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>
74 static int naccess(struct nchandle
*nch
, int vmode
, struct ucred
*cred
,
78 * Initialize a nlookup() structure, early error return for copyin faults
79 * or a degenerate empty string (which is not allowed).
81 * The first process proc0's credentials are used if the calling thread
82 * is not associated with a process context.
87 nlookup_init(struct nlookupdata
*nd
,
88 const char *path
, enum uio_seg seg
, int flags
)
99 * note: the pathlen set by copy*str() includes the terminating \0.
101 bzero(nd
, sizeof(struct nlookupdata
));
102 nd
->nl_path
= objcache_get(namei_oc
, M_WAITOK
);
103 nd
->nl_flags
|= NLC_HASBUF
;
104 if (seg
== UIO_SYSSPACE
)
105 error
= copystr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
107 error
= copyinstr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
110 * Don't allow empty pathnames.
111 * POSIX.1 requirement: "" is not a vaild file name.
113 if (error
== 0 && pathlen
<= 1)
118 cache_copy(&p
->p_fd
->fd_ncdir
, &nd
->nl_nch
);
119 cache_copy(&p
->p_fd
->fd_nrdir
, &nd
->nl_rootnch
);
120 if (p
->p_fd
->fd_njdir
.ncp
)
121 cache_copy(&p
->p_fd
->fd_njdir
, &nd
->nl_jailnch
);
122 nd
->nl_cred
= crhold(p
->p_ucred
);
124 cache_copy(&rootnch
, &nd
->nl_nch
);
125 cache_copy(&nd
->nl_nch
, &nd
->nl_rootnch
);
126 cache_copy(&nd
->nl_nch
, &nd
->nl_jailnch
);
127 nd
->nl_cred
= crhold(proc0
.p_ucred
);
130 nd
->nl_flags
|= flags
;
139 * nlookup_init() for "at" family of syscalls.
141 * Works similarly to nlookup_init() but if path is relative and fd is not
142 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd.
143 * In this case, the file entry pointed to by fd is ref'ed and returned in
146 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd
147 * and release the ref to the file entry.
150 nlookup_init_at(struct nlookupdata
*nd
, struct file
**fpp
, int fd
,
151 const char *path
, enum uio_seg seg
, int flags
)
153 struct thread
*td
= curthread
;
154 struct proc
*p
= td
->td_proc
;
161 if ((error
= nlookup_init(nd
, path
, seg
, flags
)) != 0) {
165 if (nd
->nl_path
[0] != '/' && fd
!= AT_FDCWD
) {
166 if ((error
= holdvnode(p
->p_fd
, fd
, &fp
)) != 0)
168 vp
= (struct vnode
*)fp
->f_data
;
169 if (vp
->v_type
!= VDIR
|| fp
->f_nchandle
.ncp
== NULL
) {
175 cache_drop(&nd
->nl_nch
);
176 cache_copy(&fp
->f_nchandle
, &nd
->nl_nch
);
189 * This works similarly to nlookup_init() but does not assume a process
190 * context. rootnch is always chosen for the root directory and the cred
191 * and starting directory are supplied in arguments.
194 nlookup_init_raw(struct nlookupdata
*nd
,
195 const char *path
, enum uio_seg seg
, int flags
,
196 struct ucred
*cred
, struct nchandle
*ncstart
)
204 bzero(nd
, sizeof(struct nlookupdata
));
205 nd
->nl_path
= objcache_get(namei_oc
, M_WAITOK
);
206 nd
->nl_flags
|= NLC_HASBUF
;
207 if (seg
== UIO_SYSSPACE
)
208 error
= copystr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
210 error
= copyinstr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
213 * Don't allow empty pathnames.
214 * POSIX.1 requirement: "" is not a vaild file name.
216 if (error
== 0 && pathlen
<= 1)
220 cache_copy(ncstart
, &nd
->nl_nch
);
221 cache_copy(&rootnch
, &nd
->nl_rootnch
);
222 cache_copy(&rootnch
, &nd
->nl_jailnch
);
223 nd
->nl_cred
= crhold(cred
);
225 nd
->nl_flags
|= flags
;
233 * This works similarly to nlookup_init_raw() but does not rely
234 * on rootnch being initialized yet.
237 nlookup_init_root(struct nlookupdata
*nd
,
238 const char *path
, enum uio_seg seg
, int flags
,
239 struct ucred
*cred
, struct nchandle
*ncstart
,
240 struct nchandle
*ncroot
)
248 bzero(nd
, sizeof(struct nlookupdata
));
249 nd
->nl_path
= objcache_get(namei_oc
, M_WAITOK
);
250 nd
->nl_flags
|= NLC_HASBUF
;
251 if (seg
== UIO_SYSSPACE
)
252 error
= copystr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
254 error
= copyinstr(path
, nd
->nl_path
, MAXPATHLEN
, &pathlen
);
257 * Don't allow empty pathnames.
258 * POSIX.1 requirement: "" is not a vaild file name.
260 if (error
== 0 && pathlen
<= 1)
264 cache_copy(ncstart
, &nd
->nl_nch
);
265 cache_copy(ncroot
, &nd
->nl_rootnch
);
266 cache_copy(ncroot
, &nd
->nl_jailnch
);
267 nd
->nl_cred
= crhold(cred
);
269 nd
->nl_flags
|= flags
;
277 * Set a different credential; this credential will be used by future
278 * operations performed on nd.nl_open_vp and nlookupdata structure.
281 nlookup_set_cred(struct nlookupdata
*nd
, struct ucred
*cred
)
283 KKASSERT(nd
->nl_cred
!= NULL
);
285 if (nd
->nl_cred
!= cred
) {
293 * Cleanup a nlookupdata structure after we are through with it. This may
294 * be called on any nlookupdata structure initialized with nlookup_init().
295 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
296 * returns an error, even if as a consumer you believe you have taken all
297 * dynamic elements out of the nlookupdata structure.
300 nlookup_done(struct nlookupdata
*nd
)
302 if (nd
->nl_nch
.ncp
) {
303 if (nd
->nl_flags
& NLC_NCPISLOCKED
) {
304 nd
->nl_flags
&= ~NLC_NCPISLOCKED
;
305 cache_unlock(&nd
->nl_nch
);
307 cache_drop(&nd
->nl_nch
); /* NULL's out the nch */
309 if (nd
->nl_rootnch
.ncp
)
310 cache_drop(&nd
->nl_rootnch
);
311 if (nd
->nl_jailnch
.ncp
)
312 cache_drop(&nd
->nl_jailnch
);
313 if ((nd
->nl_flags
& NLC_HASBUF
) && nd
->nl_path
) {
314 objcache_put(namei_oc
, nd
->nl_path
);
321 if (nd
->nl_open_vp
) {
322 if (nd
->nl_flags
& NLC_LOCKVP
) {
323 vn_unlock(nd
->nl_open_vp
);
324 nd
->nl_flags
&= ~NLC_LOCKVP
;
326 vn_close(nd
->nl_open_vp
, nd
->nl_vp_fmode
);
327 nd
->nl_open_vp
= NULL
;
333 nd
->nl_flags
= 0; /* clear remaining flags (just clear everything) */
337 * Works similarly to nlookup_done() when nd initialized with
341 nlookup_done_at(struct nlookupdata
*nd
, struct file
*fp
)
349 nlookup_zero(struct nlookupdata
*nd
)
351 bzero(nd
, sizeof(struct nlookupdata
));
355 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
356 * if an error occured.
358 * Note that the returned ncp is not checked for permissions, though VEXEC
359 * is checked on the directory path leading up to the result. The caller
360 * must call naccess() to check the permissions of the returned leaf.
363 nlookup_simple(const char *str
, enum uio_seg seg
,
364 int niflags
, int *error
)
366 struct nlookupdata nd
;
369 *error
= nlookup_init(&nd
, str
, seg
, niflags
);
371 if ((*error
= nlookup(&nd
)) == 0) {
372 nch
= nd
.nl_nch
; /* keep hold ref from structure */
373 cache_zero(&nd
.nl_nch
); /* and NULL out */
385 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
386 * on return, even if an error occurs. If no error occurs the returned
387 * nl_nch is always referenced and locked, otherwise it may or may not be.
389 * Intermediate directory elements, including the current directory, require
390 * execute (search) permission. nlookup does not examine the access
391 * permissions on the returned element.
393 * If NLC_CREATE is set the last directory must allow node creation,
394 * and an error code of 0 will be returned for a non-existant
395 * target (not ENOENT).
397 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
398 * plus the sticky check is made, and an error code of 0 will be returned
399 * for a non-existant target (not ENOENT).
401 * If NLC_DELETE is set the last directory mut allow node deletion,
402 * plus the sticky check is made.
404 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
405 * of the returned entry. The vnode will be referenced, but not locked,
406 * and will be released by nlookup_done() along with everything else.
409 nlookup(struct nlookupdata
*nd
)
411 struct nlcomponent nlc
;
414 struct nchandle nctmp
;
424 if (KTRPOINT(nd
->nl_td
, KTR_NAMEI
))
425 ktrnamei(nd
->nl_td
->td_lwp
, nd
->nl_path
);
427 bzero(&nlc
, sizeof(nlc
));
430 * Setup for the loop. The current working namecache element is
431 * always at least referenced. We lock it as required, but always
432 * return a locked, resolved namecache entry.
442 * Loop on the path components. At the top of the loop nd->nl_nch
443 * is ref'd and unlocked and represents our current position.
447 * Make sure nl_nch is locked so we can access the vnode, resolution
450 if ((nd
->nl_flags
& NLC_NCPISLOCKED
) == 0) {
451 nd
->nl_flags
|= NLC_NCPISLOCKED
;
452 cache_lock(&nd
->nl_nch
);
456 * Check if the root directory should replace the current
457 * directory. This is done at the start of a translation
458 * or after a symbolic link has been found. In other cases
459 * ptr will never be pointing at a '/'.
464 } while (*ptr
== '/');
465 cache_get(&nd
->nl_rootnch
, &nch
);
466 cache_put(&nd
->nl_nch
);
467 nd
->nl_nch
= nch
; /* remains locked */
470 * Fast-track termination. There is no parent directory of
471 * the root in the same mount from the point of view of
472 * the caller so return EPERM if NLC_REFDVP is specified.
473 * e.g. 'rmdir /' is not allowed.
476 if (nd
->nl_flags
& NLC_REFDVP
)
486 * Check directory search permissions.
489 error
= naccess(&nd
->nl_nch
, NLC_EXEC
, nd
->nl_cred
, &dflags
);
494 * Extract the path component. Path components are limited to
497 nlc
.nlc_nameptr
= ptr
;
498 while (*ptr
&& *ptr
!= '/')
500 nlc
.nlc_namelen
= ptr
- nlc
.nlc_nameptr
;
501 if (nlc
.nlc_namelen
>= 256) {
502 error
= ENAMETOOLONG
;
507 * Lookup the path component in the cache, creating an unresolved
508 * entry if necessary. We have to handle "." and ".." as special
511 * When handling ".." we have to detect a traversal back through a
512 * mount point. If we are at the root, ".." just returns the root.
514 * When handling "." or ".." we also have to recalculate dflags
515 * since our dflags will be for some sub-directory instead of the
518 * This subsection returns a locked, refd 'nch' unless it errors out.
519 * The namecache topology is not allowed to be disconnected, so
520 * encountering a NULL parent will generate EINVAL. This typically
521 * occurs when a directory is removed out from under a process.
523 if (nlc
.nlc_namelen
== 1 && nlc
.nlc_nameptr
[0] == '.') {
524 cache_get(&nd
->nl_nch
, &nch
);
526 } else if (nlc
.nlc_namelen
== 2 &&
527 nlc
.nlc_nameptr
[0] == '.' && nlc
.nlc_nameptr
[1] == '.') {
528 if (nd
->nl_nch
.mount
== nd
->nl_rootnch
.mount
&&
529 nd
->nl_nch
.ncp
== nd
->nl_rootnch
.ncp
532 * ".." at the root returns the root
534 cache_get(&nd
->nl_nch
, &nch
);
537 * Locate the parent ncp. If we are at the root of a
538 * filesystem mount we have to skip to the mounted-on
539 * point in the underlying filesystem.
541 * Expect the parent to always be good since the
542 * mountpoint doesn't go away. XXX hack. cache_get()
543 * requires the ncp to already have a ref as a safety.
546 while (nctmp
.ncp
== nctmp
.mount
->mnt_ncmountpt
.ncp
)
547 nctmp
= nctmp
.mount
->mnt_ncmounton
;
548 nctmp
.ncp
= nctmp
.ncp
->nc_parent
;
549 KKASSERT(nctmp
.ncp
!= NULL
);
551 cache_get(&nctmp
, &nch
);
552 cache_drop(&nctmp
); /* NOTE: zero's nctmp */
557 * Must unlock nl_nch when traversing down the path.
559 cache_unlock(&nd
->nl_nch
);
560 nd
->nl_flags
&= ~NLC_NCPISLOCKED
;
561 nch
= cache_nlookup(&nd
->nl_nch
, &nlc
);
562 while ((error
= cache_resolve(&nch
, nd
->nl_cred
)) == EAGAIN
) {
563 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
564 nch
.ncp
->nc_nlen
, nch
.ncp
->nc_nlen
, nch
.ncp
->nc_name
);
566 nch
= cache_nlookup(&nd
->nl_nch
, &nlc
);
572 * If the last component was "." or ".." our dflags no longer
573 * represents the parent directory and we have to explicitly
576 * Expect the parent to be good since nch is locked.
578 if (wasdotordotdot
&& error
== 0) {
580 if ((par
.ncp
= nch
.ncp
->nc_parent
) != NULL
) {
581 par
.mount
= nch
.mount
;
584 error
= naccess(&par
, 0, nd
->nl_cred
, &dflags
);
588 if (nd
->nl_flags
& NLC_NCPISLOCKED
) {
589 cache_unlock(&nd
->nl_nch
);
590 nd
->nl_flags
&= ~NLC_NCPISLOCKED
;
594 * [end of subsection]
596 * nch is locked and referenced.
597 * nd->nl_nch is unlocked and referenced.
599 * nl_nch must be unlocked or we could chain lock to the root
600 * if a resolve gets stuck (e.g. in NFS).
604 * Resolve the namespace if necessary. The ncp returned by
605 * cache_nlookup() is referenced and locked.
607 * XXX neither '.' nor '..' should return EAGAIN since they were
608 * previously resolved and thus cannot be newly created ncp's.
610 if (nch
.ncp
->nc_flag
& NCF_UNRESOLVED
) {
611 error
= cache_resolve(&nch
, nd
->nl_cred
);
612 KKASSERT(error
!= EAGAIN
);
614 error
= nch
.ncp
->nc_error
;
618 * Early completion. ENOENT is not an error if this is the last
619 * component and NLC_CREATE or NLC_RENAME (rename target) was
620 * requested. Note that ncp->nc_error is left as ENOENT in that
621 * case, which we check later on.
623 * Also handle invalid '.' or '..' components terminating a path
624 * for a create/rename/delete. The standard requires this and pax
625 * pretty stupidly depends on it.
627 for (xptr
= ptr
; *xptr
== '/'; ++xptr
)
630 if (error
== ENOENT
&&
631 (nd
->nl_flags
& (NLC_CREATE
| NLC_RENAME_DST
))
633 if (nd
->nl_flags
& NLC_NFS_RDONLY
) {
636 error
= naccess(&nch
, nd
->nl_flags
| dflags
,
640 if (error
== 0 && wasdotordotdot
&&
641 (nd
->nl_flags
& (NLC_CREATE
| NLC_DELETE
|
642 NLC_RENAME_SRC
| NLC_RENAME_DST
))) {
646 if (nd
->nl_flags
& NLC_CREATE
)
648 else if (nd
->nl_flags
& NLC_DELETE
)
649 error
= (wasdotordotdot
== 1) ? EINVAL
: ENOTEMPTY
;
656 * Early completion on error.
664 * If the element is a symlink and it is either not the last
665 * element or it is the last element and we are allowed to
666 * follow symlinks, resolve the symlink.
668 if ((nch
.ncp
->nc_flag
& NCF_ISSYMLINK
) &&
669 (*ptr
|| (nd
->nl_flags
& NLC_FOLLOW
))
671 if (nd
->nl_loopcnt
++ >= MAXSYMLINKS
) {
676 error
= nreadsymlink(nd
, &nch
, &nlc
);
682 * Concatenate trailing path elements onto the returned symlink.
683 * Note that if the path component (ptr) is not exhausted, it
684 * will being with a '/', so we do not have to add another one.
686 * The symlink may not be empty.
689 if (nlc
.nlc_namelen
== 0 || nlc
.nlc_namelen
+ len
>= MAXPATHLEN
) {
690 error
= nlc
.nlc_namelen
? ENAMETOOLONG
: ENOENT
;
691 objcache_put(namei_oc
, nlc
.nlc_nameptr
);
694 bcopy(ptr
, nlc
.nlc_nameptr
+ nlc
.nlc_namelen
, len
+ 1);
695 if (nd
->nl_flags
& NLC_HASBUF
)
696 objcache_put(namei_oc
, nd
->nl_path
);
697 nd
->nl_path
= nlc
.nlc_nameptr
;
698 nd
->nl_flags
|= NLC_HASBUF
;
702 * Go back up to the top to resolve any initial '/'s in the
709 * If the element is a directory and we are crossing a mount point,
712 while ((nch
.ncp
->nc_flag
& NCF_ISMOUNTPT
) &&
713 (nd
->nl_flags
& NLC_NOCROSSMOUNT
) == 0 &&
714 (mp
= cache_findmount(&nch
)) != NULL
719 cache_get(&mp
->mnt_ncmountpt
, &nch
);
721 if (nch
.ncp
->nc_flag
& NCF_UNRESOLVED
) {
722 while (vfs_busy(mp
, 0))
724 error
= VFS_ROOT(mp
, &tdp
);
728 cache_setvp(&nch
, tdp
);
738 * Skip any slashes to get to the next element. If there
739 * are any slashes at all the current element must be a
740 * directory or, in the create case, intended to become a directory.
741 * If it isn't we break without incrementing ptr and fall through
742 * to the failure case below.
744 while (*ptr
== '/') {
745 if ((nch
.ncp
->nc_flag
& NCF_ISDIR
) == 0 &&
746 !(nd
->nl_flags
& NLC_WILLBEDIR
)
754 * Continuation case: additional elements and the current
755 * element is a directory.
757 if (*ptr
&& (nch
.ncp
->nc_flag
& NCF_ISDIR
)) {
758 cache_drop(&nd
->nl_nch
);
760 KKASSERT((nd
->nl_flags
& NLC_NCPISLOCKED
) == 0);
766 * Failure case: additional elements and the current element
776 * Successful lookup of last element.
778 * Check permissions if the target exists. If the target does not
779 * exist directory permissions were already tested in the early
780 * completion code above.
782 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
783 * if the file is marked append-only, and NLC_STICKY if the directory
784 * containing the file is sticky.
786 if (nch
.ncp
->nc_vp
&& (nd
->nl_flags
& NLC_ALLCHKS
)) {
787 error
= naccess(&nch
, nd
->nl_flags
| dflags
,
796 * Termination: no more elements.
798 * If NLC_REFDVP is set acquire a referenced parent dvp.
800 if (nd
->nl_flags
& NLC_REFDVP
) {
801 cache_lock(&nd
->nl_nch
);
802 error
= cache_vref(&nd
->nl_nch
, nd
->nl_cred
, &nd
->nl_dvp
);
803 cache_unlock(&nd
->nl_nch
);
805 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch
.ncp
);
810 cache_drop(&nd
->nl_nch
);
812 nd
->nl_flags
|= NLC_NCPISLOCKED
;
818 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
819 * (ncp->nc_error will be ENOENT), but we will still return an error
826 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
827 * of continuity in the namecache tree by connecting the ncp related to the
828 * vnode under the mount to the ncp related to the mount's root vnode.
830 * If no error occured a locked, ref'd ncp is stored in *ncpp.
833 nlookup_mp(struct mount
*mp
, struct nchandle
*nch
)
839 cache_get(&mp
->mnt_ncmountpt
, nch
);
840 if (nch
->ncp
->nc_flag
& NCF_UNRESOLVED
) {
841 while (vfs_busy(mp
, 0))
843 error
= VFS_ROOT(mp
, &vp
);
848 cache_setvp(nch
, vp
);
856 * Read the contents of a symlink, allocate a path buffer out of the
857 * namei_oc and initialize the supplied nlcomponent with the result.
859 * If an error occurs no buffer will be allocated or returned in the nlc.
862 nreadsymlink(struct nlookupdata
*nd
, struct nchandle
*nch
,
863 struct nlcomponent
*nlc
)
872 nlc
->nlc_nameptr
= NULL
;
873 nlc
->nlc_namelen
= 0;
874 if (nch
->ncp
->nc_vp
== NULL
)
876 if ((error
= cache_vget(nch
, nd
->nl_cred
, LK_SHARED
, &vp
)) != 0)
878 cp
= objcache_get(namei_oc
, M_WAITOK
);
880 aiov
.iov_len
= MAXPATHLEN
;
881 auio
.uio_iov
= &aiov
;
884 auio
.uio_rw
= UIO_READ
;
885 auio
.uio_segflg
= UIO_SYSSPACE
;
886 auio
.uio_td
= nd
->nl_td
;
887 auio
.uio_resid
= MAXPATHLEN
- 1;
888 error
= VOP_READLINK(vp
, &auio
, nd
->nl_cred
);
891 linklen
= MAXPATHLEN
- 1 - auio
.uio_resid
;
893 linklen
= varsymreplace(cp
, linklen
, MAXPATHLEN
- 1);
895 error
= ENAMETOOLONG
;
900 nlc
->nlc_nameptr
= cp
;
901 nlc
->nlc_namelen
= linklen
;
905 objcache_put(namei_oc
, cp
);
911 * Check access [XXX cache vattr!] [XXX quota]
913 * Generally check the NLC_* access bits. All specified bits must pass
914 * for this function to return 0.
916 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
917 * access, otherwise it must exist. No error is returned in this case.
919 * The file must not exist if NLC_EXCL is specified.
921 * Directory permissions in general are tested for NLC_CREATE if the file
922 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
923 * whether the file exists or not.
925 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
926 * the latter is only tested if the target exists.
928 * The passed ncp must be referenced and locked.
931 naccess(struct nchandle
*nch
, int nflags
, struct ucred
*cred
, int *nflagsp
)
935 struct namecache
*ncp
;
939 ASSERT_NCH_LOCKED(nch
);
941 if (ncp
->nc_flag
& NCF_UNRESOLVED
) {
942 cache_resolve(nch
, cred
);
945 error
= ncp
->nc_error
;
948 * Directory permissions checks. Silently ignore ENOENT if these
949 * tests pass. It isn't an error.
951 * We can safely resolve ncp->nc_parent because ncp is currently
954 if (nflags
& (NLC_CREATE
| NLC_DELETE
| NLC_RENAME_SRC
| NLC_RENAME_DST
)) {
955 if (((nflags
& NLC_CREATE
) && ncp
->nc_vp
== NULL
) ||
956 ((nflags
& NLC_DELETE
) && ncp
->nc_vp
!= NULL
) ||
957 ((nflags
& NLC_RENAME_SRC
) && ncp
->nc_vp
!= NULL
) ||
958 (nflags
& NLC_RENAME_DST
)
962 if ((par
.ncp
= ncp
->nc_parent
) == NULL
) {
965 } else if (error
== 0 || error
== ENOENT
) {
966 par
.mount
= nch
->mount
;
969 error
= naccess(&par
, NLC_WRITE
, cred
, NULL
);
976 * NLC_EXCL check. Target file must not exist.
978 if (error
== 0 && (nflags
& NLC_EXCL
) && ncp
->nc_vp
!= NULL
)
982 * Get the vnode attributes so we can do the rest of our checks.
984 * NOTE: We only call naccess_va() if the target exists.
987 error
= cache_vget(nch
, cred
, LK_SHARED
, &vp
);
988 if (error
== ENOENT
) {
990 * Silently zero-out ENOENT if creating or renaming
991 * (rename target). It isn't an error.
993 if (nflags
& (NLC_CREATE
| NLC_RENAME_DST
))
995 } else if (error
== 0) {
997 * Get the vnode attributes and check for illegal O_TRUNC
998 * requests and read-only mounts.
1000 * NOTE: You can still open devices on read-only mounts for
1003 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
1004 * check on the parent directory above.
1006 * XXX cache the va in the namecache or in the vnode
1008 error
= VOP_GETATTR(vp
, &va
);
1009 if (error
== 0 && (nflags
& NLC_TRUNCATE
)) {
1010 switch(va
.va_type
) {
1025 if (error
== 0 && (nflags
& NLC_WRITE
) && vp
->v_mount
&&
1026 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
)
1028 switch(va
.va_type
) {
1042 * Check permissions based on file attributes. The passed
1043 * flags (*nflagsp) are modified with feedback based on
1044 * special attributes and requirements.
1048 * Adjust the returned (*nflagsp) if non-NULL.
1051 if ((va
.va_mode
& VSVTX
) && va
.va_uid
!= cred
->cr_uid
)
1052 *nflagsp
|= NLC_STICKY
;
1053 if (va
.va_flags
& APPEND
)
1054 *nflagsp
|= NLC_APPENDONLY
;
1055 if (va
.va_flags
& IMMUTABLE
)
1056 *nflagsp
|= NLC_IMMUTABLE
;
1060 * Track swapcache management flags in the namecache.
1062 * Calculate the flags based on the current vattr info
1063 * and recalculate the inherited flags from the parent
1064 * (the original cache linkage may have occurred without
1065 * getattrs and thus have stale flags).
1068 if (va
.va_flags
& SF_NOCACHE
)
1069 cflags
|= NCF_SF_NOCACHE
;
1070 if (va
.va_flags
& UF_CACHE
)
1071 cflags
|= NCF_UF_CACHE
;
1072 if (ncp
->nc_parent
) {
1073 if (ncp
->nc_parent
->nc_flag
&
1074 (NCF_SF_NOCACHE
| NCF_SF_PNOCACHE
)) {
1075 cflags
|= NCF_SF_PNOCACHE
;
1077 if (ncp
->nc_parent
->nc_flag
&
1078 (NCF_UF_CACHE
| NCF_UF_PCACHE
)) {
1079 cflags
|= NCF_UF_PCACHE
;
1082 ncp
->nc_flag
&= ~(NCF_SF_NOCACHE
| NCF_UF_CACHE
|
1083 NCF_SF_PNOCACHE
| NCF_UF_PCACHE
);
1084 ncp
->nc_flag
|= cflags
;
1087 * Process general access.
1089 error
= naccess_va(&va
, nflags
, cred
);
1097 * Check the requested access against the given vattr using cred.
1100 naccess_va(struct vattr
*va
, int nflags
, struct ucred
*cred
)
1106 * Test the immutable bit. Creations, deletions, renames (source
1107 * or destination) are not allowed. chown/chmod/other is also not
1108 * allowed but is handled by SETATTR. Hardlinks to the immutable
1111 * If the directory is set to immutable then creations, deletions,
1112 * renames (source or dest) and hardlinks to files within the directory
1113 * are not allowed, and regular files opened through the directory may
1114 * not be written to or truncated (unless a special device).
1116 * NOTE! New hardlinks to immutable files work but new hardlinks to
1117 * files, immutable or not, sitting inside an immutable directory are
1118 * not allowed. As always if the file is hardlinked via some other
1119 * path additional hardlinks may be possible even if the file is marked
1120 * immutable. The sysop needs to create a closure by checking the hard
1121 * link count. Once closure is achieved you are good, and security
1122 * scripts should check link counts anyway.
1124 * Writes and truncations are only allowed on special devices.
1126 if ((va
->va_flags
& IMMUTABLE
) || (nflags
& NLC_IMMUTABLE
)) {
1127 if ((nflags
& NLC_IMMUTABLE
) && (nflags
& NLC_HLINK
))
1129 if (nflags
& (NLC_CREATE
| NLC_DELETE
|
1130 NLC_RENAME_SRC
| NLC_RENAME_DST
)) {
1133 if (nflags
& (NLC_WRITE
| NLC_TRUNCATE
)) {
1134 switch(va
->va_type
) {
1148 * Test the no-unlink and append-only bits for opens, rename targets,
1149 * and deletions. These bits are not tested for creations or
1152 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1153 * If you do not wish this you must also set NOUNLINK.
1155 * If the governing directory is marked APPEND-only it implies
1156 * NOUNLINK for all entries in the directory.
1158 if (((va
->va_flags
& NOUNLINK
) || (nflags
& NLC_APPENDONLY
)) &&
1159 (nflags
& (NLC_DELETE
| NLC_RENAME_SRC
| NLC_RENAME_DST
))
1165 * A file marked append-only may not be deleted but can be renamed.
1167 if ((va
->va_flags
& APPEND
) &&
1168 (nflags
& (NLC_DELETE
| NLC_RENAME_DST
))
1174 * A file marked append-only which is opened for writing must also
1175 * be opened O_APPEND.
1177 if ((va
->va_flags
& APPEND
) && (nflags
& (NLC_OPEN
| NLC_TRUNCATE
))) {
1178 if (nflags
& NLC_TRUNCATE
)
1180 if ((nflags
& (NLC_OPEN
| NLC_WRITE
)) == (NLC_OPEN
| NLC_WRITE
)) {
1181 if ((nflags
& NLC_APPEND
) == 0)
1187 * root gets universal access
1189 if (cred
->cr_uid
== 0)
1193 * Check owner perms.
1195 * If NLC_OWN is set the owner of the file is allowed no matter when
1196 * the owner-mode bits say (utimes).
1199 if (nflags
& NLC_READ
)
1201 if (nflags
& NLC_WRITE
)
1203 if (nflags
& NLC_EXEC
)
1206 if (cred
->cr_uid
== va
->va_uid
) {
1207 if ((nflags
& NLC_OWN
) == 0) {
1208 if ((vmode
& va
->va_mode
) != vmode
)
1215 * If NLC_STICKY is set only the owner may delete or rename a file.
1216 * This bit is typically set on /tmp.
1218 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1219 * the specific delete or rename case. For deletions and renames we
1220 * usually just care about directory permissions, not file permissions.
1222 if ((nflags
& NLC_STICKY
) &&
1223 (nflags
& (NLC_RENAME_SRC
| NLC_RENAME_DST
| NLC_DELETE
))) {
1231 for (i
= 0; i
< cred
->cr_ngroups
; ++i
) {
1232 if (va
->va_gid
== cred
->cr_groups
[i
]) {
1233 if ((vmode
& va
->va_mode
) != vmode
)
1243 if ((vmode
& va
->va_mode
) != vmode
)