2 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/kern/vfs_nlookup.c,v 1.25 2008/07/19 04:43:33 dillon Exp $
37 * nlookup() is the 'new' namei interface. Rather then return directory and
38 * leaf vnodes (in various lock states) the new interface instead deals in
39 * namecache records. Namecache records may represent both a positive or
40 * a negative hit. The namespace is locked via the namecache record instead
41 * of via the vnode, and only the leaf namecache record (representing the
42 * filename) needs to be locked.
44 * This greatly improves filesystem parallelism and is a huge simplification
45 * of the API verses the old vnode locking / namei scheme.
47 * Filesystems must actively control the caching aspects of the namecache,
48 * and since namecache pointers are used as handles they are non-optional
49 * even for filesystems which do not generally wish to cache things. It is
50 * intended that a separate cache coherency API will be constructed to handle
54 #include "opt_ktrace.h"
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/kernel.h>
59 #include <sys/vnode.h>
60 #include <sys/mount.h>
61 #include <sys/filedesc.h>
63 #include <sys/namei.h>
64 #include <sys/nlookup.h>
65 #include <sys/malloc.h>
67 #include <sys/objcache.h>
71 #include <sys/ktrace.h>
75 * Initialize a nlookup() structure, early error return for copyin faults
76 * or a degenerate empty string (which is not allowed).
78 * The first process proc0's credentials are used if the calling thread
79 * is not associated with a process context.
84 nlookup_init(struct nlookupdata
*nd
,
85 const char *path
, enum uio_seg seg
, int flags
)
96 * note: the pathlen set by copy*str() includes the terminating \0.
98 bzero(nd
, sizeof(struct nlookupdata
));
99 nd
->nl_path
= objcache_get(namei_oc
, M_WAITOK
);
100 nd
->nl_flags
|= NLC_HASBUF
;
101 if (seg
== UIO_SYSSPACE
)
102 error
= copystr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
104 error
= copyinstr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
107 * Don't allow empty pathnames.
108 * POSIX.1 requirement: "" is not a vaild file name.
110 if (error
== 0 && pathlen
<= 1)
115 cache_copy(&p
->p_fd
->fd_ncdir
, &nd
->nl_nch
);
116 cache_copy(&p
->p_fd
->fd_nrdir
, &nd
->nl_rootnch
);
117 if (p
->p_fd
->fd_njdir
.ncp
)
118 cache_copy(&p
->p_fd
->fd_njdir
, &nd
->nl_jailnch
);
119 nd
->nl_cred
= crhold(p
->p_ucred
);
121 cache_copy(&rootnch
, &nd
->nl_nch
);
122 cache_copy(&nd
->nl_nch
, &nd
->nl_rootnch
);
123 cache_copy(&nd
->nl_nch
, &nd
->nl_jailnch
);
124 nd
->nl_cred
= crhold(proc0
.p_ucred
);
127 nd
->nl_flags
|= flags
;
136 * nlookup_init() for "at" family of syscalls.
138 * Works similarly to nlookup_init() but if path is relative and fd is not
139 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd.
140 * In this case, the file entry pointed to by fd is ref'ed and returned in
143 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd
144 * and release the ref to the file entry.
147 nlookup_init_at(struct nlookupdata
*nd
, struct file
**fpp
, int fd
,
148 const char *path
, enum uio_seg seg
, int flags
)
150 struct thread
*td
= curthread
;
151 struct proc
*p
= td
->td_proc
;
158 if ((error
= nlookup_init(nd
, path
, seg
, flags
)) != 0) {
162 if (nd
->nl_path
[0] != '/' && fd
!= AT_FDCWD
) {
163 if ((error
= holdvnode(p
->p_fd
, fd
, &fp
)) != 0)
165 vp
= (struct vnode
*)fp
->f_data
;
166 if (vp
->v_type
!= VDIR
|| fp
->f_nchandle
.ncp
== NULL
) {
172 cache_drop(&nd
->nl_nch
);
173 cache_copy(&fp
->f_nchandle
, &nd
->nl_nch
);
186 * This works similarly to nlookup_init() but does not assume a process
187 * context. rootnch is always chosen for the root directory and the cred
188 * and starting directory are supplied in arguments.
191 nlookup_init_raw(struct nlookupdata
*nd
,
192 const char *path
, enum uio_seg seg
, int flags
,
193 struct ucred
*cred
, struct nchandle
*ncstart
)
201 bzero(nd
, sizeof(struct nlookupdata
));
202 nd
->nl_path
= objcache_get(namei_oc
, M_WAITOK
);
203 nd
->nl_flags
|= NLC_HASBUF
;
204 if (seg
== UIO_SYSSPACE
)
205 error
= copystr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
207 error
= copyinstr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
210 * Don't allow empty pathnames.
211 * POSIX.1 requirement: "" is not a vaild file name.
213 if (error
== 0 && pathlen
<= 1)
217 cache_copy(ncstart
, &nd
->nl_nch
);
218 cache_copy(&rootnch
, &nd
->nl_rootnch
);
219 cache_copy(&rootnch
, &nd
->nl_jailnch
);
220 nd
->nl_cred
= crhold(cred
);
222 nd
->nl_flags
|= flags
;
230 * Set a different credential; this credential will be used by future
231 * operations performed on nd.nl_open_vp and nlookupdata structure.
234 nlookup_set_cred(struct nlookupdata
*nd
, struct ucred
*cred
)
236 KKASSERT(nd
->nl_cred
!= NULL
);
238 if (nd
->nl_cred
!= cred
) {
246 * Cleanup a nlookupdata structure after we are through with it. This may
247 * be called on any nlookupdata structure initialized with nlookup_init().
248 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
249 * returns an error, even if as a consumer you believe you have taken all
250 * dynamic elements out of the nlookupdata structure.
253 nlookup_done(struct nlookupdata
*nd
)
255 if (nd
->nl_nch
.ncp
) {
256 if (nd
->nl_flags
& NLC_NCPISLOCKED
) {
257 nd
->nl_flags
&= ~NLC_NCPISLOCKED
;
258 cache_unlock(&nd
->nl_nch
);
260 cache_drop(&nd
->nl_nch
);
262 if (nd
->nl_rootnch
.ncp
)
263 cache_drop(&nd
->nl_rootnch
);
264 if (nd
->nl_jailnch
.ncp
)
265 cache_drop(&nd
->nl_jailnch
);
266 if ((nd
->nl_flags
& NLC_HASBUF
) && nd
->nl_path
) {
267 objcache_put(namei_oc
, nd
->nl_path
);
274 if (nd
->nl_open_vp
) {
275 if (nd
->nl_flags
& NLC_LOCKVP
) {
276 vn_unlock(nd
->nl_open_vp
);
277 nd
->nl_flags
&= ~NLC_LOCKVP
;
279 vn_close(nd
->nl_open_vp
, nd
->nl_vp_fmode
);
280 nd
->nl_open_vp
= NULL
;
286 nd
->nl_flags
= 0; /* clear remaining flags (just clear everything) */
290 * Works similarly to nlookup_done() when nd initialized with
294 nlookup_done_at(struct nlookupdata
*nd
, struct file
*fp
)
302 nlookup_zero(struct nlookupdata
*nd
)
304 bzero(nd
, sizeof(struct nlookupdata
));
308 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
309 * if an error occured.
311 * Note that the returned ncp is not checked for permissions, though VEXEC
312 * is checked on the directory path leading up to the result. The caller
313 * must call naccess() to check the permissions of the returned leaf.
316 nlookup_simple(const char *str
, enum uio_seg seg
,
317 int niflags
, int *error
)
319 struct nlookupdata nd
;
322 *error
= nlookup_init(&nd
, str
, seg
, niflags
);
324 if ((*error
= nlookup(&nd
)) == 0) {
325 nch
= nd
.nl_nch
; /* keep hold ref from structure */
326 cache_zero(&nd
.nl_nch
); /* and NULL out */
338 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
339 * on return, even if an error occurs. If no error occurs the returned
340 * nl_nch is always referenced and locked, otherwise it may or may not be.
342 * Intermediate directory elements, including the current directory, require
343 * execute (search) permission. nlookup does not examine the access
344 * permissions on the returned element.
346 * If NLC_CREATE is set the last directory must allow node creation,
347 * and an error code of 0 will be returned for a non-existant
348 * target (not ENOENT).
350 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
351 * plus the sticky check is made, and an error code of 0 will be returned
352 * for a non-existant target (not ENOENT).
354 * If NLC_DELETE is set the last directory mut allow node deletion,
355 * plus the sticky check is made.
357 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
358 * of the returned entry. The vnode will be referenced, but not locked,
359 * and will be released by nlookup_done() along with everything else.
362 nlookup(struct nlookupdata
*nd
)
364 struct nlcomponent nlc
;
367 struct nchandle nctmp
;
377 if (KTRPOINT(nd
->nl_td
, KTR_NAMEI
))
378 ktrnamei(nd
->nl_td
->td_lwp
, nd
->nl_path
);
380 bzero(&nlc
, sizeof(nlc
));
383 * Setup for the loop. The current working namecache element must
384 * be in a refd + unlocked state. This typically the case on entry except
385 * when stringing nlookup()'s along in a chain, since nlookup() always
386 * returns nl_nch in a locked state.
389 if (nd
->nl_flags
& NLC_NCPISLOCKED
) {
390 nd
->nl_flags
&= ~NLC_NCPISLOCKED
;
391 cache_unlock(&nd
->nl_nch
);
400 * Loop on the path components. At the top of the loop nd->nl_nch
401 * is ref'd and unlocked and represents our current position.
405 * Check if the root directory should replace the current
406 * directory. This is done at the start of a translation
407 * or after a symbolic link has been found. In other cases
408 * ptr will never be pointing at a '/'.
413 } while (*ptr
== '/');
414 cache_copy(&nd
->nl_rootnch
, &nch
);
415 cache_drop(&nd
->nl_nch
);
419 * Fast-track termination. There is no parent directory of
420 * the root in the same mount from the point of view of
421 * the caller so return EPERM if NLC_REFDVP is specified.
422 * e.g. 'rmdir /' is not allowed.
425 if (nd
->nl_flags
& NLC_REFDVP
) {
428 cache_lock(&nd
->nl_nch
);
429 nd
->nl_flags
|= NLC_NCPISLOCKED
;
438 * Check directory search permissions.
441 if ((error
= naccess(&nd
->nl_nch
, NLC_EXEC
, nd
->nl_cred
, &dflags
)) != 0)
445 * Extract the path component. Path components are limited to
448 nlc
.nlc_nameptr
= ptr
;
449 while (*ptr
&& *ptr
!= '/')
451 nlc
.nlc_namelen
= ptr
- nlc
.nlc_nameptr
;
452 if (nlc
.nlc_namelen
>= 256) {
453 error
= ENAMETOOLONG
;
458 * Lookup the path component in the cache, creating an unresolved
459 * entry if necessary. We have to handle "." and ".." as special
462 * When handling ".." we have to detect a traversal back through a
463 * mount point. If we are at the root, ".." just returns the root.
465 * When handling "." or ".." we also have to recalculate dflags
466 * since our dflags will be for some sub-directory instead of the
469 * This subsection returns a locked, refd 'nch' unless it errors out.
470 * The namecache topology is not allowed to be disconnected, so
471 * encountering a NULL parent will generate EINVAL. This typically
472 * occurs when a directory is removed out from under a process.
474 if (nlc
.nlc_namelen
== 1 && nlc
.nlc_nameptr
[0] == '.') {
475 cache_get(&nd
->nl_nch
, &nch
);
477 } else if (nlc
.nlc_namelen
== 2 &&
478 nlc
.nlc_nameptr
[0] == '.' && nlc
.nlc_nameptr
[1] == '.') {
479 if (nd
->nl_nch
.mount
== nd
->nl_rootnch
.mount
&&
480 nd
->nl_nch
.ncp
== nd
->nl_rootnch
.ncp
483 * ".." at the root returns the root
485 cache_get(&nd
->nl_nch
, &nch
);
488 * Locate the parent ncp. If we are at the root of a
489 * filesystem mount we have to skip to the mounted-on
490 * point in the underlying filesystem.
492 * Expect the parent to always be good since the
493 * mountpoint doesn't go away. XXX hack. cache_get()
494 * requires the ncp to already have a ref as a safety.
497 while (nctmp
.ncp
== nctmp
.mount
->mnt_ncmountpt
.ncp
)
498 nctmp
= nctmp
.mount
->mnt_ncmounton
;
499 nctmp
.ncp
= nctmp
.ncp
->nc_parent
;
500 KKASSERT(nctmp
.ncp
!= NULL
);
501 cache_copy(&nctmp
, &nch
); /* XXX hack */
502 cache_get(&nch
, &nch
);
503 cache_drop(&nctmp
); /* NOTE: zero's nctmp */
507 nch
= cache_nlookup(&nd
->nl_nch
, &nlc
);
508 while ((error
= cache_resolve(&nch
, nd
->nl_cred
)) == EAGAIN
) {
509 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
510 nch
.ncp
->nc_nlen
, nch
.ncp
->nc_nlen
, nch
.ncp
->nc_name
);
512 nch
= cache_nlookup(&nd
->nl_nch
, &nlc
);
518 * If the last component was "." or ".." our dflags no longer
519 * represents the parent directory and we have to explicitly
522 * Expect the parent to be good since nch is locked.
524 if (wasdotordotdot
&& error
== 0) {
526 if ((par
.ncp
= nch
.ncp
->nc_parent
) != NULL
) {
527 par
.mount
= nch
.mount
;
530 error
= naccess(&par
, 0, nd
->nl_cred
, &dflags
);
536 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
540 * Resolve the namespace if necessary. The ncp returned by
541 * cache_nlookup() is referenced and locked.
543 * XXX neither '.' nor '..' should return EAGAIN since they were
544 * previously resolved and thus cannot be newly created ncp's.
546 if (nch
.ncp
->nc_flag
& NCF_UNRESOLVED
) {
547 error
= cache_resolve(&nch
, nd
->nl_cred
);
548 KKASSERT(error
!= EAGAIN
);
550 error
= nch
.ncp
->nc_error
;
554 * Early completion. ENOENT is not an error if this is the last
555 * component and NLC_CREATE or NLC_RENAME (rename target) was
556 * requested. Note that ncp->nc_error is left as ENOENT in that
557 * case, which we check later on.
559 * Also handle invalid '.' or '..' components terminating a path
560 * for a create/rename/delete. The standard requires this and pax
561 * pretty stupidly depends on it.
563 for (xptr
= ptr
; *xptr
== '/'; ++xptr
)
566 if (error
== ENOENT
&&
567 (nd
->nl_flags
& (NLC_CREATE
| NLC_RENAME_DST
))
569 if (nd
->nl_flags
& NLC_NFS_RDONLY
) {
572 error
= naccess(&nch
, nd
->nl_flags
| dflags
,
576 if (error
== 0 && wasdotordotdot
&&
577 (nd
->nl_flags
& (NLC_CREATE
| NLC_DELETE
|
578 NLC_RENAME_SRC
| NLC_RENAME_DST
))) {
582 if (nd
->nl_flags
& NLC_CREATE
)
584 else if (nd
->nl_flags
& NLC_DELETE
)
585 error
= (wasdotordotdot
== 1) ? EINVAL
: ENOTEMPTY
;
592 * Early completion on error.
600 * If the element is a symlink and it is either not the last
601 * element or it is the last element and we are allowed to
602 * follow symlinks, resolve the symlink.
604 if ((nch
.ncp
->nc_flag
& NCF_ISSYMLINK
) &&
605 (*ptr
|| (nd
->nl_flags
& NLC_FOLLOW
))
607 if (nd
->nl_loopcnt
++ >= MAXSYMLINKS
) {
612 error
= nreadsymlink(nd
, &nch
, &nlc
);
618 * Concatenate trailing path elements onto the returned symlink.
619 * Note that if the path component (ptr) is not exhausted, it
620 * will being with a '/', so we do not have to add another one.
622 * The symlink may not be empty.
625 if (nlc
.nlc_namelen
== 0 || nlc
.nlc_namelen
+ len
>= MAXPATHLEN
) {
626 error
= nlc
.nlc_namelen
? ENAMETOOLONG
: ENOENT
;
627 objcache_put(namei_oc
, nlc
.nlc_nameptr
);
630 bcopy(ptr
, nlc
.nlc_nameptr
+ nlc
.nlc_namelen
, len
+ 1);
631 if (nd
->nl_flags
& NLC_HASBUF
)
632 objcache_put(namei_oc
, nd
->nl_path
);
633 nd
->nl_path
= nlc
.nlc_nameptr
;
634 nd
->nl_flags
|= NLC_HASBUF
;
638 * Go back up to the top to resolve any initial '/'s in the
645 * If the element is a directory and we are crossing a mount point,
648 while ((nch
.ncp
->nc_flag
& NCF_ISMOUNTPT
) &&
649 (nd
->nl_flags
& NLC_NOCROSSMOUNT
) == 0 &&
650 (mp
= cache_findmount(&nch
)) != NULL
655 cache_get(&mp
->mnt_ncmountpt
, &nch
);
657 if (nch
.ncp
->nc_flag
& NCF_UNRESOLVED
) {
658 while (vfs_busy(mp
, 0))
660 error
= VFS_ROOT(mp
, &tdp
);
664 cache_setvp(&nch
, tdp
);
674 * Skip any slashes to get to the next element. If there
675 * are any slashes at all the current element must be a
676 * directory or, in the create case, intended to become a directory.
677 * If it isn't we break without incrementing ptr and fall through
678 * to the failure case below.
680 while (*ptr
== '/') {
681 if ((nch
.ncp
->nc_flag
& NCF_ISDIR
) == 0 &&
682 !(nd
->nl_flags
& NLC_WILLBEDIR
)
690 * Continuation case: additional elements and the current
691 * element is a directory.
693 if (*ptr
&& (nch
.ncp
->nc_flag
& NCF_ISDIR
)) {
694 cache_drop(&nd
->nl_nch
);
701 * Failure case: additional elements and the current element
711 * Successful lookup of last element.
713 * Check permissions if the target exists. If the target does not
714 * exist directory permissions were already tested in the early
715 * completion code above.
717 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
718 * if the file is marked append-only, and NLC_STICKY if the directory
719 * containing the file is sticky.
721 if (nch
.ncp
->nc_vp
&& (nd
->nl_flags
& NLC_ALLCHKS
)) {
722 error
= naccess(&nch
, nd
->nl_flags
| dflags
,
731 * Termination: no more elements.
733 * If NLC_REFDVP is set acquire a referenced parent dvp.
735 if (nd
->nl_flags
& NLC_REFDVP
) {
736 error
= cache_vref(&nd
->nl_nch
, nd
->nl_cred
, &nd
->nl_dvp
);
738 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch
.ncp
);
743 cache_drop(&nd
->nl_nch
);
745 nd
->nl_flags
|= NLC_NCPISLOCKED
;
751 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
752 * (ncp->nc_error will be ENOENT), but we will still return an error
759 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
760 * of continuity in the namecache tree by connecting the ncp related to the
761 * vnode under the mount to the ncp related to the mount's root vnode.
763 * If no error occured a locked, ref'd ncp is stored in *ncpp.
766 nlookup_mp(struct mount
*mp
, struct nchandle
*nch
)
772 cache_get(&mp
->mnt_ncmountpt
, nch
);
773 if (nch
->ncp
->nc_flag
& NCF_UNRESOLVED
) {
774 while (vfs_busy(mp
, 0))
776 error
= VFS_ROOT(mp
, &vp
);
781 cache_setvp(nch
, vp
);
789 * Read the contents of a symlink, allocate a path buffer out of the
790 * namei_oc and initialize the supplied nlcomponent with the result.
792 * If an error occurs no buffer will be allocated or returned in the nlc.
795 nreadsymlink(struct nlookupdata
*nd
, struct nchandle
*nch
,
796 struct nlcomponent
*nlc
)
805 nlc
->nlc_nameptr
= NULL
;
806 nlc
->nlc_namelen
= 0;
807 if (nch
->ncp
->nc_vp
== NULL
)
809 if ((error
= cache_vget(nch
, nd
->nl_cred
, LK_SHARED
, &vp
)) != 0)
811 cp
= objcache_get(namei_oc
, M_WAITOK
);
813 aiov
.iov_len
= MAXPATHLEN
;
814 auio
.uio_iov
= &aiov
;
817 auio
.uio_rw
= UIO_READ
;
818 auio
.uio_segflg
= UIO_SYSSPACE
;
819 auio
.uio_td
= nd
->nl_td
;
820 auio
.uio_resid
= MAXPATHLEN
- 1;
821 error
= VOP_READLINK(vp
, &auio
, nd
->nl_cred
);
824 linklen
= MAXPATHLEN
- 1 - auio
.uio_resid
;
826 linklen
= varsymreplace(cp
, linklen
, MAXPATHLEN
- 1);
828 error
= ENAMETOOLONG
;
833 nlc
->nlc_nameptr
= cp
;
834 nlc
->nlc_namelen
= linklen
;
838 objcache_put(namei_oc
, cp
);
844 * Check access [XXX cache vattr!] [XXX quota]
846 * Generally check the NLC_* access bits. All specified bits must pass
847 * for this function to return 0.
849 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
850 * access, otherwise it must exist. No error is returned in this case.
852 * The file must not exist if NLC_EXCL is specified.
854 * Directory permissions in general are tested for NLC_CREATE if the file
855 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
856 * whether the file exists or not.
858 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
859 * the latter is only tested if the target exists.
861 * The passed ncp may or may not be locked. The caller should use a
862 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
863 * NLC_DELETE, and NLC_EXCL checks.
866 naccess(struct nchandle
*nch
, int nflags
, struct ucred
*cred
, int *nflagsp
)
873 if (nch
->ncp
->nc_flag
& NCF_UNRESOLVED
) {
875 cache_resolve(nch
, cred
);
878 error
= nch
->ncp
->nc_error
;
881 * Directory permissions checks. Silently ignore ENOENT if these
882 * tests pass. It isn't an error.
884 if (nflags
& (NLC_CREATE
| NLC_DELETE
| NLC_RENAME_SRC
| NLC_RENAME_DST
)) {
885 if (((nflags
& NLC_CREATE
) && nch
->ncp
->nc_vp
== NULL
) ||
886 ((nflags
& NLC_DELETE
) && nch
->ncp
->nc_vp
!= NULL
) ||
887 ((nflags
& NLC_RENAME_SRC
) && nch
->ncp
->nc_vp
!= NULL
) ||
888 (nflags
& NLC_RENAME_DST
)
893 lwkt_gettoken(&nlock
, &vfs_token
);
894 if ((par
.ncp
= nch
->ncp
->nc_parent
) == NULL
) {
897 } else if (error
== 0 || error
== ENOENT
) {
898 par
.mount
= nch
->mount
;
901 error
= naccess(&par
, NLC_WRITE
, cred
, NULL
);
904 lwkt_reltoken(&nlock
);
909 * NLC_EXCL check. Target file must not exist.
911 if (error
== 0 && (nflags
& NLC_EXCL
) && nch
->ncp
->nc_vp
!= NULL
)
915 * Get the vnode attributes so we can do the rest of our checks.
917 * NOTE: We only call naccess_va() if the target exists.
920 error
= cache_vget(nch
, cred
, LK_SHARED
, &vp
);
921 if (error
== ENOENT
) {
923 * Silently zero-out ENOENT if creating or renaming
924 * (rename target). It isn't an error.
926 if (nflags
& (NLC_CREATE
| NLC_RENAME_DST
))
928 } else if (error
== 0) {
930 * Get the vnode attributes and check for illegal O_TRUNC
931 * requests and read-only mounts.
933 * NOTE: You can still open devices on read-only mounts for
936 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
937 * check on the parent directory above.
939 * XXX cache the va in the namecache or in the vnode
941 error
= VOP_GETATTR(vp
, &va
);
942 if (error
== 0 && (nflags
& NLC_TRUNCATE
)) {
958 if (error
== 0 && (nflags
& NLC_WRITE
) && vp
->v_mount
&&
959 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
)
975 * Check permissions based on file attributes. The passed
976 * flags (*nflagsp) are modified with feedback based on
977 * special attributes and requirements.
981 * Adjust the returned (*nflagsp) if non-NULL.
984 if ((va
.va_mode
& VSVTX
) && va
.va_uid
!= cred
->cr_uid
)
985 *nflagsp
|= NLC_STICKY
;
986 if (va
.va_flags
& APPEND
)
987 *nflagsp
|= NLC_APPENDONLY
;
988 if (va
.va_flags
& IMMUTABLE
)
989 *nflagsp
|= NLC_IMMUTABLE
;
993 * Process general access.
995 error
= naccess_va(&va
, nflags
, cred
);
1003 * Check the requested access against the given vattr using cred.
1006 naccess_va(struct vattr
*va
, int nflags
, struct ucred
*cred
)
1012 * Test the immutable bit. Creations, deletions, renames (source
1013 * or destination) are not allowed. chown/chmod/other is also not
1014 * allowed but is handled by SETATTR. Hardlinks to the immutable
1017 * If the directory is set to immutable then creations, deletions,
1018 * renames (source or dest) and hardlinks to files within the directory
1019 * are not allowed, and regular files opened through the directory may
1020 * not be written to or truncated (unless a special device).
1022 * NOTE! New hardlinks to immutable files work but new hardlinks to
1023 * files, immutable or not, sitting inside an immutable directory are
1024 * not allowed. As always if the file is hardlinked via some other
1025 * path additional hardlinks may be possible even if the file is marked
1026 * immutable. The sysop needs to create a closure by checking the hard
1027 * link count. Once closure is achieved you are good, and security
1028 * scripts should check link counts anyway.
1030 * Writes and truncations are only allowed on special devices.
1032 if ((va
->va_flags
& IMMUTABLE
) || (nflags
& NLC_IMMUTABLE
)) {
1033 if ((nflags
& NLC_IMMUTABLE
) && (nflags
& NLC_HLINK
))
1035 if (nflags
& (NLC_CREATE
| NLC_DELETE
|
1036 NLC_RENAME_SRC
| NLC_RENAME_DST
)) {
1039 if (nflags
& (NLC_WRITE
| NLC_TRUNCATE
)) {
1040 switch(va
->va_type
) {
1054 * Test the no-unlink and append-only bits for opens, rename targets,
1055 * and deletions. These bits are not tested for creations or
1058 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1059 * If you do not wish this you must also set NOUNLINK.
1061 * If the governing directory is marked APPEND-only it implies
1062 * NOUNLINK for all entries in the directory.
1064 if (((va
->va_flags
& NOUNLINK
) || (nflags
& NLC_APPENDONLY
)) &&
1065 (nflags
& (NLC_DELETE
| NLC_RENAME_SRC
| NLC_RENAME_DST
))
1071 * A file marked append-only may not be deleted but can be renamed.
1073 if ((va
->va_flags
& APPEND
) &&
1074 (nflags
& (NLC_DELETE
| NLC_RENAME_DST
))
1080 * A file marked append-only which is opened for writing must also
1081 * be opened O_APPEND.
1083 if ((va
->va_flags
& APPEND
) && (nflags
& (NLC_OPEN
| NLC_TRUNCATE
))) {
1084 if (nflags
& NLC_TRUNCATE
)
1086 if ((nflags
& (NLC_OPEN
| NLC_WRITE
)) == (NLC_OPEN
| NLC_WRITE
)) {
1087 if ((nflags
& NLC_APPEND
) == 0)
1093 * root gets universal access
1095 if (cred
->cr_uid
== 0)
1099 * Check owner perms.
1101 * If NLC_OWN is set the owner of the file is allowed no matter when
1102 * the owner-mode bits say (utimes).
1105 if (nflags
& NLC_READ
)
1107 if (nflags
& NLC_WRITE
)
1109 if (nflags
& NLC_EXEC
)
1112 if (cred
->cr_uid
== va
->va_uid
) {
1113 if ((nflags
& NLC_OWN
) == 0) {
1114 if ((vmode
& va
->va_mode
) != vmode
)
1121 * If NLC_STICKY is set only the owner may delete or rename a file.
1122 * This bit is typically set on /tmp.
1124 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1125 * the specific delete or rename case. For deletions and renames we
1126 * usually just care about directory permissions, not file permissions.
1128 if ((nflags
& NLC_STICKY
) &&
1129 (nflags
& (NLC_RENAME_SRC
| NLC_RENAME_DST
| NLC_DELETE
))) {
1137 for (i
= 0; i
< cred
->cr_ngroups
; ++i
) {
1138 if (va
->va_gid
== cred
->cr_groups
[i
]) {
1139 if ((vmode
& va
->va_mode
) != vmode
)
1149 if ((vmode
& va
->va_mode
) != vmode
)