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>
70 #include <sys/ktrace.h>
74 * Initialize a nlookup() structure, early error return for copyin faults
75 * or a degenerate empty string (which is not allowed).
77 * The first process proc0's credentials are used if the calling thread
78 * is not associated with a process context.
81 nlookup_init(struct nlookupdata *nd,
82 const char *path, enum uio_seg seg, int flags)
93 * note: the pathlen set by copy*str() includes the terminating \0.
95 bzero(nd, sizeof(struct nlookupdata));
96 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
97 nd->nl_flags |= NLC_HASBUF;
98 if (seg == UIO_SYSSPACE)
99 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
101 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
104 * Don't allow empty pathnames.
105 * POSIX.1 requirement: "" is not a vaild file name.
107 if (error == 0 && pathlen <= 1)
112 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch);
113 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch);
114 if (p->p_fd->fd_njdir.ncp)
115 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch);
116 nd->nl_cred = crhold(p->p_ucred);
118 cache_copy(&rootnch, &nd->nl_nch);
119 cache_copy(&nd->nl_nch, &nd->nl_rootnch);
120 cache_copy(&nd->nl_nch, &nd->nl_jailnch);
121 nd->nl_cred = crhold(proc0.p_ucred);
124 nd->nl_flags |= flags;
132 * This works similarly to nlookup_init() but does not assume a process
133 * context. rootnch is always chosen for the root directory and the cred
134 * and starting directory are supplied in arguments.
137 nlookup_init_raw(struct nlookupdata *nd,
138 const char *path, enum uio_seg seg, int flags,
139 struct ucred *cred, struct nchandle *ncstart)
147 bzero(nd, sizeof(struct nlookupdata));
148 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
149 nd->nl_flags |= NLC_HASBUF;
150 if (seg == UIO_SYSSPACE)
151 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
153 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
156 * Don't allow empty pathnames.
157 * POSIX.1 requirement: "" is not a vaild file name.
159 if (error == 0 && pathlen <= 1)
163 cache_copy(ncstart, &nd->nl_nch);
164 cache_copy(&rootnch, &nd->nl_rootnch);
165 cache_copy(&rootnch, &nd->nl_jailnch);
166 nd->nl_cred = crhold(cred);
168 nd->nl_flags |= flags;
176 * Set a different credential; this credential will be used by future
177 * operations performed on nd.nl_open_vp and nlookupdata structure.
180 nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
182 KKASSERT(nd->nl_cred != NULL);
184 if (nd->nl_cred != cred) {
192 * Cleanup a nlookupdata structure after we are through with it. This may
193 * be called on any nlookupdata structure initialized with nlookup_init().
194 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
195 * returns an error, even if as a consumer you believe you have taken all
196 * dynamic elements out of the nlookupdata structure.
199 nlookup_done(struct nlookupdata *nd)
201 if (nd->nl_nch.ncp) {
202 if (nd->nl_flags & NLC_NCPISLOCKED) {
203 nd->nl_flags &= ~NLC_NCPISLOCKED;
204 cache_unlock(&nd->nl_nch);
206 cache_drop(&nd->nl_nch);
208 if (nd->nl_rootnch.ncp)
209 cache_drop(&nd->nl_rootnch);
210 if (nd->nl_jailnch.ncp)
211 cache_drop(&nd->nl_jailnch);
212 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
213 objcache_put(namei_oc, nd->nl_path);
220 if (nd->nl_open_vp) {
221 if (nd->nl_flags & NLC_LOCKVP) {
222 vn_unlock(nd->nl_open_vp);
223 nd->nl_flags &= ~NLC_LOCKVP;
225 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
226 nd->nl_open_vp = NULL;
232 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
236 nlookup_zero(struct nlookupdata *nd)
238 bzero(nd, sizeof(struct nlookupdata));
242 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
243 * if an error occured.
245 * Note that the returned ncp is not checked for permissions, though VEXEC
246 * is checked on the directory path leading up to the result. The caller
247 * must call naccess() to check the permissions of the returned leaf.
250 nlookup_simple(const char *str, enum uio_seg seg,
251 int niflags, int *error)
253 struct nlookupdata nd;
256 *error = nlookup_init(&nd, str, seg, niflags);
258 if ((*error = nlookup(&nd)) == 0) {
259 nch = nd.nl_nch; /* keep hold ref from structure */
260 cache_zero(&nd.nl_nch); /* and NULL out */
272 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
273 * on return, even if an error occurs. If no error occurs the returned
274 * nl_nch is always referenced and locked, otherwise it may or may not be.
276 * Intermediate directory elements, including the current directory, require
277 * execute (search) permission. nlookup does not examine the access
278 * permissions on the returned element.
280 * If NLC_CREATE is set the last directory must allow node creation,
281 * and an error code of 0 will be returned for a non-existant
282 * target (not ENOENT).
284 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
285 * plus the sticky check is made, and an error code of 0 will be returned
286 * for a non-existant target (not ENOENT).
288 * If NLC_DELETE is set the last directory mut allow node deletion,
289 * plus the sticky check is made.
291 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
292 * of the returned entry. The vnode will be referenced, but not locked,
293 * and will be released by nlookup_done() along with everything else.
296 nlookup(struct nlookupdata *nd)
298 struct nlcomponent nlc;
310 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
311 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
313 bzero(&nlc, sizeof(nlc));
316 * Setup for the loop. The current working namecache element must
317 * be in a refd + unlocked state. This typically the case on entry except
318 * when stringing nlookup()'s along in a chain, since nlookup() always
319 * returns nl_nch in a locked state.
322 if (nd->nl_flags & NLC_NCPISLOCKED) {
323 nd->nl_flags &= ~NLC_NCPISLOCKED;
324 cache_unlock(&nd->nl_nch);
333 * Loop on the path components. At the top of the loop nd->nl_nch
334 * is ref'd and unlocked and represents our current position.
338 * Check if the root directory should replace the current
339 * directory. This is done at the start of a translation
340 * or after a symbolic link has been found. In other cases
341 * ptr will never be pointing at a '/'.
346 } while (*ptr == '/');
347 cache_copy(&nd->nl_rootnch, &nch);
348 cache_drop(&nd->nl_nch);
352 * Fast-track termination. There is no parent directory of
353 * the root in the same mount from the point of view of
354 * the caller so return EPERM if NLC_REFDVP is specified.
355 * e.g. 'rmdir /' is not allowed.
358 if (nd->nl_flags & NLC_REFDVP) {
361 cache_lock(&nd->nl_nch);
362 nd->nl_flags |= NLC_NCPISLOCKED;
371 * Check directory search permissions.
374 if ((error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags)) != 0)
378 * Extract the path component
380 nlc.nlc_nameptr = ptr;
381 while (*ptr && *ptr != '/')
383 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
386 * Lookup the path component in the cache, creating an unresolved
387 * entry if necessary. We have to handle "." and ".." as special
390 * When handling ".." we have to detect a traversal back through a
391 * mount point. If we are at the root, ".." just returns the root.
393 * When handling "." or ".." we also have to recalculate dflags
394 * since our dflags will be for some sub-directory instead of the
397 * This subsection returns a locked, refd 'nch' unless it errors out.
398 * The namecache topology is not allowed to be disconnected, so
399 * encountering a NULL parent will generate EINVAL. This typically
400 * occurs when a directory is removed out from under a process.
402 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
403 cache_get(&nd->nl_nch, &nch);
405 } else if (nlc.nlc_namelen == 2 &&
406 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
407 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
408 nd->nl_nch.ncp == nd->nl_rootnch.ncp
411 * ".." at the root returns the root
413 cache_get(&nd->nl_nch, &nch);
416 * Locate the parent ncp. If we are at the root of a
417 * filesystem mount we have to skip to the mounted-on
418 * point in the underlying filesystem.
421 while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
422 nch = nch.mount->mnt_ncmounton;
423 nch.ncp = nch.ncp->nc_parent;
424 KKASSERT(nch.ncp != NULL);
425 cache_get(&nch, &nch);
429 nch = cache_nlookup(&nd->nl_nch, &nlc);
430 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
431 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
432 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
434 nch = cache_nlookup(&nd->nl_nch, &nlc);
440 * If the last component was "." or ".." our dflags no longer
441 * represents the parent directory and we have to explicitly
444 if (wasdotordotdot && error == 0) {
446 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
447 par.mount = nch.mount;
450 error = naccess(&par, 0, nd->nl_cred, &dflags);
456 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
460 * Resolve the namespace if necessary. The ncp returned by
461 * cache_nlookup() is referenced and locked.
463 * XXX neither '.' nor '..' should return EAGAIN since they were
464 * previously resolved and thus cannot be newly created ncp's.
466 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
467 error = cache_resolve(&nch, nd->nl_cred);
468 KKASSERT(error != EAGAIN);
470 error = nch.ncp->nc_error;
474 * Early completion. ENOENT is not an error if this is the last
475 * component and NLC_CREATE or NLC_RENAME (rename target) was
476 * requested. Note that ncp->nc_error is left as ENOENT in that
477 * case, which we check later on.
479 * Also handle invalid '.' or '..' components terminating a path
480 * for a create/rename/delete. The standard requires this and pax
481 * pretty stupidly depends on it.
483 for (xptr = ptr; *xptr == '/'; ++xptr)
486 if (error == ENOENT &&
487 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
489 if (nd->nl_flags & NLC_NFS_RDONLY) {
492 error = naccess(&nch, nd->nl_flags | dflags,
496 if (error == 0 && wasdotordotdot &&
497 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
498 NLC_RENAME_SRC | NLC_RENAME_DST))) {
504 * Early completion on error.
512 * If the element is a symlink and it is either not the last
513 * element or it is the last element and we are allowed to
514 * follow symlinks, resolve the symlink.
516 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
517 (*ptr || (nd->nl_flags & NLC_FOLLOW))
519 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
524 error = nreadsymlink(nd, &nch, &nlc);
530 * Concatenate trailing path elements onto the returned symlink.
531 * Note that if the path component (ptr) is not exhausted, it
532 * will being with a '/', so we do not have to add another one.
534 * The symlink may not be empty.
537 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
538 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
539 objcache_put(namei_oc, nlc.nlc_nameptr);
542 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
543 if (nd->nl_flags & NLC_HASBUF)
544 objcache_put(namei_oc, nd->nl_path);
545 nd->nl_path = nlc.nlc_nameptr;
546 nd->nl_flags |= NLC_HASBUF;
550 * Go back up to the top to resolve any initial '/'s in the
557 * If the element is a directory and we are crossing a mount point,
560 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
561 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
562 (mp = cache_findmount(&nch)) != NULL
567 cache_get(&mp->mnt_ncmountpt, &nch);
569 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
570 while (vfs_busy(mp, 0))
572 error = VFS_ROOT(mp, &tdp);
576 cache_setvp(&nch, tdp);
586 * Skip any slashes to get to the next element. If there
587 * are any slashes at all the current element must be a
588 * directory or, in the create case, intended to become a directory.
589 * If it isn't we break without incrementing ptr and fall through
590 * to the failure case below.
592 while (*ptr == '/') {
593 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
594 !(nd->nl_flags & NLC_WILLBEDIR)
602 * Continuation case: additional elements and the current
603 * element is a directory.
605 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
606 cache_drop(&nd->nl_nch);
613 * Failure case: additional elements and the current element
623 * Successful lookup of last element.
625 * Check permissions if the target exists. If the target does not
626 * exist directory permissions were already tested in the early
627 * completion code above.
629 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
630 * if the file is marked append-only, and NLC_STICKY if the directory
631 * containing the file is sticky.
633 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
634 error = naccess(&nch, nd->nl_flags | dflags,
643 * Termination: no more elements.
645 * If NLC_REFDVP is set acquire a referenced parent dvp.
647 if (nd->nl_flags & NLC_REFDVP) {
648 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
650 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
655 cache_drop(&nd->nl_nch);
657 nd->nl_flags |= NLC_NCPISLOCKED;
663 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
664 * (ncp->nc_error will be ENOENT), but we will still return an error
671 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
672 * of continuity in the namecache tree by connecting the ncp related to the
673 * vnode under the mount to the ncp related to the mount's root vnode.
675 * If no error occured a locked, ref'd ncp is stored in *ncpp.
678 nlookup_mp(struct mount *mp, struct nchandle *nch)
684 cache_get(&mp->mnt_ncmountpt, nch);
685 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
686 while (vfs_busy(mp, 0))
688 error = VFS_ROOT(mp, &vp);
693 cache_setvp(nch, vp);
701 * Read the contents of a symlink, allocate a path buffer out of the
702 * namei_oc and initialize the supplied nlcomponent with the result.
704 * If an error occurs no buffer will be allocated or returned in the nlc.
707 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
708 struct nlcomponent *nlc)
717 nlc->nlc_nameptr = NULL;
718 nlc->nlc_namelen = 0;
719 if (nch->ncp->nc_vp == NULL)
721 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
723 cp = objcache_get(namei_oc, M_WAITOK);
725 aiov.iov_len = MAXPATHLEN;
726 auio.uio_iov = &aiov;
729 auio.uio_rw = UIO_READ;
730 auio.uio_segflg = UIO_SYSSPACE;
731 auio.uio_td = nd->nl_td;
732 auio.uio_resid = MAXPATHLEN - 1;
733 error = VOP_READLINK(vp, &auio, nd->nl_cred);
736 linklen = MAXPATHLEN - 1 - auio.uio_resid;
738 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
740 error = ENAMETOOLONG;
745 nlc->nlc_nameptr = cp;
746 nlc->nlc_namelen = linklen;
750 objcache_put(namei_oc, cp);
756 * Check access [XXX cache vattr!] [XXX quota]
758 * Generally check the NLC_* access bits. All specified bits must pass
759 * for this function to return 0.
761 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
762 * access, otherwise it must exist. No error is returned in this case.
764 * The file must not exist if NLC_EXCL is specified.
766 * Directory permissions in general are tested for NLC_CREATE if the file
767 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
768 * whether the file exists or not.
770 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
771 * the latter is only tested if the target exists.
773 * The passed ncp may or may not be locked. The caller should use a
774 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
775 * NLC_DELETE, and NLC_EXCL checks.
778 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
786 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
788 cache_resolve(nch, cred);
791 error = nch->ncp->nc_error;
794 * Directory permissions checks. Silently ignore ENOENT if these
795 * tests pass. It isn't an error.
797 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
798 if (((nflags & NLC_CREATE) && nch->ncp->nc_vp == NULL) ||
799 ((nflags & NLC_DELETE) && nch->ncp->nc_vp != NULL) ||
800 ((nflags & NLC_RENAME_SRC) && nch->ncp->nc_vp != NULL) ||
801 (nflags & NLC_RENAME_DST)
803 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
806 } else if (error == 0 || error == ENOENT) {
807 par.mount = nch->mount;
810 error = naccess(&par, NLC_WRITE, cred, NULL);
817 * NLC_EXCL check. Target file must not exist.
819 if (error == 0 && (nflags & NLC_EXCL) && nch->ncp->nc_vp != NULL)
823 * Get the vnode attributes so we can do the rest of our checks.
825 * NOTE: We only call naccess_va() if the target exists.
828 error = cache_vget(nch, cred, LK_SHARED, &vp);
829 if (error == ENOENT) {
831 * Silently zero-out ENOENT if creating or renaming
832 * (rename target). It isn't an error.
834 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
836 } else if (error == 0) {
838 * Get the vnode attributes and check for illegal O_TRUNC
839 * requests and read-only mounts.
841 * NOTE: You can still open devices on read-only mounts for
844 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
845 * check on the parent directory above.
847 * XXX cache the va in the namecache or in the vnode
849 error = VOP_GETATTR(vp, &va);
850 if (error == 0 && (nflags & NLC_TRUNCATE)) {
862 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
863 (vp->v_mount->mnt_flag & MNT_RDONLY)
879 * Check permissions based on file attributes. The passed
880 * flags (*nflagsp) are modified with feedback based on
881 * special attributes and requirements.
885 * Adjust the returned (*nflagsp) if non-NULL.
888 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
889 *nflagsp |= NLC_STICKY;
890 if (va.va_flags & APPEND)
891 *nflagsp |= NLC_APPENDONLY;
892 if (va.va_flags & IMMUTABLE)
893 *nflagsp |= NLC_IMMUTABLE;
897 * Process general access.
899 error = naccess_va(&va, nflags, cred);
907 * Check the requested access against the given vattr using cred.
910 naccess_va(struct vattr *va, int nflags, struct ucred *cred)
916 * Test the immutable bit. Creations, deletions, renames (source
917 * or destination) are not allowed. chown/chmod/other is also not
918 * allowed but is handled by SETATTR. Hardlinks to the immutable
921 * If the directory is set to immutable then creations, deletions,
922 * renames (source or dest) and hardlinks to files within the directory
923 * are not allowed, and regular files opened through the directory may
924 * not be written to or truncated (unless a special device).
926 * NOTE! New hardlinks to immutable files work but new hardlinks to
927 * files, immutable or not, sitting inside an immutable directory are
928 * not allowed. As always if the file is hardlinked via some other
929 * path additional hardlinks may be possible even if the file is marked
930 * immutable. The sysop needs to create a closure by checking the hard
931 * link count. Once closure is achieved you are good, and security
932 * scripts should check link counts anyway.
934 * Writes and truncations are only allowed on special devices.
936 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
937 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
939 if (nflags & (NLC_CREATE | NLC_DELETE |
940 NLC_RENAME_SRC | NLC_RENAME_DST)) {
943 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
944 switch(va->va_type) {
957 * Test the no-unlink and append-only bits for opens, rename targets,
958 * and deletions. These bits are not tested for creations or
961 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
962 * If you do not wish this you must also set NOUNLINK.
964 * If the governing directory is marked APPEND-only it implies
965 * NOUNLINK for all entries in the directory.
967 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
968 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
974 * A file marked append-only may not be deleted but can be renamed.
976 if ((va->va_flags & APPEND) &&
977 (nflags & (NLC_DELETE | NLC_RENAME_DST))
983 * A file marked append-only which is opened for writing must also
984 * be opened O_APPEND.
986 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
987 if (nflags & NLC_TRUNCATE)
989 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
990 if ((nflags & NLC_APPEND) == 0)
996 * root gets universal access
998 if (cred->cr_uid == 0)
1002 * Check owner perms.
1004 * If NLC_OWN is set the owner of the file is allowed no matter when
1005 * the owner-mode bits say (utimes).
1008 if (nflags & NLC_READ)
1010 if (nflags & NLC_WRITE)
1012 if (nflags & NLC_EXEC)
1015 if (cred->cr_uid == va->va_uid) {
1016 if ((nflags & NLC_OWN) == 0) {
1017 if ((vmode & va->va_mode) != vmode)
1024 * If NLC_STICKY is set only the owner may delete or rename a file.
1025 * This bit is typically set on /tmp.
1027 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1028 * the specific delete or rename case. For deletions and renames we
1029 * usually just care about directory permissions, not file permissions.
1031 if ((nflags & NLC_STICKY) &&
1032 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
1040 for (i = 0; i < cred->cr_ngroups; ++i) {
1041 if (va->va_gid == cred->cr_groups[i]) {
1042 if ((vmode & va->va_mode) != vmode)
1052 if ((vmode & va->va_mode) != vmode)