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 globaldata_t gd = mycpu;
412 struct nlcomponent nlc;
415 struct nchandle nctmp;
426 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
427 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
429 bzero(&nlc, sizeof(nlc));
432 * Setup for the loop. The current working namecache element is
433 * always at least referenced. We lock it as required, but always
434 * return a locked, resolved namecache entry.
444 * Loop on the path components. At the top of the loop nd->nl_nch
445 * is ref'd and unlocked and represents our current position.
449 * Make sure nl_nch is locked so we can access the vnode, resolution
452 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) {
453 nd->nl_flags |= NLC_NCPISLOCKED;
454 cache_lock(&nd->nl_nch);
458 * Check if the root directory should replace the current
459 * directory. This is done at the start of a translation
460 * or after a symbolic link has been found. In other cases
461 * ptr will never be pointing at a '/'.
466 } while (*ptr == '/');
467 cache_get(&nd->nl_rootnch, &nch);
468 cache_put(&nd->nl_nch);
469 nd->nl_nch = nch; /* remains locked */
472 * Fast-track termination. There is no parent directory of
473 * the root in the same mount from the point of view of
474 * the caller so return EPERM if NLC_REFDVP is specified.
475 * e.g. 'rmdir /' is not allowed.
478 if (nd->nl_flags & NLC_REFDVP)
488 * Check directory search permissions.
491 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags);
496 * Extract the path component. Path components are limited to
499 nlc.nlc_nameptr = ptr;
500 while (*ptr && *ptr != '/')
502 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
503 if (nlc.nlc_namelen >= 256) {
504 error = ENAMETOOLONG;
509 * Lookup the path component in the cache, creating an unresolved
510 * entry if necessary. We have to handle "." and ".." as special
513 * When handling ".." we have to detect a traversal back through a
514 * mount point. If we are at the root, ".." just returns the root.
516 * When handling "." or ".." we also have to recalculate dflags
517 * since our dflags will be for some sub-directory instead of the
520 * This subsection returns a locked, refd 'nch' unless it errors out.
521 * The namecache topology is not allowed to be disconnected, so
522 * encountering a NULL parent will generate EINVAL. This typically
523 * occurs when a directory is removed out from under a process.
525 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
526 cache_get(&nd->nl_nch, &nch);
528 } else if (nlc.nlc_namelen == 2 &&
529 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
530 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
531 nd->nl_nch.ncp == nd->nl_rootnch.ncp
534 * ".." at the root returns the root
536 cache_get(&nd->nl_nch, &nch);
539 * Locate the parent ncp. If we are at the root of a
540 * filesystem mount we have to skip to the mounted-on
541 * point in the underlying filesystem.
543 * Expect the parent to always be good since the
544 * mountpoint doesn't go away. XXX hack. cache_get()
545 * requires the ncp to already have a ref as a safety.
548 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
549 nctmp = nctmp.mount->mnt_ncmounton;
550 nctmp.ncp = nctmp.ncp->nc_parent;
551 KKASSERT(nctmp.ncp != NULL);
553 cache_get(&nctmp, &nch);
554 cache_drop(&nctmp); /* NOTE: zero's nctmp */
559 * Must unlock nl_nch when traversing down the path.
561 cache_unlock(&nd->nl_nch);
562 nd->nl_flags &= ~NLC_NCPISLOCKED;
563 nch = cache_nlookup(&nd->nl_nch, &nlc);
564 if (nch.ncp->nc_flag & NCF_UNRESOLVED)
566 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
567 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
568 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
570 nch = cache_nlookup(&nd->nl_nch, &nlc);
576 * If the last component was "." or ".." our dflags no longer
577 * represents the parent directory and we have to explicitly
580 * Expect the parent to be good since nch is locked.
582 if (wasdotordotdot && error == 0) {
584 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
585 par.mount = nch.mount;
588 error = naccess(&par, 0, nd->nl_cred, &dflags);
592 if (nd->nl_flags & NLC_NCPISLOCKED) {
593 cache_unlock(&nd->nl_nch);
594 nd->nl_flags &= ~NLC_NCPISLOCKED;
598 * [end of subsection]
600 * nch is locked and referenced.
601 * nd->nl_nch is unlocked and referenced.
603 * nl_nch must be unlocked or we could chain lock to the root
604 * if a resolve gets stuck (e.g. in NFS).
608 * Resolve the namespace if necessary. The ncp returned by
609 * cache_nlookup() is referenced and locked.
611 * XXX neither '.' nor '..' should return EAGAIN since they were
612 * previously resolved and thus cannot be newly created ncp's.
614 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
616 error = cache_resolve(&nch, nd->nl_cred);
617 KKASSERT(error != EAGAIN);
619 error = nch.ncp->nc_error;
623 * Early completion. ENOENT is not an error if this is the last
624 * component and NLC_CREATE or NLC_RENAME (rename target) was
625 * requested. Note that ncp->nc_error is left as ENOENT in that
626 * case, which we check later on.
628 * Also handle invalid '.' or '..' components terminating a path
629 * for a create/rename/delete. The standard requires this and pax
630 * pretty stupidly depends on it.
632 for (xptr = ptr; *xptr == '/'; ++xptr)
635 if (error == ENOENT &&
636 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
638 if (nd->nl_flags & NLC_NFS_RDONLY) {
641 error = naccess(&nch, nd->nl_flags | dflags,
645 if (error == 0 && wasdotordotdot &&
646 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
647 NLC_RENAME_SRC | NLC_RENAME_DST))) {
651 if (nd->nl_flags & NLC_CREATE)
653 else if (nd->nl_flags & NLC_DELETE)
654 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
661 * Early completion on error.
669 * If the element is a symlink and it is either not the last
670 * element or it is the last element and we are allowed to
671 * follow symlinks, resolve the symlink.
673 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
674 (*ptr || (nd->nl_flags & NLC_FOLLOW))
676 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
681 error = nreadsymlink(nd, &nch, &nlc);
687 * Concatenate trailing path elements onto the returned symlink.
688 * Note that if the path component (ptr) is not exhausted, it
689 * will being with a '/', so we do not have to add another one.
691 * The symlink may not be empty.
694 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
695 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
696 objcache_put(namei_oc, nlc.nlc_nameptr);
699 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
700 if (nd->nl_flags & NLC_HASBUF)
701 objcache_put(namei_oc, nd->nl_path);
702 nd->nl_path = nlc.nlc_nameptr;
703 nd->nl_flags |= NLC_HASBUF;
707 * Go back up to the top to resolve any initial '/'s in the
714 * If the element is a directory and we are crossing a mount point,
717 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
718 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
719 (mp = cache_findmount(&nch)) != NULL
724 cache_get(&mp->mnt_ncmountpt, &nch);
726 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
727 while (vfs_busy(mp, 0))
729 error = VFS_ROOT(mp, &tdp);
733 cache_setvp(&nch, tdp);
743 * Skip any slashes to get to the next element. If there
744 * are any slashes at all the current element must be a
745 * directory or, in the create case, intended to become a directory.
746 * If it isn't we break without incrementing ptr and fall through
747 * to the failure case below.
749 while (*ptr == '/') {
750 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
751 !(nd->nl_flags & NLC_WILLBEDIR)
759 * Continuation case: additional elements and the current
760 * element is a directory.
762 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
763 cache_drop(&nd->nl_nch);
765 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
771 * Failure case: additional elements and the current element
781 * Successful lookup of last element.
783 * Check permissions if the target exists. If the target does not
784 * exist directory permissions were already tested in the early
785 * completion code above.
787 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
788 * if the file is marked append-only, and NLC_STICKY if the directory
789 * containing the file is sticky.
791 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
792 error = naccess(&nch, nd->nl_flags | dflags,
801 * Termination: no more elements.
803 * If NLC_REFDVP is set acquire a referenced parent dvp.
805 if (nd->nl_flags & NLC_REFDVP) {
806 cache_lock(&nd->nl_nch);
807 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
808 cache_unlock(&nd->nl_nch);
810 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
815 cache_drop(&nd->nl_nch);
817 nd->nl_flags |= NLC_NCPISLOCKED;
823 ++gd->gd_nchstats->ncs_longhits;
825 ++gd->gd_nchstats->ncs_longmiss;
828 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
829 * (ncp->nc_error will be ENOENT), but we will still return an error
836 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
837 * of continuity in the namecache tree by connecting the ncp related to the
838 * vnode under the mount to the ncp related to the mount's root vnode.
840 * If no error occured a locked, ref'd ncp is stored in *ncpp.
843 nlookup_mp(struct mount *mp, struct nchandle *nch)
849 cache_get(&mp->mnt_ncmountpt, nch);
850 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
851 while (vfs_busy(mp, 0))
853 error = VFS_ROOT(mp, &vp);
858 cache_setvp(nch, vp);
866 * Read the contents of a symlink, allocate a path buffer out of the
867 * namei_oc and initialize the supplied nlcomponent with the result.
869 * If an error occurs no buffer will be allocated or returned in the nlc.
872 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
873 struct nlcomponent *nlc)
882 nlc->nlc_nameptr = NULL;
883 nlc->nlc_namelen = 0;
884 if (nch->ncp->nc_vp == NULL)
886 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
888 cp = objcache_get(namei_oc, M_WAITOK);
890 aiov.iov_len = MAXPATHLEN;
891 auio.uio_iov = &aiov;
894 auio.uio_rw = UIO_READ;
895 auio.uio_segflg = UIO_SYSSPACE;
896 auio.uio_td = nd->nl_td;
897 auio.uio_resid = MAXPATHLEN - 1;
898 error = VOP_READLINK(vp, &auio, nd->nl_cred);
901 linklen = MAXPATHLEN - 1 - auio.uio_resid;
903 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
905 error = ENAMETOOLONG;
910 nlc->nlc_nameptr = cp;
911 nlc->nlc_namelen = linklen;
915 objcache_put(namei_oc, cp);
921 * Check access [XXX cache vattr!] [XXX quota]
923 * Generally check the NLC_* access bits. All specified bits must pass
924 * for this function to return 0.
926 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
927 * access, otherwise it must exist. No error is returned in this case.
929 * The file must not exist if NLC_EXCL is specified.
931 * Directory permissions in general are tested for NLC_CREATE if the file
932 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
933 * whether the file exists or not.
935 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
936 * the latter is only tested if the target exists.
938 * The passed ncp must be referenced and locked.
941 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
945 struct namecache *ncp;
949 ASSERT_NCH_LOCKED(nch);
951 if (ncp->nc_flag & NCF_UNRESOLVED) {
952 cache_resolve(nch, cred);
955 error = ncp->nc_error;
958 * Directory permissions checks. Silently ignore ENOENT if these
959 * tests pass. It isn't an error.
961 * We can safely resolve ncp->nc_parent because ncp is currently
964 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
965 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) ||
966 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) ||
967 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) ||
968 (nflags & NLC_RENAME_DST)
972 if ((par.ncp = ncp->nc_parent) == NULL) {
975 } else if (error == 0 || error == ENOENT) {
976 par.mount = nch->mount;
979 error = naccess(&par, NLC_WRITE, cred, NULL);
986 * NLC_EXCL check. Target file must not exist.
988 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL)
992 * Get the vnode attributes so we can do the rest of our checks.
994 * NOTE: We only call naccess_va() if the target exists.
997 error = cache_vget(nch, cred, LK_SHARED, &vp);
998 if (error == ENOENT) {
1000 * Silently zero-out ENOENT if creating or renaming
1001 * (rename target). It isn't an error.
1003 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
1005 } else if (error == 0) {
1007 * Get the vnode attributes and check for illegal O_TRUNC
1008 * requests and read-only mounts.
1010 * NOTE: You can still open devices on read-only mounts for
1013 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
1014 * check on the parent directory above.
1016 * XXX cache the va in the namecache or in the vnode
1018 error = VOP_GETATTR(vp, &va);
1019 if (error == 0 && (nflags & NLC_TRUNCATE)) {
1020 switch(va.va_type) {
1035 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
1036 (vp->v_mount->mnt_flag & MNT_RDONLY)
1038 switch(va.va_type) {
1052 * Check permissions based on file attributes. The passed
1053 * flags (*nflagsp) are modified with feedback based on
1054 * special attributes and requirements.
1058 * Adjust the returned (*nflagsp) if non-NULL.
1061 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
1062 *nflagsp |= NLC_STICKY;
1063 if (va.va_flags & APPEND)
1064 *nflagsp |= NLC_APPENDONLY;
1065 if (va.va_flags & IMMUTABLE)
1066 *nflagsp |= NLC_IMMUTABLE;
1070 * Track swapcache management flags in the namecache.
1072 * Calculate the flags based on the current vattr info
1073 * and recalculate the inherited flags from the parent
1074 * (the original cache linkage may have occurred without
1075 * getattrs and thus have stale flags).
1078 if (va.va_flags & SF_NOCACHE)
1079 cflags |= NCF_SF_NOCACHE;
1080 if (va.va_flags & UF_CACHE)
1081 cflags |= NCF_UF_CACHE;
1082 if (ncp->nc_parent) {
1083 if (ncp->nc_parent->nc_flag &
1084 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) {
1085 cflags |= NCF_SF_PNOCACHE;
1087 if (ncp->nc_parent->nc_flag &
1088 (NCF_UF_CACHE | NCF_UF_PCACHE)) {
1089 cflags |= NCF_UF_PCACHE;
1092 ncp->nc_flag &= ~(NCF_SF_NOCACHE | NCF_UF_CACHE |
1093 NCF_SF_PNOCACHE | NCF_UF_PCACHE);
1094 ncp->nc_flag |= cflags;
1097 * Process general access.
1099 error = naccess_va(&va, nflags, cred);
1107 * Check the requested access against the given vattr using cred.
1110 naccess_va(struct vattr *va, int nflags, struct ucred *cred)
1116 * Test the immutable bit. Creations, deletions, renames (source
1117 * or destination) are not allowed. chown/chmod/other is also not
1118 * allowed but is handled by SETATTR. Hardlinks to the immutable
1121 * If the directory is set to immutable then creations, deletions,
1122 * renames (source or dest) and hardlinks to files within the directory
1123 * are not allowed, and regular files opened through the directory may
1124 * not be written to or truncated (unless a special device).
1126 * NOTE! New hardlinks to immutable files work but new hardlinks to
1127 * files, immutable or not, sitting inside an immutable directory are
1128 * not allowed. As always if the file is hardlinked via some other
1129 * path additional hardlinks may be possible even if the file is marked
1130 * immutable. The sysop needs to create a closure by checking the hard
1131 * link count. Once closure is achieved you are good, and security
1132 * scripts should check link counts anyway.
1134 * Writes and truncations are only allowed on special devices.
1136 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1137 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1139 if (nflags & (NLC_CREATE | NLC_DELETE |
1140 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1143 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1144 switch(va->va_type) {
1158 * Test the no-unlink and append-only bits for opens, rename targets,
1159 * and deletions. These bits are not tested for creations or
1162 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1163 * If you do not wish this you must also set NOUNLINK.
1165 * If the governing directory is marked APPEND-only it implies
1166 * NOUNLINK for all entries in the directory.
1168 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1169 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1175 * A file marked append-only may not be deleted but can be renamed.
1177 if ((va->va_flags & APPEND) &&
1178 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1184 * A file marked append-only which is opened for writing must also
1185 * be opened O_APPEND.
1187 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1188 if (nflags & NLC_TRUNCATE)
1190 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1191 if ((nflags & NLC_APPEND) == 0)
1197 * root gets universal access
1199 if (cred->cr_uid == 0)
1203 * Check owner perms.
1205 * If NLC_OWN is set the owner of the file is allowed no matter when
1206 * the owner-mode bits say (utimes).
1209 if (nflags & NLC_READ)
1211 if (nflags & NLC_WRITE)
1213 if (nflags & NLC_EXEC)
1216 if (cred->cr_uid == va->va_uid) {
1217 if ((nflags & NLC_OWN) == 0) {
1218 if ((vmode & va->va_mode) != vmode)
1225 * If NLC_STICKY is set only the owner may delete or rename a file.
1226 * This bit is typically set on /tmp.
1228 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1229 * the specific delete or rename case. For deletions and renames we
1230 * usually just care about directory permissions, not file permissions.
1232 if ((nflags & NLC_STICKY) &&
1233 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
1241 for (i = 0; i < cred->cr_ngroups; ++i) {
1242 if (va->va_gid == cred->cr_groups[i]) {
1243 if ((vmode & va->va_mode) != vmode)
1253 if ((vmode & va->va_mode) != vmode)