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 * Set a different credential; this credential will be used by future
234 * operations performed on nd.nl_open_vp and nlookupdata structure.
237 nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
239 KKASSERT(nd->nl_cred != NULL);
241 if (nd->nl_cred != cred) {
249 * Cleanup a nlookupdata structure after we are through with it. This may
250 * be called on any nlookupdata structure initialized with nlookup_init().
251 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
252 * returns an error, even if as a consumer you believe you have taken all
253 * dynamic elements out of the nlookupdata structure.
256 nlookup_done(struct nlookupdata *nd)
258 if (nd->nl_nch.ncp) {
259 if (nd->nl_flags & NLC_NCPISLOCKED) {
260 nd->nl_flags &= ~NLC_NCPISLOCKED;
261 cache_unlock(&nd->nl_nch);
263 cache_drop(&nd->nl_nch); /* NULL's out the nch */
265 if (nd->nl_rootnch.ncp)
266 cache_drop(&nd->nl_rootnch);
267 if (nd->nl_jailnch.ncp)
268 cache_drop(&nd->nl_jailnch);
269 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
270 objcache_put(namei_oc, nd->nl_path);
277 if (nd->nl_open_vp) {
278 if (nd->nl_flags & NLC_LOCKVP) {
279 vn_unlock(nd->nl_open_vp);
280 nd->nl_flags &= ~NLC_LOCKVP;
282 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
283 nd->nl_open_vp = NULL;
289 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
293 * Works similarly to nlookup_done() when nd initialized with
297 nlookup_done_at(struct nlookupdata *nd, struct file *fp)
305 nlookup_zero(struct nlookupdata *nd)
307 bzero(nd, sizeof(struct nlookupdata));
311 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
312 * if an error occured.
314 * Note that the returned ncp is not checked for permissions, though VEXEC
315 * is checked on the directory path leading up to the result. The caller
316 * must call naccess() to check the permissions of the returned leaf.
319 nlookup_simple(const char *str, enum uio_seg seg,
320 int niflags, int *error)
322 struct nlookupdata nd;
325 *error = nlookup_init(&nd, str, seg, niflags);
327 if ((*error = nlookup(&nd)) == 0) {
328 nch = nd.nl_nch; /* keep hold ref from structure */
329 cache_zero(&nd.nl_nch); /* and NULL out */
341 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
342 * on return, even if an error occurs. If no error occurs the returned
343 * nl_nch is always referenced and locked, otherwise it may or may not be.
345 * Intermediate directory elements, including the current directory, require
346 * execute (search) permission. nlookup does not examine the access
347 * permissions on the returned element.
349 * If NLC_CREATE is set the last directory must allow node creation,
350 * and an error code of 0 will be returned for a non-existant
351 * target (not ENOENT).
353 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
354 * plus the sticky check is made, and an error code of 0 will be returned
355 * for a non-existant target (not ENOENT).
357 * If NLC_DELETE is set the last directory mut allow node deletion,
358 * plus the sticky check is made.
360 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
361 * of the returned entry. The vnode will be referenced, but not locked,
362 * and will be released by nlookup_done() along with everything else.
365 nlookup(struct nlookupdata *nd)
367 struct nlcomponent nlc;
370 struct nchandle nctmp;
380 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
381 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
383 bzero(&nlc, sizeof(nlc));
386 * Setup for the loop. The current working namecache element is
387 * always at least referenced. We lock it as required, but always
388 * return a locked, resolved namecache entry.
398 * Loop on the path components. At the top of the loop nd->nl_nch
399 * is ref'd and unlocked and represents our current position.
403 * Make sure nl_nch is locked so we can access the vnode, resolution
406 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) {
407 nd->nl_flags |= NLC_NCPISLOCKED;
408 cache_lock(&nd->nl_nch);
412 * Check if the root directory should replace the current
413 * directory. This is done at the start of a translation
414 * or after a symbolic link has been found. In other cases
415 * ptr will never be pointing at a '/'.
420 } while (*ptr == '/');
421 cache_get(&nd->nl_rootnch, &nch);
422 cache_put(&nd->nl_nch);
423 nd->nl_nch = nch; /* remains locked */
426 * Fast-track termination. There is no parent directory of
427 * the root in the same mount from the point of view of
428 * the caller so return EPERM if NLC_REFDVP is specified.
429 * e.g. 'rmdir /' is not allowed.
432 if (nd->nl_flags & NLC_REFDVP)
442 * Check directory search permissions.
445 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags);
450 * Extract the path component. Path components are limited to
453 nlc.nlc_nameptr = ptr;
454 while (*ptr && *ptr != '/')
456 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
457 if (nlc.nlc_namelen >= 256) {
458 error = ENAMETOOLONG;
463 * Lookup the path component in the cache, creating an unresolved
464 * entry if necessary. We have to handle "." and ".." as special
467 * When handling ".." we have to detect a traversal back through a
468 * mount point. If we are at the root, ".." just returns the root.
470 * When handling "." or ".." we also have to recalculate dflags
471 * since our dflags will be for some sub-directory instead of the
474 * This subsection returns a locked, refd 'nch' unless it errors out.
475 * The namecache topology is not allowed to be disconnected, so
476 * encountering a NULL parent will generate EINVAL. This typically
477 * occurs when a directory is removed out from under a process.
479 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
480 cache_get(&nd->nl_nch, &nch);
482 } else if (nlc.nlc_namelen == 2 &&
483 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
484 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
485 nd->nl_nch.ncp == nd->nl_rootnch.ncp
488 * ".." at the root returns the root
490 cache_get(&nd->nl_nch, &nch);
493 * Locate the parent ncp. If we are at the root of a
494 * filesystem mount we have to skip to the mounted-on
495 * point in the underlying filesystem.
497 * Expect the parent to always be good since the
498 * mountpoint doesn't go away. XXX hack. cache_get()
499 * requires the ncp to already have a ref as a safety.
502 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
503 nctmp = nctmp.mount->mnt_ncmounton;
504 nctmp.ncp = nctmp.ncp->nc_parent;
505 KKASSERT(nctmp.ncp != NULL);
507 cache_get(&nctmp, &nch);
508 cache_drop(&nctmp); /* NOTE: zero's nctmp */
513 * Must unlock nl_nch when traversing down the path.
515 cache_unlock(&nd->nl_nch);
516 nd->nl_flags &= ~NLC_NCPISLOCKED;
517 nch = cache_nlookup(&nd->nl_nch, &nlc);
518 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
519 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
520 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
522 nch = cache_nlookup(&nd->nl_nch, &nlc);
528 * If the last component was "." or ".." our dflags no longer
529 * represents the parent directory and we have to explicitly
532 * Expect the parent to be good since nch is locked.
534 if (wasdotordotdot && error == 0) {
536 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
537 par.mount = nch.mount;
540 error = naccess(&par, 0, nd->nl_cred, &dflags);
544 if (nd->nl_flags & NLC_NCPISLOCKED) {
545 cache_unlock(&nd->nl_nch);
546 nd->nl_flags &= ~NLC_NCPISLOCKED;
550 * [end of subsection]
552 * nch is locked and referenced.
553 * nd->nl_nch is unlocked and referenced.
555 * nl_nch must be unlocked or we could chain lock to the root
556 * if a resolve gets stuck (e.g. in NFS).
560 * Resolve the namespace if necessary. The ncp returned by
561 * cache_nlookup() is referenced and locked.
563 * XXX neither '.' nor '..' should return EAGAIN since they were
564 * previously resolved and thus cannot be newly created ncp's.
566 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
567 error = cache_resolve(&nch, nd->nl_cred);
568 KKASSERT(error != EAGAIN);
570 error = nch.ncp->nc_error;
574 * Early completion. ENOENT is not an error if this is the last
575 * component and NLC_CREATE or NLC_RENAME (rename target) was
576 * requested. Note that ncp->nc_error is left as ENOENT in that
577 * case, which we check later on.
579 * Also handle invalid '.' or '..' components terminating a path
580 * for a create/rename/delete. The standard requires this and pax
581 * pretty stupidly depends on it.
583 for (xptr = ptr; *xptr == '/'; ++xptr)
586 if (error == ENOENT &&
587 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
589 if (nd->nl_flags & NLC_NFS_RDONLY) {
592 error = naccess(&nch, nd->nl_flags | dflags,
596 if (error == 0 && wasdotordotdot &&
597 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
598 NLC_RENAME_SRC | NLC_RENAME_DST))) {
602 if (nd->nl_flags & NLC_CREATE)
604 else if (nd->nl_flags & NLC_DELETE)
605 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
612 * Early completion on error.
620 * If the element is a symlink and it is either not the last
621 * element or it is the last element and we are allowed to
622 * follow symlinks, resolve the symlink.
624 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
625 (*ptr || (nd->nl_flags & NLC_FOLLOW))
627 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
632 error = nreadsymlink(nd, &nch, &nlc);
638 * Concatenate trailing path elements onto the returned symlink.
639 * Note that if the path component (ptr) is not exhausted, it
640 * will being with a '/', so we do not have to add another one.
642 * The symlink may not be empty.
645 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
646 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
647 objcache_put(namei_oc, nlc.nlc_nameptr);
650 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
651 if (nd->nl_flags & NLC_HASBUF)
652 objcache_put(namei_oc, nd->nl_path);
653 nd->nl_path = nlc.nlc_nameptr;
654 nd->nl_flags |= NLC_HASBUF;
658 * Go back up to the top to resolve any initial '/'s in the
665 * If the element is a directory and we are crossing a mount point,
668 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
669 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
670 (mp = cache_findmount(&nch)) != NULL
675 cache_get(&mp->mnt_ncmountpt, &nch);
677 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
678 while (vfs_busy(mp, 0))
680 error = VFS_ROOT(mp, &tdp);
684 cache_setvp(&nch, tdp);
694 * Skip any slashes to get to the next element. If there
695 * are any slashes at all the current element must be a
696 * directory or, in the create case, intended to become a directory.
697 * If it isn't we break without incrementing ptr and fall through
698 * to the failure case below.
700 while (*ptr == '/') {
701 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
702 !(nd->nl_flags & NLC_WILLBEDIR)
710 * Continuation case: additional elements and the current
711 * element is a directory.
713 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
714 cache_drop(&nd->nl_nch);
716 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
722 * Failure case: additional elements and the current element
732 * Successful lookup of last element.
734 * Check permissions if the target exists. If the target does not
735 * exist directory permissions were already tested in the early
736 * completion code above.
738 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
739 * if the file is marked append-only, and NLC_STICKY if the directory
740 * containing the file is sticky.
742 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
743 error = naccess(&nch, nd->nl_flags | dflags,
752 * Termination: no more elements.
754 * If NLC_REFDVP is set acquire a referenced parent dvp.
756 if (nd->nl_flags & NLC_REFDVP) {
757 cache_lock(&nd->nl_nch);
758 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
759 cache_unlock(&nd->nl_nch);
761 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
766 cache_drop(&nd->nl_nch);
768 nd->nl_flags |= NLC_NCPISLOCKED;
774 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
775 * (ncp->nc_error will be ENOENT), but we will still return an error
782 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
783 * of continuity in the namecache tree by connecting the ncp related to the
784 * vnode under the mount to the ncp related to the mount's root vnode.
786 * If no error occured a locked, ref'd ncp is stored in *ncpp.
789 nlookup_mp(struct mount *mp, struct nchandle *nch)
795 cache_get(&mp->mnt_ncmountpt, nch);
796 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
797 while (vfs_busy(mp, 0))
799 error = VFS_ROOT(mp, &vp);
804 cache_setvp(nch, vp);
812 * Read the contents of a symlink, allocate a path buffer out of the
813 * namei_oc and initialize the supplied nlcomponent with the result.
815 * If an error occurs no buffer will be allocated or returned in the nlc.
818 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
819 struct nlcomponent *nlc)
828 nlc->nlc_nameptr = NULL;
829 nlc->nlc_namelen = 0;
830 if (nch->ncp->nc_vp == NULL)
832 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
834 cp = objcache_get(namei_oc, M_WAITOK);
836 aiov.iov_len = MAXPATHLEN;
837 auio.uio_iov = &aiov;
840 auio.uio_rw = UIO_READ;
841 auio.uio_segflg = UIO_SYSSPACE;
842 auio.uio_td = nd->nl_td;
843 auio.uio_resid = MAXPATHLEN - 1;
844 error = VOP_READLINK(vp, &auio, nd->nl_cred);
847 linklen = MAXPATHLEN - 1 - auio.uio_resid;
849 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
851 error = ENAMETOOLONG;
856 nlc->nlc_nameptr = cp;
857 nlc->nlc_namelen = linklen;
861 objcache_put(namei_oc, cp);
867 * Check access [XXX cache vattr!] [XXX quota]
869 * Generally check the NLC_* access bits. All specified bits must pass
870 * for this function to return 0.
872 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
873 * access, otherwise it must exist. No error is returned in this case.
875 * The file must not exist if NLC_EXCL is specified.
877 * Directory permissions in general are tested for NLC_CREATE if the file
878 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
879 * whether the file exists or not.
881 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
882 * the latter is only tested if the target exists.
884 * The passed ncp must be referenced and locked.
887 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
891 struct namecache *ncp;
895 ASSERT_NCH_LOCKED(nch);
897 if (ncp->nc_flag & NCF_UNRESOLVED) {
898 cache_resolve(nch, cred);
901 error = ncp->nc_error;
904 * Directory permissions checks. Silently ignore ENOENT if these
905 * tests pass. It isn't an error.
907 * We can safely resolve ncp->nc_parent because ncp is currently
910 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
911 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) ||
912 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) ||
913 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) ||
914 (nflags & NLC_RENAME_DST)
918 if ((par.ncp = ncp->nc_parent) == NULL) {
921 } else if (error == 0 || error == ENOENT) {
922 par.mount = nch->mount;
925 error = naccess(&par, NLC_WRITE, cred, NULL);
932 * NLC_EXCL check. Target file must not exist.
934 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL)
938 * Get the vnode attributes so we can do the rest of our checks.
940 * NOTE: We only call naccess_va() if the target exists.
943 error = cache_vget(nch, cred, LK_SHARED, &vp);
944 if (error == ENOENT) {
946 * Silently zero-out ENOENT if creating or renaming
947 * (rename target). It isn't an error.
949 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
951 } else if (error == 0) {
953 * Get the vnode attributes and check for illegal O_TRUNC
954 * requests and read-only mounts.
956 * NOTE: You can still open devices on read-only mounts for
959 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
960 * check on the parent directory above.
962 * XXX cache the va in the namecache or in the vnode
964 error = VOP_GETATTR(vp, &va);
965 if (error == 0 && (nflags & NLC_TRUNCATE)) {
981 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
982 (vp->v_mount->mnt_flag & MNT_RDONLY)
998 * Check permissions based on file attributes. The passed
999 * flags (*nflagsp) are modified with feedback based on
1000 * special attributes and requirements.
1004 * Adjust the returned (*nflagsp) if non-NULL.
1007 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
1008 *nflagsp |= NLC_STICKY;
1009 if (va.va_flags & APPEND)
1010 *nflagsp |= NLC_APPENDONLY;
1011 if (va.va_flags & IMMUTABLE)
1012 *nflagsp |= NLC_IMMUTABLE;
1016 * Track swapcache management flags in the namecache.
1018 * Calculate the flags based on the current vattr info
1019 * and recalculate the inherited flags from the parent
1020 * (the original cache linkage may have occurred without
1021 * getattrs and thus have stale flags).
1024 if (va.va_flags & SF_NOCACHE)
1025 cflags |= NCF_SF_NOCACHE;
1026 if (va.va_flags & UF_CACHE)
1027 cflags |= NCF_UF_CACHE;
1028 if (ncp->nc_parent) {
1029 if (ncp->nc_parent->nc_flag &
1030 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) {
1031 cflags |= NCF_SF_PNOCACHE;
1033 if (ncp->nc_parent->nc_flag &
1034 (NCF_UF_CACHE | NCF_UF_PCACHE)) {
1035 cflags |= NCF_UF_PCACHE;
1038 ncp->nc_flag &= ~(NCF_SF_NOCACHE | NCF_UF_CACHE |
1039 NCF_SF_PNOCACHE | NCF_UF_PCACHE);
1040 ncp->nc_flag |= cflags;
1043 * Process general access.
1045 error = naccess_va(&va, nflags, cred);
1053 * Check the requested access against the given vattr using cred.
1056 naccess_va(struct vattr *va, int nflags, struct ucred *cred)
1062 * Test the immutable bit. Creations, deletions, renames (source
1063 * or destination) are not allowed. chown/chmod/other is also not
1064 * allowed but is handled by SETATTR. Hardlinks to the immutable
1067 * If the directory is set to immutable then creations, deletions,
1068 * renames (source or dest) and hardlinks to files within the directory
1069 * are not allowed, and regular files opened through the directory may
1070 * not be written to or truncated (unless a special device).
1072 * NOTE! New hardlinks to immutable files work but new hardlinks to
1073 * files, immutable or not, sitting inside an immutable directory are
1074 * not allowed. As always if the file is hardlinked via some other
1075 * path additional hardlinks may be possible even if the file is marked
1076 * immutable. The sysop needs to create a closure by checking the hard
1077 * link count. Once closure is achieved you are good, and security
1078 * scripts should check link counts anyway.
1080 * Writes and truncations are only allowed on special devices.
1082 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1083 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1085 if (nflags & (NLC_CREATE | NLC_DELETE |
1086 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1089 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1090 switch(va->va_type) {
1104 * Test the no-unlink and append-only bits for opens, rename targets,
1105 * and deletions. These bits are not tested for creations or
1108 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1109 * If you do not wish this you must also set NOUNLINK.
1111 * If the governing directory is marked APPEND-only it implies
1112 * NOUNLINK for all entries in the directory.
1114 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1115 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1121 * A file marked append-only may not be deleted but can be renamed.
1123 if ((va->va_flags & APPEND) &&
1124 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1130 * A file marked append-only which is opened for writing must also
1131 * be opened O_APPEND.
1133 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1134 if (nflags & NLC_TRUNCATE)
1136 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1137 if ((nflags & NLC_APPEND) == 0)
1143 * root gets universal access
1145 if (cred->cr_uid == 0)
1149 * Check owner perms.
1151 * If NLC_OWN is set the owner of the file is allowed no matter when
1152 * the owner-mode bits say (utimes).
1155 if (nflags & NLC_READ)
1157 if (nflags & NLC_WRITE)
1159 if (nflags & NLC_EXEC)
1162 if (cred->cr_uid == va->va_uid) {
1163 if ((nflags & NLC_OWN) == 0) {
1164 if ((vmode & va->va_mode) != vmode)
1171 * If NLC_STICKY is set only the owner may delete or rename a file.
1172 * This bit is typically set on /tmp.
1174 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1175 * the specific delete or rename case. For deletions and renames we
1176 * usually just care about directory permissions, not file permissions.
1178 if ((nflags & NLC_STICKY) &&
1179 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
1187 for (i = 0; i < cred->cr_ngroups; ++i) {
1188 if (va->va_gid == cred->cr_groups[i]) {
1189 if ((vmode & va->va_mode) != vmode)
1199 if ((vmode & va->va_mode) != vmode)