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;
376 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
377 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
379 bzero(&nlc, sizeof(nlc));
382 * Setup for the loop. The current working namecache element must
383 * be in a refd + unlocked state. This typically the case on entry except
384 * when stringing nlookup()'s along in a chain, since nlookup() always
385 * returns nl_nch in a locked state.
388 if (nd->nl_flags & NLC_NCPISLOCKED) {
389 nd->nl_flags &= ~NLC_NCPISLOCKED;
390 cache_unlock(&nd->nl_nch);
399 * Loop on the path components. At the top of the loop nd->nl_nch
400 * is ref'd and unlocked and represents our current position.
404 * Check if the root directory should replace the current
405 * directory. This is done at the start of a translation
406 * or after a symbolic link has been found. In other cases
407 * ptr will never be pointing at a '/'.
412 } while (*ptr == '/');
413 cache_copy(&nd->nl_rootnch, &nch);
414 cache_drop(&nd->nl_nch);
418 * Fast-track termination. There is no parent directory of
419 * the root in the same mount from the point of view of
420 * the caller so return EPERM if NLC_REFDVP is specified.
421 * e.g. 'rmdir /' is not allowed.
424 if (nd->nl_flags & NLC_REFDVP) {
427 cache_lock(&nd->nl_nch);
428 nd->nl_flags |= NLC_NCPISLOCKED;
437 * Check directory search permissions.
440 if ((error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags)) != 0)
444 * Extract the path component. Path components are limited to
447 nlc.nlc_nameptr = ptr;
448 while (*ptr && *ptr != '/')
450 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
451 if (nlc.nlc_namelen >= 256) {
452 error = ENAMETOOLONG;
457 * Lookup the path component in the cache, creating an unresolved
458 * entry if necessary. We have to handle "." and ".." as special
461 * When handling ".." we have to detect a traversal back through a
462 * mount point. If we are at the root, ".." just returns the root.
464 * When handling "." or ".." we also have to recalculate dflags
465 * since our dflags will be for some sub-directory instead of the
468 * This subsection returns a locked, refd 'nch' unless it errors out.
469 * The namecache topology is not allowed to be disconnected, so
470 * encountering a NULL parent will generate EINVAL. This typically
471 * occurs when a directory is removed out from under a process.
473 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
474 cache_get(&nd->nl_nch, &nch);
476 } else if (nlc.nlc_namelen == 2 &&
477 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
478 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
479 nd->nl_nch.ncp == nd->nl_rootnch.ncp
482 * ".." at the root returns the root
484 cache_get(&nd->nl_nch, &nch);
487 * Locate the parent ncp. If we are at the root of a
488 * filesystem mount we have to skip to the mounted-on
489 * point in the underlying filesystem.
492 while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
493 nch = nch.mount->mnt_ncmounton;
494 nch.ncp = nch.ncp->nc_parent;
495 KKASSERT(nch.ncp != NULL);
496 cache_get(&nch, &nch);
500 nch = cache_nlookup(&nd->nl_nch, &nlc);
501 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
502 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
503 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
505 nch = cache_nlookup(&nd->nl_nch, &nlc);
511 * If the last component was "." or ".." our dflags no longer
512 * represents the parent directory and we have to explicitly
515 if (wasdotordotdot && error == 0) {
517 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
518 par.mount = nch.mount;
521 error = naccess(&par, 0, nd->nl_cred, &dflags);
527 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
531 * Resolve the namespace if necessary. The ncp returned by
532 * cache_nlookup() is referenced and locked.
534 * XXX neither '.' nor '..' should return EAGAIN since they were
535 * previously resolved and thus cannot be newly created ncp's.
537 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
538 error = cache_resolve(&nch, nd->nl_cred);
539 KKASSERT(error != EAGAIN);
541 error = nch.ncp->nc_error;
545 * Early completion. ENOENT is not an error if this is the last
546 * component and NLC_CREATE or NLC_RENAME (rename target) was
547 * requested. Note that ncp->nc_error is left as ENOENT in that
548 * case, which we check later on.
550 * Also handle invalid '.' or '..' components terminating a path
551 * for a create/rename/delete. The standard requires this and pax
552 * pretty stupidly depends on it.
554 for (xptr = ptr; *xptr == '/'; ++xptr)
557 if (error == ENOENT &&
558 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
560 if (nd->nl_flags & NLC_NFS_RDONLY) {
563 error = naccess(&nch, nd->nl_flags | dflags,
567 if (error == 0 && wasdotordotdot &&
568 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
569 NLC_RENAME_SRC | NLC_RENAME_DST))) {
573 if (nd->nl_flags & NLC_CREATE)
575 else if (nd->nl_flags & NLC_DELETE)
576 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
583 * Early completion on error.
591 * If the element is a symlink and it is either not the last
592 * element or it is the last element and we are allowed to
593 * follow symlinks, resolve the symlink.
595 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
596 (*ptr || (nd->nl_flags & NLC_FOLLOW))
598 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
603 error = nreadsymlink(nd, &nch, &nlc);
609 * Concatenate trailing path elements onto the returned symlink.
610 * Note that if the path component (ptr) is not exhausted, it
611 * will being with a '/', so we do not have to add another one.
613 * The symlink may not be empty.
616 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
617 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
618 objcache_put(namei_oc, nlc.nlc_nameptr);
621 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
622 if (nd->nl_flags & NLC_HASBUF)
623 objcache_put(namei_oc, nd->nl_path);
624 nd->nl_path = nlc.nlc_nameptr;
625 nd->nl_flags |= NLC_HASBUF;
629 * Go back up to the top to resolve any initial '/'s in the
636 * If the element is a directory and we are crossing a mount point,
639 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
640 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
641 (mp = cache_findmount(&nch)) != NULL
646 cache_get(&mp->mnt_ncmountpt, &nch);
648 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
649 while (vfs_busy(mp, 0))
651 error = VFS_ROOT(mp, &tdp);
655 cache_setvp(&nch, tdp);
665 * Skip any slashes to get to the next element. If there
666 * are any slashes at all the current element must be a
667 * directory or, in the create case, intended to become a directory.
668 * If it isn't we break without incrementing ptr and fall through
669 * to the failure case below.
671 while (*ptr == '/') {
672 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
673 !(nd->nl_flags & NLC_WILLBEDIR)
681 * Continuation case: additional elements and the current
682 * element is a directory.
684 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
685 cache_drop(&nd->nl_nch);
692 * Failure case: additional elements and the current element
702 * Successful lookup of last element.
704 * Check permissions if the target exists. If the target does not
705 * exist directory permissions were already tested in the early
706 * completion code above.
708 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
709 * if the file is marked append-only, and NLC_STICKY if the directory
710 * containing the file is sticky.
712 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
713 error = naccess(&nch, nd->nl_flags | dflags,
722 * Termination: no more elements.
724 * If NLC_REFDVP is set acquire a referenced parent dvp.
726 if (nd->nl_flags & NLC_REFDVP) {
727 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
729 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
734 cache_drop(&nd->nl_nch);
736 nd->nl_flags |= NLC_NCPISLOCKED;
742 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
743 * (ncp->nc_error will be ENOENT), but we will still return an error
750 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
751 * of continuity in the namecache tree by connecting the ncp related to the
752 * vnode under the mount to the ncp related to the mount's root vnode.
754 * If no error occured a locked, ref'd ncp is stored in *ncpp.
757 nlookup_mp(struct mount *mp, struct nchandle *nch)
763 cache_get(&mp->mnt_ncmountpt, nch);
764 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
765 while (vfs_busy(mp, 0))
767 error = VFS_ROOT(mp, &vp);
772 cache_setvp(nch, vp);
780 * Read the contents of a symlink, allocate a path buffer out of the
781 * namei_oc and initialize the supplied nlcomponent with the result.
783 * If an error occurs no buffer will be allocated or returned in the nlc.
786 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
787 struct nlcomponent *nlc)
796 nlc->nlc_nameptr = NULL;
797 nlc->nlc_namelen = 0;
798 if (nch->ncp->nc_vp == NULL)
800 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
802 cp = objcache_get(namei_oc, M_WAITOK);
804 aiov.iov_len = MAXPATHLEN;
805 auio.uio_iov = &aiov;
808 auio.uio_rw = UIO_READ;
809 auio.uio_segflg = UIO_SYSSPACE;
810 auio.uio_td = nd->nl_td;
811 auio.uio_resid = MAXPATHLEN - 1;
812 error = VOP_READLINK(vp, &auio, nd->nl_cred);
815 linklen = MAXPATHLEN - 1 - auio.uio_resid;
817 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
819 error = ENAMETOOLONG;
824 nlc->nlc_nameptr = cp;
825 nlc->nlc_namelen = linklen;
829 objcache_put(namei_oc, cp);
835 * Check access [XXX cache vattr!] [XXX quota]
837 * Generally check the NLC_* access bits. All specified bits must pass
838 * for this function to return 0.
840 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
841 * access, otherwise it must exist. No error is returned in this case.
843 * The file must not exist if NLC_EXCL is specified.
845 * Directory permissions in general are tested for NLC_CREATE if the file
846 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
847 * whether the file exists or not.
849 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
850 * the latter is only tested if the target exists.
852 * The passed ncp may or may not be locked. The caller should use a
853 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
854 * NLC_DELETE, and NLC_EXCL checks.
857 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
865 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
867 cache_resolve(nch, cred);
870 error = nch->ncp->nc_error;
873 * Directory permissions checks. Silently ignore ENOENT if these
874 * tests pass. It isn't an error.
876 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
877 if (((nflags & NLC_CREATE) && nch->ncp->nc_vp == NULL) ||
878 ((nflags & NLC_DELETE) && nch->ncp->nc_vp != NULL) ||
879 ((nflags & NLC_RENAME_SRC) && nch->ncp->nc_vp != NULL) ||
880 (nflags & NLC_RENAME_DST)
882 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
885 } else if (error == 0 || error == ENOENT) {
886 par.mount = nch->mount;
889 error = naccess(&par, NLC_WRITE, cred, NULL);
896 * NLC_EXCL check. Target file must not exist.
898 if (error == 0 && (nflags & NLC_EXCL) && nch->ncp->nc_vp != NULL)
902 * Get the vnode attributes so we can do the rest of our checks.
904 * NOTE: We only call naccess_va() if the target exists.
907 error = cache_vget(nch, cred, LK_SHARED, &vp);
908 if (error == ENOENT) {
910 * Silently zero-out ENOENT if creating or renaming
911 * (rename target). It isn't an error.
913 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
915 } else if (error == 0) {
917 * Get the vnode attributes and check for illegal O_TRUNC
918 * requests and read-only mounts.
920 * NOTE: You can still open devices on read-only mounts for
923 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
924 * check on the parent directory above.
926 * XXX cache the va in the namecache or in the vnode
928 error = VOP_GETATTR(vp, &va);
929 if (error == 0 && (nflags & NLC_TRUNCATE)) {
945 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
946 (vp->v_mount->mnt_flag & MNT_RDONLY)
962 * Check permissions based on file attributes. The passed
963 * flags (*nflagsp) are modified with feedback based on
964 * special attributes and requirements.
968 * Adjust the returned (*nflagsp) if non-NULL.
971 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
972 *nflagsp |= NLC_STICKY;
973 if (va.va_flags & APPEND)
974 *nflagsp |= NLC_APPENDONLY;
975 if (va.va_flags & IMMUTABLE)
976 *nflagsp |= NLC_IMMUTABLE;
980 * Process general access.
982 error = naccess_va(&va, nflags, cred);
990 * Check the requested access against the given vattr using cred.
993 naccess_va(struct vattr *va, int nflags, struct ucred *cred)
999 * Test the immutable bit. Creations, deletions, renames (source
1000 * or destination) are not allowed. chown/chmod/other is also not
1001 * allowed but is handled by SETATTR. Hardlinks to the immutable
1004 * If the directory is set to immutable then creations, deletions,
1005 * renames (source or dest) and hardlinks to files within the directory
1006 * are not allowed, and regular files opened through the directory may
1007 * not be written to or truncated (unless a special device).
1009 * NOTE! New hardlinks to immutable files work but new hardlinks to
1010 * files, immutable or not, sitting inside an immutable directory are
1011 * not allowed. As always if the file is hardlinked via some other
1012 * path additional hardlinks may be possible even if the file is marked
1013 * immutable. The sysop needs to create a closure by checking the hard
1014 * link count. Once closure is achieved you are good, and security
1015 * scripts should check link counts anyway.
1017 * Writes and truncations are only allowed on special devices.
1019 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1020 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1022 if (nflags & (NLC_CREATE | NLC_DELETE |
1023 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1026 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1027 switch(va->va_type) {
1041 * Test the no-unlink and append-only bits for opens, rename targets,
1042 * and deletions. These bits are not tested for creations or
1045 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1046 * If you do not wish this you must also set NOUNLINK.
1048 * If the governing directory is marked APPEND-only it implies
1049 * NOUNLINK for all entries in the directory.
1051 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1052 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1058 * A file marked append-only may not be deleted but can be renamed.
1060 if ((va->va_flags & APPEND) &&
1061 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1067 * A file marked append-only which is opened for writing must also
1068 * be opened O_APPEND.
1070 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1071 if (nflags & NLC_TRUNCATE)
1073 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1074 if ((nflags & NLC_APPEND) == 0)
1080 * root gets universal access
1082 if (cred->cr_uid == 0)
1086 * Check owner perms.
1088 * If NLC_OWN is set the owner of the file is allowed no matter when
1089 * the owner-mode bits say (utimes).
1092 if (nflags & NLC_READ)
1094 if (nflags & NLC_WRITE)
1096 if (nflags & NLC_EXEC)
1099 if (cred->cr_uid == va->va_uid) {
1100 if ((nflags & NLC_OWN) == 0) {
1101 if ((vmode & va->va_mode) != vmode)
1108 * If NLC_STICKY is set only the owner may delete or rename a file.
1109 * This bit is typically set on /tmp.
1111 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1112 * the specific delete or rename case. For deletions and renames we
1113 * usually just care about directory permissions, not file permissions.
1115 if ((nflags & NLC_STICKY) &&
1116 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
1124 for (i = 0; i < cred->cr_ngroups; ++i) {
1125 if (va->va_gid == cred->cr_groups[i]) {
1126 if ((vmode & va->va_mode) != vmode)
1136 if ((vmode & va->va_mode) != vmode)