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
35 * nlookup() is the 'new' namei interface. Rather then return directory and
36 * leaf vnodes (in various lock states) the new interface instead deals in
37 * namecache records. Namecache records may represent both a positive or
38 * a negative hit. The namespace is locked via the namecache record instead
39 * of via the vnode, and only the leaf namecache record (representing the
40 * filename) needs to be locked.
42 * This greatly improves filesystem parallelism and is a huge simplification
43 * of the API verses the old vnode locking / namei scheme.
45 * Filesystems must actively control the caching aspects of the namecache,
46 * and since namecache pointers are used as handles they are non-optional
47 * even for filesystems which do not generally wish to cache things. It is
48 * intended that a separate cache coherency API will be constructed to handle
52 #include "opt_ktrace.h"
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/kernel.h>
57 #include <sys/vnode.h>
58 #include <sys/mount.h>
59 #include <sys/filedesc.h>
61 #include <sys/namei.h>
62 #include <sys/nlookup.h>
63 #include <sys/malloc.h>
65 #include <sys/objcache.h>
69 #include <sys/ktrace.h>
72 static int naccess(struct nchandle *nch, int vmode, struct ucred *cred,
76 * Initialize a nlookup() structure, early error return for copyin faults
77 * or a degenerate empty string (which is not allowed).
79 * The first process proc0's credentials are used if the calling thread
80 * is not associated with a process context.
85 nlookup_init(struct nlookupdata *nd,
86 const char *path, enum uio_seg seg, int flags)
97 * note: the pathlen set by copy*str() includes the terminating \0.
99 bzero(nd, sizeof(struct nlookupdata));
100 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
101 nd->nl_flags |= NLC_HASBUF;
102 if (seg == UIO_SYSSPACE)
103 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
105 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
108 * Don't allow empty pathnames.
109 * POSIX.1 requirement: "" is not a vaild file name.
111 if (error == 0 && pathlen <= 1)
116 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch);
117 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch);
118 if (p->p_fd->fd_njdir.ncp)
119 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch);
120 nd->nl_cred = crhold(p->p_ucred);
122 cache_copy(&rootnch, &nd->nl_nch);
123 cache_copy(&nd->nl_nch, &nd->nl_rootnch);
124 cache_copy(&nd->nl_nch, &nd->nl_jailnch);
125 nd->nl_cred = crhold(proc0.p_ucred);
128 nd->nl_flags |= flags;
137 * nlookup_init() for "at" family of syscalls.
139 * Works similarly to nlookup_init() but if path is relative and fd is not
140 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd.
141 * In this case, the file entry pointed to by fd is ref'ed and returned in
144 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd
145 * and release the ref to the file entry.
148 nlookup_init_at(struct nlookupdata *nd, struct file **fpp, int fd,
149 const char *path, enum uio_seg seg, int flags)
151 struct thread *td = curthread;
152 struct proc *p = td->td_proc;
159 if ((error = nlookup_init(nd, path, seg, flags)) != 0) {
163 if (nd->nl_path[0] != '/' && fd != AT_FDCWD) {
164 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
166 vp = (struct vnode*)fp->f_data;
167 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL) {
173 cache_drop(&nd->nl_nch);
174 cache_copy(&fp->f_nchandle, &nd->nl_nch);
187 * This works similarly to nlookup_init() but does not assume a process
188 * context. rootnch is always chosen for the root directory and the cred
189 * and starting directory are supplied in arguments.
192 nlookup_init_raw(struct nlookupdata *nd,
193 const char *path, enum uio_seg seg, int flags,
194 struct ucred *cred, struct nchandle *ncstart)
202 bzero(nd, sizeof(struct nlookupdata));
203 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
204 nd->nl_flags |= NLC_HASBUF;
205 if (seg == UIO_SYSSPACE)
206 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
208 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
211 * Don't allow empty pathnames.
212 * POSIX.1 requirement: "" is not a vaild file name.
214 if (error == 0 && pathlen <= 1)
218 cache_copy(ncstart, &nd->nl_nch);
219 cache_copy(&rootnch, &nd->nl_rootnch);
220 cache_copy(&rootnch, &nd->nl_jailnch);
221 nd->nl_cred = crhold(cred);
223 nd->nl_flags |= flags;
231 * This works similarly to nlookup_init_raw() but does not rely
232 * on rootnch being initialized yet.
235 nlookup_init_root(struct nlookupdata *nd,
236 const char *path, enum uio_seg seg, int flags,
237 struct ucred *cred, struct nchandle *ncstart,
238 struct nchandle *ncroot)
246 bzero(nd, sizeof(struct nlookupdata));
247 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
248 nd->nl_flags |= NLC_HASBUF;
249 if (seg == UIO_SYSSPACE)
250 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
252 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
255 * Don't allow empty pathnames.
256 * POSIX.1 requirement: "" is not a vaild file name.
258 if (error == 0 && pathlen <= 1)
262 cache_copy(ncstart, &nd->nl_nch);
263 cache_copy(ncroot, &nd->nl_rootnch);
264 cache_copy(ncroot, &nd->nl_jailnch);
265 nd->nl_cred = crhold(cred);
267 nd->nl_flags |= flags;
275 * Set a different credential; this credential will be used by future
276 * operations performed on nd.nl_open_vp and nlookupdata structure.
279 nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
281 KKASSERT(nd->nl_cred != NULL);
283 if (nd->nl_cred != cred) {
291 * Cleanup a nlookupdata structure after we are through with it. This may
292 * be called on any nlookupdata structure initialized with nlookup_init().
293 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
294 * returns an error, even if as a consumer you believe you have taken all
295 * dynamic elements out of the nlookupdata structure.
298 nlookup_done(struct nlookupdata *nd)
300 if (nd->nl_nch.ncp) {
301 if (nd->nl_flags & NLC_NCPISLOCKED) {
302 nd->nl_flags &= ~NLC_NCPISLOCKED;
303 cache_unlock(&nd->nl_nch);
305 cache_drop(&nd->nl_nch); /* NULL's out the nch */
307 if (nd->nl_rootnch.ncp)
308 cache_drop(&nd->nl_rootnch);
309 if (nd->nl_jailnch.ncp)
310 cache_drop(&nd->nl_jailnch);
311 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
312 objcache_put(namei_oc, nd->nl_path);
319 if (nd->nl_open_vp) {
320 if (nd->nl_flags & NLC_LOCKVP) {
321 vn_unlock(nd->nl_open_vp);
322 nd->nl_flags &= ~NLC_LOCKVP;
324 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
325 nd->nl_open_vp = NULL;
331 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
335 * Works similarly to nlookup_done() when nd initialized with
339 nlookup_done_at(struct nlookupdata *nd, struct file *fp)
347 nlookup_zero(struct nlookupdata *nd)
349 bzero(nd, sizeof(struct nlookupdata));
353 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
354 * if an error occured.
356 * Note that the returned ncp is not checked for permissions, though VEXEC
357 * is checked on the directory path leading up to the result. The caller
358 * must call naccess() to check the permissions of the returned leaf.
361 nlookup_simple(const char *str, enum uio_seg seg,
362 int niflags, int *error)
364 struct nlookupdata nd;
367 *error = nlookup_init(&nd, str, seg, niflags);
369 if ((*error = nlookup(&nd)) == 0) {
370 nch = nd.nl_nch; /* keep hold ref from structure */
371 cache_zero(&nd.nl_nch); /* and NULL out */
383 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
384 * on return, even if an error occurs. If no error occurs or NLC_CREATE
385 * is flagged and ENOENT is returned, then the returned nl_nch is always
386 * referenced and locked exclusively.
388 * WARNING: For any general error other than ENOENT w/NLC_CREATE, the
389 * the resulting nl_nch may or may not be locked and if locked
390 * might be locked either shared or exclusive.
392 * Intermediate directory elements, including the current directory, require
393 * execute (search) permission. nlookup does not examine the access
394 * permissions on the returned element.
396 * If NLC_CREATE is set the last directory must allow node creation,
397 * and an error code of 0 will be returned for a non-existant
398 * target (not ENOENT).
400 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
401 * plus the sticky check is made, and an error code of 0 will be returned
402 * for a non-existant target (not ENOENT).
404 * If NLC_DELETE is set the last directory mut allow node deletion,
405 * plus the sticky check is made.
407 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
408 * of the returned entry. The vnode will be referenced, but not locked,
409 * and will be released by nlookup_done() along with everything else.
411 * NOTE: As an optimization we attempt to obtain a shared namecache lock
412 * on any intermediate elements. On success, the returned element
413 * is ALWAYS locked exclusively.
417 islastelement(const char *ptr)
425 nlookup(struct nlookupdata *nd)
427 globaldata_t gd = mycpu;
428 struct nlcomponent nlc;
431 struct nchandle nctmp;
433 struct vnode *hvp; /* hold to prevent recyclement */
442 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
443 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
445 bzero(&nlc, sizeof(nlc));
448 * Setup for the loop. The current working namecache element is
449 * always at least referenced. We lock it as required, but always
450 * return a locked, resolved namecache entry.
460 * Loop on the path components. At the top of the loop nd->nl_nch
461 * is ref'd and unlocked and represents our current position.
465 * Make sure nl_nch is locked so we can access the vnode, resolution
468 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) {
469 nd->nl_flags |= NLC_NCPISLOCKED;
470 cache_lock_maybe_shared(&nd->nl_nch, islastelement(ptr));
474 * Check if the root directory should replace the current
475 * directory. This is done at the start of a translation
476 * or after a symbolic link has been found. In other cases
477 * ptr will never be pointing at a '/'.
482 } while (*ptr == '/');
483 cache_unlock(&nd->nl_nch);
484 cache_get_maybe_shared(&nd->nl_rootnch, &nch,
486 cache_drop(&nd->nl_nch);
487 nd->nl_nch = nch; /* remains locked */
490 * Fast-track termination. There is no parent directory of
491 * the root in the same mount from the point of view of
492 * the caller so return EACCES if NLC_REFDVP is specified,
493 * and EEXIST if NLC_CREATE is also specified.
494 * e.g. 'rmdir /' or 'mkdir /' are not allowed.
497 if (nd->nl_flags & NLC_REFDVP)
498 error = (nd->nl_flags & NLC_CREATE) ? EEXIST : EACCES;
507 * Check directory search permissions (nd->nl_nch is locked & refd)
510 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags);
515 * Extract the path component. Path components are limited to
518 nlc.nlc_nameptr = ptr;
519 while (*ptr && *ptr != '/')
521 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
522 if (nlc.nlc_namelen >= 256) {
523 error = ENAMETOOLONG;
528 * Lookup the path component in the cache, creating an unresolved
529 * entry if necessary. We have to handle "." and ".." as special
532 * When handling ".." we have to detect a traversal back through a
533 * mount point. If we are at the root, ".." just returns the root.
535 * When handling "." or ".." we also have to recalculate dflags
536 * since our dflags will be for some sub-directory instead of the
539 * This subsection returns a locked, refd 'nch' unless it errors out,
540 * and an unlocked but still ref'd nd->nl_nch.
542 * The namecache topology is not allowed to be disconnected, so
543 * encountering a NULL parent will generate EINVAL. This typically
544 * occurs when a directory is removed out from under a process.
546 * WARNING! The unlocking of nd->nl_nch is sensitive code.
548 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
550 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
551 cache_unlock(&nd->nl_nch);
552 nd->nl_flags &= ~NLC_NCPISLOCKED;
553 cache_get_maybe_shared(&nd->nl_nch, &nch, islastelement(ptr));
555 } else if (nlc.nlc_namelen == 2 &&
556 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
557 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
558 nd->nl_nch.ncp == nd->nl_rootnch.ncp
561 * ".." at the root returns the root
563 cache_unlock(&nd->nl_nch);
564 nd->nl_flags &= ~NLC_NCPISLOCKED;
565 cache_get_maybe_shared(&nd->nl_nch, &nch, islastelement(ptr));
568 * Locate the parent ncp. If we are at the root of a
569 * filesystem mount we have to skip to the mounted-on
570 * point in the underlying filesystem.
572 * Expect the parent to always be good since the
573 * mountpoint doesn't go away. XXX hack. cache_get()
574 * requires the ncp to already have a ref as a safety.
577 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
578 nctmp = nctmp.mount->mnt_ncmounton;
579 nctmp.ncp = nctmp.ncp->nc_parent;
580 KKASSERT(nctmp.ncp != NULL);
582 cache_unlock(&nd->nl_nch);
583 nd->nl_flags &= ~NLC_NCPISLOCKED;
584 cache_get_maybe_shared(&nctmp, &nch, islastelement(ptr));
585 cache_drop(&nctmp); /* NOTE: zero's nctmp */
590 * Must unlock nl_nch when traversing down the path. However,
591 * the child ncp has not yet been found/created and the parent's
592 * child list might be empty. Thus releasing the lock can
593 * allow a race whereby the parent ncp's vnode is recycled.
594 * This case can occur especially when maxvnodes is set very low.
596 * We need the parent's ncp to remain resolved for all normal
597 * filesystem activities, so we vhold() the vp during the lookup
598 * to prevent recyclement due to vnlru / maxvnodes.
600 * If we race an unlink or rename the ncp might be marked
601 * DESTROYED after resolution, requiring a retry.
603 if ((hvp = nd->nl_nch.ncp->nc_vp) != NULL)
605 cache_unlock(&nd->nl_nch);
606 nd->nl_flags &= ~NLC_NCPISLOCKED;
607 error = cache_nlookup_maybe_shared(&nd->nl_nch, &nlc,
608 islastelement(ptr), &nch);
609 if (error == EWOULDBLOCK) {
610 nch = cache_nlookup(&nd->nl_nch, &nlc);
611 if (nch.ncp->nc_flag & NCF_UNRESOLVED)
614 error = cache_resolve(&nch, nd->nl_cred);
615 if (error != EAGAIN &&
616 (nch.ncp->nc_flag & NCF_DESTROYED) == 0) {
619 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
620 nch.ncp->nc_nlen, nch.ncp->nc_nlen,
623 nch = cache_nlookup(&nd->nl_nch, &nlc);
632 * If the last component was "." or ".." our dflags no longer
633 * represents the parent directory and we have to explicitly
636 * Expect the parent to be good since nch is locked.
638 if (wasdotordotdot && error == 0) {
640 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
641 par.mount = nch.mount;
643 cache_lock_maybe_shared(&par, islastelement(ptr));
644 error = naccess(&par, 0, nd->nl_cred, &dflags);
650 * [end of subsection]
652 * nch is locked and referenced.
653 * nd->nl_nch is unlocked and referenced.
655 * nl_nch must be unlocked or we could chain lock to the root
656 * if a resolve gets stuck (e.g. in NFS).
658 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
661 * Resolve the namespace if necessary. The ncp returned by
662 * cache_nlookup() is referenced and locked.
664 * XXX neither '.' nor '..' should return EAGAIN since they were
665 * previously resolved and thus cannot be newly created ncp's.
667 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
669 error = cache_resolve(&nch, nd->nl_cred);
670 KKASSERT(error != EAGAIN);
672 error = nch.ncp->nc_error;
676 * Early completion. ENOENT is not an error if this is the last
677 * component and NLC_CREATE or NLC_RENAME (rename target) was
678 * requested. Note that ncp->nc_error is left as ENOENT in that
679 * case, which we check later on.
681 * Also handle invalid '.' or '..' components terminating a path
682 * for a create/rename/delete. The standard requires this and pax
683 * pretty stupidly depends on it.
685 if (islastelement(ptr)) {
686 if (error == ENOENT &&
687 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
689 if (nd->nl_flags & NLC_NFS_RDONLY) {
692 error = naccess(&nch, nd->nl_flags | dflags,
696 if (error == 0 && wasdotordotdot &&
697 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
698 NLC_RENAME_SRC | NLC_RENAME_DST))) {
702 if (nd->nl_flags & NLC_CREATE)
704 else if (nd->nl_flags & NLC_DELETE)
705 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
712 * Early completion on error.
720 * If the element is a symlink and it is either not the last
721 * element or it is the last element and we are allowed to
722 * follow symlinks, resolve the symlink.
724 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
725 (*ptr || (nd->nl_flags & NLC_FOLLOW))
727 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
732 error = nreadsymlink(nd, &nch, &nlc);
738 * Concatenate trailing path elements onto the returned symlink.
739 * Note that if the path component (ptr) is not exhausted, it
740 * will being with a '/', so we do not have to add another one.
742 * The symlink may not be empty.
745 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
746 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
747 objcache_put(namei_oc, nlc.nlc_nameptr);
750 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
751 if (nd->nl_flags & NLC_HASBUF)
752 objcache_put(namei_oc, nd->nl_path);
753 nd->nl_path = nlc.nlc_nameptr;
754 nd->nl_flags |= NLC_HASBUF;
758 * Go back up to the top to resolve any initial '/'s in the
765 * If the element is a directory and we are crossing a mount point,
768 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
769 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
770 (mp = cache_findmount(&nch)) != NULL
776 * VFS must be busied before the namecache entry is locked,
777 * but we don't want to waste time calling vfs_busy() if the
778 * mount point is already resolved.
783 while (vfs_busy(mp, 0)) {
784 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
785 kprintf("nlookup: warning umount race avoided\n");
793 cache_get_maybe_shared(&mp->mnt_ncmountpt, &nch,
796 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
797 if (vfs_do_busy == 0) {
801 error = VFS_ROOT(mp, &tdp);
808 cache_setvp(&nch, tdp);
823 * Skip any slashes to get to the next element. If there
824 * are any slashes at all the current element must be a
825 * directory or, in the create case, intended to become a directory.
826 * If it isn't we break without incrementing ptr and fall through
827 * to the failure case below.
829 while (*ptr == '/') {
830 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
831 !(nd->nl_flags & NLC_WILLBEDIR)
839 * Continuation case: additional elements and the current
840 * element is a directory.
842 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
843 cache_drop(&nd->nl_nch);
845 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
851 * Failure case: additional elements and the current element
861 * Successful lookup of last element.
863 * Check permissions if the target exists. If the target does not
864 * exist directory permissions were already tested in the early
865 * completion code above.
867 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
868 * if the file is marked append-only, and NLC_STICKY if the directory
869 * containing the file is sticky.
871 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
872 error = naccess(&nch, nd->nl_flags | dflags,
881 * Termination: no more elements.
883 * If NLC_REFDVP is set acquire a referenced parent dvp.
885 if (nd->nl_flags & NLC_REFDVP) {
886 cache_lock(&nd->nl_nch);
887 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
888 cache_unlock(&nd->nl_nch);
890 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
895 cache_drop(&nd->nl_nch);
897 nd->nl_flags |= NLC_NCPISLOCKED;
903 ++gd->gd_nchstats->ncs_longhits;
905 ++gd->gd_nchstats->ncs_longmiss;
907 if (nd->nl_flags & NLC_NCPISLOCKED)
908 KKASSERT(cache_lockstatus(&nd->nl_nch) > 0);
911 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
912 * (ncp->nc_error will be ENOENT), but we will still return an error
919 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
920 * of continuity in the namecache tree by connecting the ncp related to the
921 * vnode under the mount to the ncp related to the mount's root vnode.
923 * If no error occured a locked, ref'd ncp is stored in *ncpp.
926 nlookup_mp(struct mount *mp, struct nchandle *nch)
932 cache_get(&mp->mnt_ncmountpt, nch);
933 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
934 while (vfs_busy(mp, 0))
936 error = VFS_ROOT(mp, &vp);
941 cache_setvp(nch, vp);
949 * Read the contents of a symlink, allocate a path buffer out of the
950 * namei_oc and initialize the supplied nlcomponent with the result.
952 * If an error occurs no buffer will be allocated or returned in the nlc.
955 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
956 struct nlcomponent *nlc)
965 nlc->nlc_nameptr = NULL;
966 nlc->nlc_namelen = 0;
967 if (nch->ncp->nc_vp == NULL)
969 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
971 cp = objcache_get(namei_oc, M_WAITOK);
973 aiov.iov_len = MAXPATHLEN;
974 auio.uio_iov = &aiov;
977 auio.uio_rw = UIO_READ;
978 auio.uio_segflg = UIO_SYSSPACE;
979 auio.uio_td = nd->nl_td;
980 auio.uio_resid = MAXPATHLEN - 1;
981 error = VOP_READLINK(vp, &auio, nd->nl_cred);
984 linklen = MAXPATHLEN - 1 - auio.uio_resid;
986 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
988 error = ENAMETOOLONG;
993 nlc->nlc_nameptr = cp;
994 nlc->nlc_namelen = linklen;
998 objcache_put(namei_oc, cp);
1004 * Check access [XXX cache vattr!] [XXX quota]
1006 * Generally check the NLC_* access bits. All specified bits must pass
1007 * for this function to return 0.
1009 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
1010 * access, otherwise it must exist. No error is returned in this case.
1012 * The file must not exist if NLC_EXCL is specified.
1014 * Directory permissions in general are tested for NLC_CREATE if the file
1015 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
1016 * whether the file exists or not.
1018 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
1019 * the latter is only tested if the target exists.
1021 * The passed ncp must be referenced and locked. If it is already resolved
1022 * it may be locked shared but otherwise should be locked exclusively.
1025 naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
1029 struct namecache *ncp;
1033 KKASSERT(cache_lockstatus(nch) > 0);
1036 if (ncp->nc_flag & NCF_UNRESOLVED) {
1037 cache_resolve(nch, cred);
1040 error = ncp->nc_error;
1043 * Directory permissions checks. Silently ignore ENOENT if these
1044 * tests pass. It isn't an error.
1046 * We can safely resolve ncp->nc_parent because ncp is currently
1049 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
1050 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) ||
1051 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) ||
1052 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) ||
1053 (nflags & NLC_RENAME_DST)
1055 struct nchandle par;
1057 if ((par.ncp = ncp->nc_parent) == NULL) {
1058 if (error != EAGAIN)
1060 } else if (error == 0 || error == ENOENT) {
1061 par.mount = nch->mount;
1063 cache_lock_maybe_shared(&par, 0);
1064 error = naccess(&par, NLC_WRITE, cred, NULL);
1071 * NLC_EXCL check. Target file must not exist.
1073 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL)
1077 * Get the vnode attributes so we can do the rest of our checks.
1079 * NOTE: We only call naccess_va() if the target exists.
1082 error = cache_vget(nch, cred, LK_SHARED, &vp);
1083 if (error == ENOENT) {
1085 * Silently zero-out ENOENT if creating or renaming
1086 * (rename target). It isn't an error.
1088 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
1090 } else if (error == 0) {
1092 * Get the vnode attributes and check for illegal O_TRUNC
1093 * requests and read-only mounts.
1095 * NOTE: You can still open devices on read-only mounts for
1098 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
1099 * check on the parent directory above.
1101 * XXX cache the va in the namecache or in the vnode
1103 error = VOP_GETATTR(vp, &va);
1104 if (error == 0 && (nflags & NLC_TRUNCATE)) {
1105 switch(va.va_type) {
1120 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
1121 (vp->v_mount->mnt_flag & MNT_RDONLY)
1123 switch(va.va_type) {
1137 * Check permissions based on file attributes. The passed
1138 * flags (*nflagsp) are modified with feedback based on
1139 * special attributes and requirements.
1143 * Adjust the returned (*nflagsp) if non-NULL.
1146 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
1147 *nflagsp |= NLC_STICKY;
1148 if (va.va_flags & APPEND)
1149 *nflagsp |= NLC_APPENDONLY;
1150 if (va.va_flags & IMMUTABLE)
1151 *nflagsp |= NLC_IMMUTABLE;
1155 * Track swapcache management flags in the namecache.
1157 * Calculate the flags based on the current vattr info
1158 * and recalculate the inherited flags from the parent
1159 * (the original cache linkage may have occurred without
1160 * getattrs and thus have stale flags).
1163 if (va.va_flags & SF_NOCACHE)
1164 cflags |= NCF_SF_NOCACHE;
1165 if (va.va_flags & UF_CACHE)
1166 cflags |= NCF_UF_CACHE;
1167 if (ncp->nc_parent) {
1168 if (ncp->nc_parent->nc_flag &
1169 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) {
1170 cflags |= NCF_SF_PNOCACHE;
1172 if (ncp->nc_parent->nc_flag &
1173 (NCF_UF_CACHE | NCF_UF_PCACHE)) {
1174 cflags |= NCF_UF_PCACHE;
1179 * XXX we're not supposed to update nc_flag when holding
1182 atomic_clear_short(&ncp->nc_flag,
1183 (NCF_SF_NOCACHE | NCF_UF_CACHE |
1184 NCF_SF_PNOCACHE | NCF_UF_PCACHE) & ~cflags);
1185 atomic_set_short(&ncp->nc_flag, cflags);
1188 * Process general access.
1190 error = naccess_va(&va, nflags, cred);
1198 * Check the requested access against the given vattr using cred.
1201 naccess_va(struct vattr *va, int nflags, struct ucred *cred)
1207 * Test the immutable bit. Creations, deletions, renames (source
1208 * or destination) are not allowed. chown/chmod/other is also not
1209 * allowed but is handled by SETATTR. Hardlinks to the immutable
1212 * If the directory is set to immutable then creations, deletions,
1213 * renames (source or dest) and hardlinks to files within the directory
1214 * are not allowed, and regular files opened through the directory may
1215 * not be written to or truncated (unless a special device).
1217 * NOTE! New hardlinks to immutable files work but new hardlinks to
1218 * files, immutable or not, sitting inside an immutable directory are
1219 * not allowed. As always if the file is hardlinked via some other
1220 * path additional hardlinks may be possible even if the file is marked
1221 * immutable. The sysop needs to create a closure by checking the hard
1222 * link count. Once closure is achieved you are good, and security
1223 * scripts should check link counts anyway.
1225 * Writes and truncations are only allowed on special devices.
1227 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1228 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1230 if (nflags & (NLC_CREATE | NLC_DELETE |
1231 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1234 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1235 switch(va->va_type) {
1249 * Test the no-unlink and append-only bits for opens, rename targets,
1250 * and deletions. These bits are not tested for creations or
1253 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1254 * If you do not wish this you must also set NOUNLINK.
1256 * If the governing directory is marked APPEND-only it implies
1257 * NOUNLINK for all entries in the directory.
1259 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1260 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1266 * A file marked append-only may not be deleted but can be renamed.
1268 if ((va->va_flags & APPEND) &&
1269 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1275 * A file marked append-only which is opened for writing must also
1276 * be opened O_APPEND.
1278 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1279 if (nflags & NLC_TRUNCATE)
1281 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1282 if ((nflags & NLC_APPEND) == 0)
1288 * root gets universal access
1290 if (cred->cr_uid == 0)
1294 * Check owner perms.
1296 * If NLC_OWN is set the owner of the file is allowed no matter when
1297 * the owner-mode bits say (utimes).
1300 if (nflags & NLC_READ)
1302 if (nflags & NLC_WRITE)
1304 if (nflags & NLC_EXEC)
1307 if (cred->cr_uid == va->va_uid) {
1308 if ((nflags & NLC_OWN) == 0) {
1309 if ((vmode & va->va_mode) != vmode)
1316 * If NLC_STICKY is set only the owner may delete or rename a file.
1317 * This bit is typically set on /tmp.
1319 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1320 * the specific delete or rename case. For deletions and renames we
1321 * usually just care about directory permissions, not file permissions.
1323 if ((nflags & NLC_STICKY) &&
1324 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
1332 for (i = 0; i < cred->cr_ngroups; ++i) {
1333 if (va->va_gid == cred->cr_groups[i]) {
1334 if ((vmode & va->va_mode) != vmode)
1344 if ((vmode & va->va_mode) != vmode)