t
*cr
) /* credential */
91 char namebuf
[TYPICALMAXPATHLEN
];
92 struct pathname lookpn
;
95 error
= pn_get_buf(fnamep
, seg
, &lookpn
, namebuf
, sizeof (namebuf
));
97 error
= lookuppnatcred(&lookpn
, NULL
, followlink
,
98 dirvpp
, compvpp
, startvp
, cr
);
100 if (error
== ENAMETOOLONG
) {
102 * This thread used a pathname > TYPICALMAXPATHLEN bytes long.
104 if (error
= pn_get(fnamep
, seg
, &lookpn
))
106 error
= lookuppnatcred(&lookpn
, NULL
, followlink
,
107 dirvpp
, compvpp
, startvp
, cr
);
115 lookupnameat(char *fnamep
, enum uio_seg seg
, int followlink
,
116 vnode_t
**dirvpp
, vnode_t
**compvpp
, vnode_t
*startvp
)
118 return (lookupnameatcred(fnamep
, seg
, followlink
, dirvpp
, compvpp
,
124 struct pathname
*pnp
,
125 struct pathname
*rpnp
,
130 return (lookuppnatcred(pnp
, rpnp
, followlink
, dirvpp
, compvpp
, NULL
,
135 * Lookup the user file name from a given vp, using a specific credential.
139 struct pathname
*pnp
, /* pathname to lookup */
140 struct pathname
*rpnp
, /* if non-NULL, return resolved path */
141 int followlink
, /* (don't) follow sym links */
142 vnode_t
**dirvpp
, /* ptr for parent vnode */
143 vnode_t
**compvpp
, /* ptr for entry vnode */
144 vnode_t
*startvp
, /* start search from this vp */
145 cred_t
*cr
) /* user credential */
147 vnode_t
*vp
; /* current directory vp */
151 if (pnp
->pn_pathlen
== 0)
154 mutex_enter(&p
->p_lock
); /* for u_rdir and u_cdir */
155 if ((rootvp
= PTOU(p
)->u_rdir
) == NULL
)
157 else if (rootvp
!= rootdir
) /* no need to VN_HOLD rootdir */
160 if (pnp
->pn_path
[0] == '/') {
163 vp
= (startvp
== NULL
) ? PTOU(p
)->u_cdir
: startvp
;
166 mutex_exit(&p
->p_lock
);
169 * Skip over leading slashes
171 if (pnp
->pn_path
[0] == '/') {
175 } while (pnp
->pn_path
[0] == '/');
178 return (lookuppnvp(pnp
, rpnp
, followlink
, dirvpp
,
179 compvpp
, rootvp
, vp
, cr
));
183 lookuppnat(struct pathname
*pnp
, struct pathname
*rpnp
,
184 int followlink
, vnode_t
**dirvpp
, vnode_t
**compvpp
,
187 return (lookuppnatcred(pnp
, rpnp
, followlink
, dirvpp
, compvpp
, startvp
,
191 /* Private flag to do our getcwd() dirty work */
192 #define LOOKUP_CHECKREAD 0x10
193 #define LOOKUP_MASK (~LOOKUP_CHECKREAD)
196 * Starting at current directory, translate pathname pnp to end.
197 * Leave pathname of final component in pnp, return the vnode
198 * for the final component in *compvpp, and return the vnode
199 * for the parent of the final component in dirvpp.
201 * This is the central routine in pathname translation and handles
202 * multiple components in pathnames, separating them at /'s. It also
203 * implements mounted file systems and processes symbolic links.
205 * vp is the vnode where the directory search should start.
207 * Reference counts: vp must be held prior to calling this function. rootvp
208 * should only be held if rootvp != rootdir.
212 struct pathname
*pnp
, /* pathname to lookup */
213 struct pathname
*rpnp
, /* if non-NULL, return resolved path */
214 int flags
, /* follow symlinks */
215 vnode_t
**dirvpp
, /* ptr for parent vnode */
216 vnode_t
**compvpp
, /* ptr for entry vnode */
217 vnode_t
*rootvp
, /* rootvp */
218 vnode_t
*vp
, /* directory to start search at */
219 cred_t
*cr
) /* user's credential */
221 vnode_t
*cvp
; /* current component vp */
222 char component
[MAXNAMELEN
]; /* buffer for component (incl null) */
226 struct pathname presrvd
; /* case preserved name */
227 struct pathname
*pp
= NULL
;
229 vnode_t
*zonevp
= curproc
->p_zone
->zone_rootvp
; /* zone root */
230 int must_be_directory
= 0;
231 boolean_t retry_with_kcred
;
232 uint32_t auditing
= AU_AUDITING();
234 CPU_STATS_ADDQ(CPU
, sys
, namei
, 1);
238 rpnp
->pn_pathlen
= 0;
240 lookup_flags
= dirvpp
? LOOKUP_DIR
: 0;
241 if (flags
& FIGNORECASE
) {
242 lookup_flags
|= FIGNORECASE
;
248 audit_anchorpath(pnp
, vp
== rootvp
);
251 * Eliminate any trailing slashes in the pathname.
252 * If there are any, we must follow all symlinks.
253 * Also, we must guarantee that the last component is a directory.
255 if (pn_fixslash(pnp
)) {
257 must_be_directory
= 1;
262 retry_with_kcred
= B_FALSE
;
265 * Make sure we have a directory.
267 if (vp
->v_type
!= VDIR
) {
272 if (rpnp
&& VN_CMP(vp
, rootvp
))
273 (void) pn_set(rpnp
, "/");
276 * Process the next component of the pathname.
278 if (error
= pn_getcomponent(pnp
, component
)) {
283 * Handle "..": two special cases.
284 * 1. If we're at the root directory (e.g. after chroot or
285 * zone_enter) then change ".." to "." so we can't get
286 * out of this subtree.
287 * 2. If this vnode is the root of a mounted file system,
288 * then replace it with the vnode that was mounted on
289 * so that we take the ".." in the other file system.
291 if (component
[0] == '.' && component
[1] == '.' && component
[2] == 0) {
293 if (VN_CMP(vp
, rootvp
) || VN_CMP(vp
, zonevp
)) {
295 } else if (vp
->v_flag
& VROOT
) {
300 * While we deal with the vfs pointer from the vnode
301 * the filesystem could have been forcefully unmounted
302 * and the vnode's v_vfsp could have been invalidated
303 * by VFS_UNMOUNT. Hence, we cache v_vfsp and use it
304 * with vfs_rlock_wait/vfs_unlock.
305 * It is safe to use the v_vfsp even it is freed by
306 * VFS_UNMOUNT because vfs_rlock_wait/vfs_unlock
307 * do not dereference v_vfsp. It is just used as a
309 * One more corner case here is the memory getting
310 * reused for another vfs structure. In this case
311 * lookuppnvp's vfs_rlock_wait will succeed, domount's
312 * vfs_lock will fail and domount will bail out with an
318 * This lock is used to synchronize
319 * mounts/unmounts and lookups.
320 * Threads doing mounts/unmounts hold the
321 * writers version vfs_lock_wait().
324 vfs_rlock_wait(vfsp
);
327 * If this vnode is on a file system that
328 * has been forcibly unmounted,
329 * we can't proceed. Cancel this operation
332 * vfs_vnodecovered is NULL if unmounted.
333 * Currently, nfs uses VFS_UNMOUNTED to
334 * check if it's a forced-umount. Keep the
335 * same checking here as well even though it
338 if (((vp
= cvp
->v_vfsp
->vfs_vnodecovered
) == NULL
) ||
339 (cvp
->v_vfsp
->vfs_flag
& VFS_UNMOUNTED
)) {
351 * Crossing mount points. For eg: We are doing
352 * a lookup of ".." for file systems root vnode
353 * mounted here, and fop_lookup() (with covered vnode)
354 * will be on underlying file systems mount point
355 * vnode. Set retry_with_kcred flag as we might end
356 * up doing fop_lookup() with kcred if required.
358 retry_with_kcred
= B_TRUE
;
364 * LOOKUP_CHECKREAD is a private flag used by vnodetopath() to indicate
365 * that we need to have read permission on every directory in the entire
366 * path. This is used to ensure that a forward-lookup of a cached value
367 * has the same effect as a reverse-lookup when the cached value cannot
370 if ((flags
& LOOKUP_CHECKREAD
) &&
371 (error
= fop_access(vp
, VREAD
, 0, cr
, NULL
)) != 0)
375 * Perform a lookup in the current directory.
377 error
= fop_lookup(vp
, component
, &cvp
, pnp
, lookup_flags
,
378 rootvp
, cr
, NULL
, NULL
, pp
);
381 * Retry with kcred - If crossing mount points & error is EACCES.
383 * If we are crossing mount points here and doing ".." lookup,
384 * fop_lookup() might fail if the underlying file systems
385 * mount point has no execute permission. In cases like these,
386 * we retry fop_lookup() by giving as much privilage as possible
387 * by passing kcred credentials.
389 * In case of hierarchical file systems, passing kcred still may
391 * For eg: UFS FS --> Mount NFS FS --> Again mount UFS on some
392 * directory inside NFS FS.
394 if ((error
== EACCES
) && retry_with_kcred
)
395 error
= fop_lookup(vp
, component
, &cvp
, pnp
, lookup_flags
,
396 rootvp
, zone_kcred(), NULL
, NULL
, pp
);
401 * On error, return hard error if
402 * (a) we're not at the end of the pathname yet, or
403 * (b) the caller didn't want the parent directory, or
404 * (c) we failed for some reason other than a missing entry.
406 if (pn_pathleft(pnp
) || dirvpp
== NULL
|| error
!= ENOENT
)
408 if (auditing
) { /* directory access */
409 if (error
= audit_savepath(pnp
, vp
, vp
, error
, cr
))
415 * We inform the caller that the desired entry must be
416 * a directory by adding a '/' to the component name.
418 if (must_be_directory
&& (error
= pn_addslash(pnp
)) != 0)
423 if (rootvp
!= rootdir
)
431 * Traverse mount points.
432 * XXX why don't we need to hold a read lock here (call vn_vfsrlock)?
433 * What prevents a concurrent update to v_vfsmountedhere?
434 * Possible answer: if mounting, we might not see the mount
435 * if it is concurrently coming into existence, but that's
436 * really not much different from the thread running a bit slower.
437 * If unmounting, we may get into traverse() when we shouldn't,
438 * but traverse() will catch this case for us.
439 * (For this to work, fetching v_vfsmountedhere had better
442 if (vn_mountedvfs(cvp
) != NULL
) {
443 if ((error
= traverse(&cvp
)) != 0)
448 * If we hit a symbolic link and there is more path to be
449 * translated or this operation does not wish to apply
450 * to a link, then place the contents of the link at the
451 * front of the remaining pathname.
453 if (cvp
->v_type
== VLNK
&& ((flags
& FOLLOW
) || pn_pathleft(pnp
))) {
454 struct pathname linkpath
;
456 if (++nlink
> MAXSYMLINKS
) {
461 if (error
= pn_getsymlink(cvp
, &linkpath
, cr
)) {
467 audit_symlink(pnp
, &linkpath
);
469 if (pn_pathleft(&linkpath
) == 0)
470 (void) pn_set(&linkpath
, ".");
471 error
= pn_insert(pnp
, &linkpath
, strlen(component
));
477 if (pnp
->pn_pathlen
== 0) {
481 if (pnp
->pn_path
[0] == '/') {
485 } while (pnp
->pn_path
[0] == '/');
491 audit_anchorpath(pnp
, vp
== rootvp
);
492 if (pn_fixslash(pnp
)) {
494 must_be_directory
= 1;
500 * If rpnp is non-NULL, remember the resolved path name therein.
501 * Do not include "." components. Collapse occurrences of
502 * "previous/..", so long as "previous" is not itself "..".
503 * Exhausting rpnp results in error ENAMETOOLONG.
505 if (rpnp
&& strcmp(component
, ".") != 0) {
508 if (strcmp(component
, "..") == 0 &&
509 rpnp
->pn_pathlen
!= 0 &&
510 !((rpnp
->pn_pathlen
> 2 &&
511 strncmp(rpnp
->pn_path
+rpnp
->pn_pathlen
-3, "/..", 3) == 0) ||
512 (rpnp
->pn_pathlen
== 2 &&
513 strncmp(rpnp
->pn_path
, "..", 2) == 0))) {
514 while (rpnp
->pn_pathlen
&&
515 rpnp
->pn_path
[rpnp
->pn_pathlen
-1] != '/')
517 if (rpnp
->pn_pathlen
> 1)
519 rpnp
->pn_path
[rpnp
->pn_pathlen
] = '\0';
521 if (rpnp
->pn_pathlen
!= 0 &&
522 rpnp
->pn_path
[rpnp
->pn_pathlen
-1] != '/')
523 rpnp
->pn_path
[rpnp
->pn_pathlen
++] = '/';
524 if (flags
& FIGNORECASE
) {
526 * Return the case-preserved name
527 * within the resolved path.
529 error
= copystr(pp
->pn_buf
,
530 rpnp
->pn_path
+ rpnp
->pn_pathlen
,
531 rpnp
->pn_bufsize
- rpnp
->pn_pathlen
, &len
);
533 error
= copystr(component
,
534 rpnp
->pn_path
+ rpnp
->pn_pathlen
,
535 rpnp
->pn_bufsize
- rpnp
->pn_pathlen
, &len
);
537 if (error
) /* copystr() returns ENAMETOOLONG */
539 rpnp
->pn_pathlen
+= (len
- 1);
540 ASSERT(rpnp
->pn_bufsize
> rpnp
->pn_pathlen
);
545 * If no more components, return last directory (if wanted) and
546 * last component (if wanted).
548 if (pn_pathleft(pnp
) == 0) {
550 * If there was a trailing slash in the pathname,
551 * make sure the last component is a directory.
553 if (must_be_directory
&& cvp
->v_type
!= VDIR
) {
557 if (dirvpp
!= NULL
) {
559 * Check that we have the real parent and not
560 * an alias of the last component.
562 if (vn_compare(vp
, cvp
)) {
564 (void) audit_savepath(pnp
, cvp
, vp
,
569 if (rootvp
!= rootdir
)
579 (void) audit_savepath(pnp
, cvp
, vp
, 0, cr
);
580 if (pnp
->pn_path
== pnp
->pn_buf
)
581 (void) pn_set(pnp
, ".");
585 if (VN_CMP(cvp
, rootvp
))
586 (void) pn_set(rpnp
, "/");
587 else if (rpnp
->pn_pathlen
== 0)
588 (void) pn_set(rpnp
, ".");
595 if (rootvp
!= rootdir
)
603 * Skip over slashes from end of last component.
605 while (pnp
->pn_path
[0] == '/') {
611 * Searched through another level of directory:
612 * release previous directory handle and save new (result
613 * of lookup) as current directory.
621 if (auditing
) /* reached end of path */
622 (void) audit_savepath(pnp
, cvp
, vp
, error
, cr
);
625 * Error. Release vnodes and return.
630 * If the error was ESTALE and the current directory to look in
631 * was the root for this lookup, the root for a mounted file
632 * system, or the starting directory for lookups, then
633 * return ENOENT instead of ESTALE. In this case, no recovery
634 * is possible by the higher level. If ESTALE was returned for
635 * some intermediate directory along the path, then recovery
636 * is potentially possible and retrying from the higher level
637 * will either correct the situation by purging stale cache
638 * entries or eventually get back to the point where no recovery
641 if (error
== ESTALE
&&
642 (VN_CMP(vp
, rootvp
) || (vp
->v_flag
& VROOT
) || vp
== startvp
))
645 if (rootvp
!= rootdir
)
653 * Traverse a mount point. Routine accepts a vnode pointer as a reference
654 * parameter and performs the indirection, releasing the original vnode.
657 traverse(vnode_t
**cvpp
)
667 * If this vnode is mounted on, then we transparently indirect
668 * to the vnode which is the root of the mounted file system.
669 * Before we do this we must check that an unmount is not in
670 * progress on this vnode.
675 * Try to read lock the vnode. If this fails because
676 * the vnode is already write locked, then check to
677 * see whether it is the current thread which locked
678 * the vnode. If it is not, then read lock the vnode
679 * by waiting to acquire the lock.
681 * The code path in domount() is an example of support
682 * which needs to look up two pathnames and locks one
683 * of them in between the two lookups.
685 error
= vn_vfsrlock(cvp
);
687 if (!vn_vfswlock_held(cvp
))
688 error
= vn_vfsrlock_wait(cvp
);
691 * lookuppn() expects a held vnode to be
692 * returned because it promptly calls
693 * VN_RELE after the error return
701 * Reached the end of the mount chain?
703 vfsp
= vn_mountedvfs(cvp
);
710 * The read lock must be held across the call to VFS_ROOT() to
711 * prevent a concurrent unmount from destroying the vfs.
713 error
= VFS_ROOT(vfsp
, &tvp
);
729 * Return the lowermost vnode if this is a mountpoint.
732 vn_under(vnode_t
*vp
)
737 while (vp
->v_flag
& VROOT
) {
740 vfs_rlock_wait(vfsp
);
741 if ((uvp
= vfsp
->vfs_vnodecovered
) == NULL
||
742 (vfsp
->vfs_flag
& VFS_UNMOUNTED
)) {
756 vnode_match(vnode_t
*v1
, vnode_t
*v2
, cred_t
*cr
)
758 vattr_t v1attr
, v2attr
;
761 * If we have a device file, check to see if is a cloned open of the
762 * same device. For self-cloning devices, the major numbers will match.
763 * For devices cloned through the 'clone' driver, the minor number of
764 * the source device will be the same as the major number of the cloned
767 if ((v1
->v_type
== VCHR
|| v1
->v_type
== VBLK
) &&
768 v1
->v_type
== v2
->v_type
) {
769 if ((spec_is_selfclone(v1
) || spec_is_selfclone(v2
)) &&
770 getmajor(v1
->v_rdev
) == getmajor(v2
->v_rdev
))
773 if (spec_is_clone(v1
) &&
774 getmajor(v1
->v_rdev
) == getminor(v2
->v_rdev
))
777 if (spec_is_clone(v2
) &&
778 getmajor(v2
->v_rdev
) == getminor(v1
->v_rdev
))
782 v1attr
.va_mask
= v2attr
.va_mask
= AT_TYPE
;
785 * This check for symbolic links handles the pseudo-symlinks in procfs.
786 * These particular links have v_type of VDIR, but the attributes have a
787 * type of VLNK. We need to avoid these links because otherwise if we
788 * are currently in '/proc/self/fd', then '/proc/self/cwd' will compare
791 if (fop_getattr(v1
, &v1attr
, 0, cr
, NULL
) != 0 ||
792 fop_getattr(v2
, &v2attr
, 0, cr
, NULL
) != 0 ||
793 v1attr
.va_type
== VLNK
|| v2attr
.va_type
== VLNK
)
796 v1attr
.va_mask
= v2attr
.va_mask
= AT_TYPE
| AT_FSID
| AT_NODEID
;
798 if (fop_getattr(v1
, &v1attr
, ATTR_REAL
, cr
, NULL
) != 0 ||
799 fop_getattr(v2
, &v2attr
, ATTR_REAL
, cr
, NULL
) != 0)
802 return (v1attr
.va_fsid
== v2attr
.va_fsid
&&
803 v1attr
.va_nodeid
== v2attr
.va_nodeid
);
808 * Find the entry in the directory corresponding to the target vnode.
811 dirfindvp(vnode_t
*vrootp
, vnode_t
*dvp
, vnode_t
*tvp
, cred_t
*cr
, char *dbuf
,
812 size_t dlen
, dirent64_t
**rdp
)
823 ASSERT(dvp
->v_type
== VDIR
);
826 * This is necessary because of the strange semantics of fop_lookup().
828 bzero(&pnp
, sizeof (pnp
));
834 uio
.uio_segflg
= UIO_SYSSPACE
;
836 uio
.uio_extflg
= UIO_COPY_CACHED
;
839 if ((error
= fop_access(dvp
, VREAD
, 0, cr
, NULL
)) != 0)
843 uio
.uio_resid
= dlen
;
847 (void) fop_rwlock(dvp
, V_WRITELOCK_FALSE
, NULL
);
848 error
= fop_readdir(dvp
, &uio
, cr
, &eof
, NULL
, 0);
849 fop_rwunlock(dvp
, V_WRITELOCK_FALSE
, NULL
);
851 dbuflen
= dlen
- uio
.uio_resid
;
853 if (error
|| dbuflen
== 0)
856 dp
= (dirent64_t
*)dbuf
;
857 while ((intptr_t)dp
< (intptr_t)dbuf
+ dbuflen
) {
859 * Ignore '.' and '..' entries
861 if (strcmp(dp
->d_name
, ".") == 0 ||
862 strcmp(dp
->d_name
, "..") == 0) {
863 dp
= (dirent64_t
*)((intptr_t)dp
+
868 error
= fop_lookup(dvp
, dp
->d_name
, &cmpvp
, &pnp
, 0,
869 vrootp
, cr
, NULL
, NULL
, NULL
);
872 * We only want to bail out if there was an error other
873 * than ENOENT. Otherwise, it could be that someone
874 * just removed an entry since the readdir() call, and
875 * the entry we want is further on in the directory.
878 if (vnode_match(tvp
, cmpvp
, cr
)) {
885 } else if (error
!= ENOENT
) {
889 dp
= (dirent64_t
*)((intptr_t)dp
+ dp
->d_reclen
);
894 * Something strange has happened, this directory does not contain the
895 * specified vnode. This should never happen in the normal case, since
896 * we ensured that dvp is the parent of vp. This is possible in some
897 * rare conditions (races and the special .zfs directory).
900 error
= fop_lookup(dvp
, ".zfs", &cmpvp
, &pnp
, 0, vrootp
, cr
,
903 if (vnode_match(tvp
, cmpvp
, cr
)) {
904 (void) strcpy(dp
->d_name
, ".zfs");
905 dp
->d_reclen
= strlen(".zfs");
920 * Given a global path (from rootdir), and a vnode that is the current root,
921 * return the portion of the path that is beneath the current root or NULL on
922 * failure. The path MUST be a resolved path (no '..' entries or symlinks),
923 * otherwise this function will fail.
926 localpath(char *path
, struct vnode
*vrootp
, cred_t
*cr
)
930 char component
[MAXNAMELEN
];
935 * We use vn_compare() instead of VN_CMP() in order to detect lofs
936 * mounts and stacked vnodes.
938 if (vn_compare(vrootp
, rootdir
))
941 if (pn_get(path
, UIO_SYSSPACE
, &pn
) != 0)
947 if (vn_ismntpt(vp
) && traverse(&vp
) != 0) {
953 while (pn_pathleft(&pn
)) {
956 if (pn_getcomponent(&pn
, component
) != 0)
959 if (fop_lookup(vp
, component
, &cvp
, &pn
, 0, rootdir
, cr
,
960 NULL
, NULL
, NULL
) != 0)
965 if (vn_ismntpt(vp
) && traverse(&vp
) != 0)
968 if (vn_compare(vp
, vrootp
)) {
969 ret
= path
+ (pn
.pn_path
- pn
.pn_buf
);
981 * Clean a stale v_path from a vnode. This is only performed if the v_path has
982 * not been altered since it was found to be stale
985 vnode_clear_vpath(vnode_t
*vp
, char *vpath_old
)
987 mutex_enter(&vp
->v_lock
);
988 if (vp
->v_path
!= vn_vpath_empty
&& vp
->v_path
== vpath_old
) {
989 vp
->v_path
= vn_vpath_empty
;
990 mutex_exit(&vp
->v_lock
);
991 kmem_free(vpath_old
, strlen(vpath_old
) + 1);
993 mutex_exit(&vp
->v_lock
);
998 * Validate that a pathname refers to a given vnode.
1001 vnode_valid_pn(vnode_t
*vp
, vnode_t
*vrootp
, pathname_t
*pn
, pathname_t
*rpn
,
1002 int flags
, cred_t
*cr
)
1006 * If we are in a zone or a chroot environment, then we have to
1007 * take additional steps, since the path to the root might not
1008 * be readable with the current credentials, even though the
1009 * process can legitmately access the file. In this case, we
1012 * lookuppnvp() with all privileges to get the resolved path.
1013 * call localpath() to get the local portion of the path, and
1014 * continue as normal.
1016 * If the the conversion to a local path fails, then we continue
1017 * as normal. This is a heuristic to make process object file
1018 * paths available from within a zone. Because lofs doesn't
1019 * support page operations, the vnode stored in the seg_t is
1020 * actually the underlying real vnode, not the lofs node itself.
1021 * Most of the time, the lofs path is the same as the underlying
1022 * vnode (for example, /usr/lib/libc.so.1).
1024 if (vrootp
!= rootdir
) {
1028 if (lookuppnvp(pn
, rpn
, FOLLOW
, NULL
, &compvp
, rootdir
,
1029 rootdir
, kcred
) == 0) {
1030 local
= localpath(rpn
->pn_path
, vrootp
, kcred
);
1035 * The original pn was changed through lookuppnvp().
1036 * Set it to local for next validation attempt.
1039 (void) pn_set(pn
, local
);
1046 * We should have a local path at this point, so start the search from
1047 * the root of the current process.
1050 if (vrootp
!= rootdir
)
1052 if (lookuppnvp(pn
, rpn
, FOLLOW
| flags
, NULL
, &compvp
, vrootp
, vrootp
,
1055 * Check to see if the returned vnode is the same as the one we
1058 if (vn_compare(vp
, compvp
) ||
1059 vnode_match(vp
, compvp
, cr
)) {
1070 * Struct for tracking vnodes with invalidated v_path entries during a
1071 * dirtopath reverse lookup. By keeping adequate state, those vnodes can be
1072 * revisted to populate v_path.
1074 struct dirpath_walk
{
1075 struct dirpath_walk
*dw_next
;
1083 * Given a directory, return the full, resolved path. This looks up "..",
1084 * searches for the given vnode in the parent, appends the component, etc. It
1085 * is used to implement vnodetopath() and getcwd() when the cached path fails.
1088 dirtopath(vnode_t
*vrootp
, vnode_t
*vp
, char *buf
, size_t buflen
, int flags
,
1091 pathname_t pn
, rpn
, emptypn
;
1092 vnode_t
*pvp
= NULL
, *startvp
= vp
;
1096 char *bufloc
, *dbuf
;
1097 const size_t dlen
= DIRENT64_RECLEN(MAXPATHLEN
);
1098 struct dirpath_walk
*dw_chain
= NULL
, *dw_entry
;
1100 /* Operation only allowed on directories */
1101 ASSERT(vp
->v_type
== VDIR
);
1103 /* We must have at least enough space for "/" */
1105 return (ENAMETOOLONG
);
1107 /* Start at end of string with terminating null */
1108 bufloc
= &buf
[buflen
- 1];
1113 dbuf
= kmem_alloc(dlen
, KM_SLEEP
);
1114 bzero(&emptypn
, sizeof (emptypn
));
1117 * Begin with an additional reference on vp. This will be decremented
1124 hrtime_t cached_stamp
;
1127 * Return if we've reached the root. If the buffer is empty,
1128 * return '/'. We explicitly don't use vn_compare(), since it
1129 * compares the real vnodes. A lofs mount of '/' would produce
1130 * incorrect results otherwise.
1132 if (VN_CMP(vrootp
, vp
)) {
1133 if (*bufloc
== '\0')
1139 * If we've reached the VFS root, something has gone wrong. We
1140 * should have reached the root in the above check. The only
1141 * explantation is that 'vp' is not contained withing the given
1142 * root, in which case we return EPERM.
1144 if (VN_CMP(rootdir
, vp
)) {
1150 * Shortcut: see if this vnode has correct v_path. If so,
1151 * we have the work done.
1153 mutex_enter(&vp
->v_lock
);
1154 if (vp
->v_path
!= vn_vpath_empty
&&
1155 pn_set(&pn
, vp
->v_path
) == 0) {
1156 cached_stamp
= vp
->v_path_stamp
;
1157 mutex_exit(&vp
->v_lock
);
1158 rpn
.pn_path
= rpn
.pn_buf
;
1160 /* Ensure the v_path pointing to correct vnode */
1161 if (vnode_valid_pn(vp
, vrootp
, &pn
, &rpn
, flags
,
1163 complen
= strlen(rpn
.pn_path
);
1169 bcopy(rpn
.pn_path
, bufloc
, complen
);
1173 * Immediately nuke cached v_path entries known
1176 vn_clearpath(vp
, cached_stamp
);
1179 mutex_exit(&vp
->v_lock
);
1183 * Shortcuts failed, search for this vnode in its parent. If
1184 * this is a mountpoint, then get the vnode underneath.
1186 if (vp
->v_flag
& VROOT
)
1188 if ((err
= fop_lookup(vp
, "..", &pvp
, &emptypn
, 0, vrootp
, cr
,
1189 NULL
, NULL
, NULL
)) != 0)
1193 * With extended attributes, it's possible for a directory to
1194 * have a parent that is a regular file. Check for that here.
1196 if (pvp
->v_type
!= VDIR
) {
1202 * If this is true, something strange has happened. This is
1203 * only true if we are the root of a filesystem, which should
1204 * have been caught by the check above.
1206 if (VN_CMP(pvp
, vp
)) {
1212 * Check if we have read and search privilege so, that
1213 * we can lookup the path in the directory
1215 vprivs
= (flags
& LOOKUP_CHECKREAD
) ? VREAD
| VEXEC
: VEXEC
;
1216 if ((err
= fop_access(pvp
, vprivs
, 0, cr
, NULL
)) != 0) {
1221 * Search the parent directory for the entry corresponding to
1224 if ((err
= dirfindvp(vrootp
, pvp
, vp
, cr
, dbuf
, dlen
, &dp
))
1227 complen
= strlen(dp
->d_name
);
1229 if (bufloc
<= buf
) {
1233 bcopy(dp
->d_name
, bufloc
, complen
);
1235 /* Prepend a slash to the current path. */
1239 * Record the name and directory for later reconstruction and
1240 * link it up with the others.
1242 dw_entry
= kmem_alloc(sizeof (*dw_entry
), KM_SLEEP
);
1243 dw_entry
->dw_name
= kmem_alloc(complen
+ 1, KM_SLEEP
);
1244 VN_HOLD(dw_entry
->dw_vnode
= vp
);
1245 VN_HOLD(dw_entry
->dw_pvnode
= pvp
);
1246 bcopy(dp
->d_name
, dw_entry
->dw_name
, complen
+ 1);
1247 dw_entry
->dw_len
= complen
;
1248 dw_entry
->dw_next
= dw_chain
;
1249 dw_chain
= dw_entry
;
1251 /* And continue with the next component */
1258 * Place the path at the beginning of the buffer.
1261 ovbcopy(bufloc
, buf
, buflen
- (bufloc
- buf
));
1265 * Walk over encountered directory entries which were afflicted with a
1266 * stale or absent v_path. If the dirtopath was successful, we should
1267 * possess the necessary information to populate all of them with a
1270 * While processing this list, it is safe to call vn_setpath despite
1271 * the fact that racing vnode actions may have altered v_path entries
1272 * while the above loopwas still executing. Any updated entries will
1273 * have a newer v_path_stamp value which prevents an invalid overwrite.
1275 * If an error was encountered during the search, freeing the chain is
1278 dw_entry
= dw_chain
;
1279 while (dw_entry
!= NULL
) {
1280 struct dirpath_walk
*next
= dw_entry
->dw_next
;
1283 vn_setpath(NULL
, dw_entry
->dw_pvnode
,
1284 dw_entry
->dw_vnode
, dw_entry
->dw_name
,
1288 VN_RELE(dw_entry
->dw_vnode
);
1289 VN_RELE(dw_entry
->dw_pvnode
);
1290 kmem_free(dw_entry
->dw_name
, dw_entry
->dw_len
+ 1);
1291 kmem_free(dw_entry
, sizeof (*dw_entry
));
1296 * If the error was ESTALE and the current directory to look in
1297 * was the root for this lookup, the root for a mounted file
1298 * system, or the starting directory for lookups, then
1299 * return ENOENT instead of ESTALE. In this case, no recovery
1300 * is possible by the higher level. If ESTALE was returned for
1301 * some intermediate directory along the path, then recovery
1302 * is potentially possible and retrying from the higher level
1303 * will either correct the situation by purging stale cache
1304 * entries or eventually get back to the point where no recovery
1307 if (err
== ESTALE
&&
1308 (VN_CMP(vp
, vrootp
) || (vp
->v_flag
& VROOT
) || vp
== startvp
))
1311 kmem_free(dbuf
, dlen
);
1322 * The additional flag, LOOKUP_CHECKREAD, is used to enforce artificial
1323 * constraints in order to be standards compliant. For example, if we have
1324 * the cached path of '/foo/bar', and '/foo' has permissions 100 (execute
1325 * only), then we can legitimately look up the path to the current working
1326 * directory without needing read permission. Existing standards tests,
1327 * however, assume that we are determining the path by repeatedly looking up
1328 * "..". We need to keep this behavior in order to maintain backwards
1332 vnodetopath_common(vnode_t
*vrootp
, vnode_t
*vp
, char *buf
, size_t buflen
,
1333 cred_t
*cr
, int flags
)
1338 boolean_t doclose
= B_FALSE
;
1341 * If vrootp is NULL, get the root for curproc. Callers with any other
1342 * requirements should pass in a different vrootp.
1344 if (vrootp
== NULL
) {
1345 proc_t
*p
= curproc
;
1347 mutex_enter(&p
->p_lock
);
1348 if ((vrootp
= PTOU(p
)->u_rdir
) == NULL
)
1351 mutex_exit(&p
->p_lock
);
1357 * This is to get around an annoying artifact of the /proc filesystem,
1358 * which is the behavior of {cwd/root}. Trying to resolve this path
1359 * will result in /proc/pid/cwd instead of whatever the real working
1360 * directory is. We can't rely on fop_realvp(), since that will break
1361 * lofs. The only difference between procfs and lofs is that opening
1362 * the file will return the underling vnode in the case of procfs.
1364 if (vp
->v_type
== VDIR
&& fop_realvp(vp
, &realvp
, NULL
) == 0 &&
1367 if (fop_open(&vp
, FREAD
, cr
, NULL
) == 0)
1374 * Check to see if we have a valid cached path in the vnode.
1377 mutex_enter(&vp
->v_lock
);
1378 if (vp
->v_path
!= vn_vpath_empty
) {
1379 hrtime_t cached_stamp
;
1382 cached_stamp
= vp
->v_path_stamp
;
1383 (void) pn_set(&pn
, vp
->v_path
);
1384 mutex_exit(&vp
->v_lock
);
1386 /* We should only cache absolute paths */
1387 ASSERT(pn
.pn_buf
[0] == '/');
1390 if (vnode_valid_pn(vp
, vrootp
, &pn
, &rpn
, flags
, cr
) == 0) {
1391 /* Return the result, if we're able. */
1392 if (buflen
> rpn
.pn_pathlen
) {
1393 bcopy(rpn
.pn_path
, buf
, rpn
.pn_pathlen
+ 1);
1402 vn_clearpath(vp
, cached_stamp
);
1404 mutex_exit(&vp
->v_lock
);
1408 if (vp
->v_type
!= VDIR
) {
1410 * The reverse lookup tricks used by dirtopath aren't possible
1411 * for non-directory entries. The best which can be done is
1412 * clearing any stale v_path so later lookups can potentially
1413 * repopulate it with a valid path.
1417 ret
= dirtopath(vrootp
, vp
, buf
, buflen
, flags
, cr
);
1423 (void) fop_close(vp
, FREAD
, 1, 0, cr
, NULL
);
1431 vnodetopath(vnode_t
*vrootp
, vnode_t
*vp
, char *buf
, size_t buflen
, cred_t
*cr
)
1433 return (vnodetopath_common(vrootp
, vp
, buf
, buflen
, cr
, 0));
1437 dogetcwd(char *buf
, size_t buflen
)
1442 refstr_t
*cwd
, *oldcwd
;
1444 pathname_t rpnp
, pnp
;
1445 proc_t
*p
= curproc
;
1448 * Check to see if there is a cached version of the cwd. If so, lookup
1449 * the cached value and make sure it is the same vnode.
1451 mutex_enter(&p
->p_lock
);
1452 if ((cwd
= PTOU(p
)->u_cwd
) != NULL
)
1454 vp
= PTOU(p
)->u_cdir
;
1456 mutex_exit(&p
->p_lock
);
1459 * Make sure we have permission to access the current directory.
1461 if ((ret
= fop_access(vp
, VEXEC
, 0, CRED(), NULL
)) != 0) {
1469 value
= refstr_value(cwd
);
1470 if ((ret
= pn_get((char *)value
, UIO_SYSSPACE
, &pnp
)) != 0) {
1478 if (lookuppn(&pnp
, &rpnp
, NO_FOLLOW
, NULL
, &compvp
) == 0) {
1480 if (VN_CMP(vp
, compvp
) &&
1481 strcmp(value
, rpnp
.pn_path
) == 0) {
1486 if (strlen(value
) + 1 > buflen
) {
1488 return (ENAMETOOLONG
);
1490 bcopy(value
, buf
, strlen(value
) + 1);
1504 ret
= vnodetopath_common(NULL
, vp
, buf
, buflen
, CRED(),
1510 * Store the new cwd and replace the existing cached copy.
1513 cwd
= refstr_alloc(buf
);
1517 mutex_enter(&p
->p_lock
);
1518 oldcwd
= PTOU(p
)->u_cwd
;
1519 PTOU(p
)->u_cwd
= cwd
;
1520 mutex_exit(&p
->p_lock
);
1523 refstr_rele(oldcwd
);