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 1.13 2006/05/05 21:15:09 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 <vm/vm_zone.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 = zalloc(namei_zone);
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 nd->nl_ncp = cache_hold(p->p_fd->fd_ncdir);
113 nd->nl_rootncp = cache_hold(p->p_fd->fd_nrdir);
114 if (p->p_fd->fd_njdir)
115 nd->nl_jailncp = cache_hold(p->p_fd->fd_njdir);
116 nd->nl_cred = crhold(p->p_ucred);
118 nd->nl_ncp = cache_hold(rootncp);
119 nd->nl_rootncp = cache_hold(nd->nl_ncp);
120 nd->nl_jailncp = cache_hold(nd->nl_ncp);
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. rootncp 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 namecache *ncstart)
147 bzero(nd, sizeof(struct nlookupdata));
148 nd->nl_path = zalloc(namei_zone);
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 nd->nl_ncp = cache_hold(ncstart);
164 nd->nl_rootncp = cache_hold(rootncp);
165 nd->nl_jailncp = cache_hold(rootncp);
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)
202 if (nd->nl_flags & NLC_NCPISLOCKED) {
203 nd->nl_flags &= ~NLC_NCPISLOCKED;
204 cache_unlock(nd->nl_ncp);
206 cache_drop(nd->nl_ncp);
209 if (nd->nl_rootncp) {
210 cache_drop(nd->nl_rootncp);
211 nd->nl_rootncp = NULL;
213 if (nd->nl_jailncp) {
214 cache_drop(nd->nl_jailncp);
215 nd->nl_jailncp = NULL;
217 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
218 zfree(namei_zone, nd->nl_path);
225 if (nd->nl_open_vp) {
226 if (nd->nl_flags & NLC_LOCKVP) {
227 VOP_UNLOCK(nd->nl_open_vp, 0);
228 nd->nl_flags &= ~NLC_LOCKVP;
230 vn_close(nd->nl_open_vp, nd->nl_vp_fmode, nd->nl_td);
231 nd->nl_open_vp = NULL;
233 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
237 nlookup_zero(struct nlookupdata *nd)
239 bzero(nd, sizeof(struct nlookupdata));
243 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
244 * if an error occured.
246 * Note that the returned ncp is not checked for permissions, though VEXEC
247 * is checked on the directory path leading up to the result. The caller
248 * must call naccess() to check the permissions of the returned leaf.
251 nlookup_simple(const char *str, enum uio_seg seg,
252 int niflags, int *error)
254 struct nlookupdata nd;
255 struct namecache *ncp;
257 *error = nlookup_init(&nd, str, seg, niflags);
259 if ((*error = nlookup(&nd)) == 0) {
260 ncp = nd.nl_ncp; /* keep hold ref from structure */
261 nd.nl_ncp = NULL; /* and NULL out */
273 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
274 * on return, even if an error occurs. If no error occurs the returned
275 * nl_ncp is always referenced and locked, otherwise it may or may not be.
277 * Intermediate directory elements, including the current directory, require
278 * execute (search) permission. nlookup does not examine the access
279 * permissions on the returned element.
281 * If NLC_CREATE or NLC_DELETE is set the last directory must allow node
282 * creation (VCREATE/VDELETE), and an error code of 0 will be returned for
283 * a non-existant target. Otherwise a non-existant target will cause
284 * ENOENT to be returned.
287 nlookup(struct nlookupdata *nd)
289 struct nlcomponent nlc;
290 struct namecache *ncp;
298 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
299 ktrnamei(nd->nl_td->td_proc->p_tracep, nd->nl_path);
301 bzero(&nlc, sizeof(nlc));
304 * Setup for the loop. The current working namecache element must
305 * be in a refd + unlocked state. This typically the case on entry except
306 * when stringing nlookup()'s along in a chain, since nlookup() always
307 * returns nl_ncp in a locked state.
310 if (nd->nl_flags & NLC_NCPISLOCKED) {
311 nd->nl_flags &= ~NLC_NCPISLOCKED;
312 cache_unlock(nd->nl_ncp);
317 * Loop on the path components. At the top of the loop nd->nl_ncp
318 * is ref'd and unlocked and represents our current position.
322 * Check if the root directory should replace the current
323 * directory. This is done at the start of a translation
324 * or after a symbolic link has been found. In other cases
325 * ptr will never be pointing at a '/'.
330 } while (*ptr == '/');
331 ncp = cache_hold(nd->nl_rootncp);
332 cache_drop(nd->nl_ncp);
335 cache_lock(nd->nl_ncp);
336 nd->nl_flags |= NLC_NCPISLOCKED;
344 * Check directory search permissions
346 if ((error = naccess(nd->nl_ncp, VEXEC, nd->nl_cred)) != 0)
350 * Extract the path component
352 nlc.nlc_nameptr = ptr;
353 while (*ptr && *ptr != '/')
355 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
358 * Lookup the path component in the cache, creating an unresolved
359 * entry if necessary. We have to handle "." and ".." as special
362 * When handling ".." we have to detect a traversal back through a
363 * mount point and skip the mount-under node. If we are at the root
364 * ".." just returns the root.
366 * This subsection returns a locked, refd 'ncp' unless it errors out.
367 * The namecache topology is not allowed to be disconnected, so
368 * encountering a NULL parent will generate EINVAL. This typically
369 * occurs when a directory is removed out from under a process.
371 * If NLC_DELETE is set neither '.' or '..' can be the last component
374 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
375 ncp = cache_get(nd->nl_ncp);
377 } else if (nlc.nlc_namelen == 2 &&
378 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
380 if (ncp == nd->nl_rootncp) {
381 ncp = cache_get(ncp);
383 while ((ncp->nc_flag & NCF_MOUNTPT) && ncp != nd->nl_rootncp) {
384 if (ncp->nc_parent->nc_flag & NCF_DESTROYED)
386 ncp = ncp->nc_parent; /* get to underlying node */
387 KKASSERT(ncp != NULL && 1);
389 if (ncp != nd->nl_rootncp) {
390 if (ncp->nc_parent->nc_flag & NCF_DESTROYED) {
394 ncp = ncp->nc_parent;
400 ncp = cache_get(ncp);
404 ncp = cache_nlookup(nd->nl_ncp, &nlc);
405 while ((error = cache_resolve(ncp, nd->nl_cred)) == EAGAIN) {
406 printf("[diagnostic] nlookup: relookup %*.*s\n",
407 ncp->nc_nlen, ncp->nc_nlen, ncp->nc_name);
409 ncp = cache_nlookup(nd->nl_ncp, &nlc);
414 * [end of subsection] ncp is locked and ref'd. nd->nl_ncp is ref'd
418 * Resolve the namespace if necessary. The ncp returned by
419 * cache_nlookup() is referenced and locked.
421 * XXX neither '.' nor '..' should return EAGAIN since they were
422 * previously resolved and thus cannot be newly created ncp's.
424 if (ncp->nc_flag & NCF_UNRESOLVED) {
425 error = cache_resolve(ncp, nd->nl_cred);
426 KKASSERT(error != EAGAIN);
428 error = ncp->nc_error;
432 * Early completion. ENOENT is not an error if this is the last
433 * component and NLC_CREATE was requested. Note that ncp->nc_error
434 * is left as ENOENT in that case, which we check later on.
436 * Also handle invalid '.' or '..' components terminating a path
437 * during removal. The standard requires this and pax pretty
438 *stupidly depends on it.
440 for (xptr = ptr; *xptr == '/'; ++xptr)
443 if (error == ENOENT && (nd->nl_flags & NLC_CREATE))
444 error = naccess(ncp, VCREATE, nd->nl_cred);
445 if (error == 0 && wasdotordotdot && (nd->nl_flags & NLC_DELETE))
450 * Early completion on error.
458 * If the element is a symlink and it is either not the last
459 * element or it is the last element and we are allowed to
460 * follow symlinks, resolve the symlink.
462 if ((ncp->nc_flag & NCF_ISSYMLINK) &&
463 (*ptr || (nd->nl_flags & NLC_FOLLOW))
465 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
470 error = nreadsymlink(nd, ncp, &nlc);
476 * Concatenate trailing path elements onto the returned symlink.
477 * Note that if the path component (ptr) is not exhausted, it
478 * will being with a '/', so we do not have to add another one.
480 * The symlink may not be empty.
483 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
484 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
485 zfree(namei_zone, nlc.nlc_nameptr);
488 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
489 if (nd->nl_flags & NLC_HASBUF)
490 zfree(namei_zone, nd->nl_path);
491 nd->nl_path = nlc.nlc_nameptr;
492 nd->nl_flags |= NLC_HASBUF;
496 * Go back up to the top to resolve any initial '/'s in the
503 * If the element is a directory and we are crossing a mount point,
504 * retrieve the root of the mounted filesystem from mnt_ncp and
505 * resolve it if necessary.
507 * XXX mnt_ncp should really be resolved in the mount code.
508 * NOTE! the normal nresolve() code cannot resolve mount point ncp's!
512 while ((ncp->nc_flag & NCF_ISDIR) && ncp->nc_vp->v_mountedhere &&
513 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0
518 mp = ncp->nc_vp->v_mountedhere;
520 ncp = cache_get(mp->mnt_ncp);
522 if (ncp->nc_flag & NCF_UNRESOLVED) {
523 while (vfs_busy(mp, 0, nd->nl_td))
525 error = VFS_ROOT(mp, &tdp);
526 vfs_unbusy(mp, nd->nl_td);
529 cache_setvp(ncp, tdp);
539 * Skip any slashes to get to the next element. If there
540 * are any slashes at all the current element must be a
541 * directory or, in the create case, intended to become a directory.
542 * If it isn't we break without incrementing ptr and fall through
543 * to the failure case below.
545 while (*ptr == '/') {
546 if ((ncp->nc_flag & NCF_ISDIR) == 0 &&
547 !(nd->nl_flags & NLC_WILLBEDIR)
555 * Continuation case: additional elements and the current
556 * element is a directory.
558 if (*ptr && (ncp->nc_flag & NCF_ISDIR)) {
559 cache_drop(nd->nl_ncp);
566 * Failure case: additional elements and the current element
576 * Successful lookup of last element.
578 * Check directory permissions if a deletion is specified.
580 if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
581 if ((error = naccess(ncp, VDELETE, nd->nl_cred)) != 0) {
588 * Termination: no more elements. If NLC_CREATE was set the
589 * ncp may represent a negative hit (ncp->nc_error will be ENOENT),
590 * but we still return an error code of 0.
592 cache_drop(nd->nl_ncp);
594 nd->nl_flags |= NLC_NCPISLOCKED;
602 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
603 * of continuity in the namecache tree by connecting the ncp related to the
604 * vnode under the mount to the ncp related to the mount's root vnode.
606 * If no error occured a locked, ref'd ncp is stored in *ncpp.
609 nlookup_mp(struct mount *mp, struct namecache **ncpp)
611 struct namecache *ncp;
618 if (ncp->nc_flag & NCF_UNRESOLVED) {
619 while (vfs_busy(mp, 0, curthread))
621 error = VFS_ROOT(mp, &vp);
622 vfs_unbusy(mp, curthread);
627 cache_setvp(ncp, vp);
636 * Read the contents of a symlink, allocate a path buffer out of the
637 * namei_zone and initialize the supplied nlcomponent with the result.
639 * If an error occurs no buffer will be allocated or returned in the nlc.
642 nreadsymlink(struct nlookupdata *nd, struct namecache *ncp,
643 struct nlcomponent *nlc)
652 nlc->nlc_nameptr = NULL;
653 nlc->nlc_namelen = 0;
654 if (ncp->nc_vp == NULL)
656 if ((error = cache_vget(ncp, nd->nl_cred, LK_SHARED, &vp)) != 0)
658 cp = zalloc(namei_zone);
660 aiov.iov_len = MAXPATHLEN;
661 auio.uio_iov = &aiov;
664 auio.uio_rw = UIO_READ;
665 auio.uio_segflg = UIO_SYSSPACE;
666 auio.uio_td = nd->nl_td;
667 auio.uio_resid = MAXPATHLEN - 1;
668 error = VOP_READLINK(vp, &auio, nd->nl_cred);
671 linklen = MAXPATHLEN - 1 - auio.uio_resid;
673 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
675 error = ENAMETOOLONG;
680 nlc->nlc_nameptr = cp;
681 nlc->nlc_namelen = linklen;
685 zfree(namei_zone, cp);
691 * Check access [XXX cache vattr!] [XXX quota]
693 * Generally check the V* access bits from sys/vnode.h. All specified bits
694 * must pass for this function to return 0.
696 * If VCREATE is specified and the target ncp represents a non-existant
697 * file or dir, or if VDELETE is specified and the target exists, the parent
698 * directory is checked for VWRITE. If VEXCL is specified and the target
699 * ncp represents a positive hit, an error is returned.
701 * If VCREATE is not specified and the target does not exist (negative hit),
702 * ENOENT is returned. Note that nlookup() does not (and should not) return
703 * ENOENT for non-existant leafs.
705 * The passed ncp may or may not be locked. The caller should use a
706 * locked ncp on leaf lookups, especially for VCREATE, VDELETE, and VEXCL
710 naccess(struct namecache *ncp, int vmode, struct ucred *cred)
712 struct namecache *par;
717 if (ncp->nc_flag & NCF_UNRESOLVED) {
719 cache_resolve(ncp, cred);
722 error = ncp->nc_error;
723 if (vmode & (VDELETE|VCREATE|VEXCL)) {
724 if (((vmode & VCREATE) && ncp->nc_vp == NULL) ||
725 ((vmode & VDELETE) && ncp->nc_vp != NULL)
727 if ((par = ncp->nc_parent) == NULL) {
732 error = naccess(par, VWRITE, cred);
736 if ((vmode & VEXCL) && ncp->nc_vp != NULL)
740 error = cache_vget(ncp, cred, LK_SHARED, &vp);
741 if (error == ENOENT) {
744 } else if (error == 0) {
745 /* XXX cache the va in the namecache or in the vnode */
746 if ((error = VOP_GETATTR(vp, &va, curthread)) == 0) {
747 if ((vmode & VWRITE) && vp->v_mount) {
748 if (vp->v_mount->mnt_flag & MNT_RDONLY)
754 error = naccess_va(&va, vmode, cred);
761 * Check the requested access against the given vattr using cred.
764 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
769 * Test the immutable bit for files, directories, and softlinks.
771 if (vmode & (VWRITE|VDELETE)) {
772 if (va->va_type == VDIR || va->va_type == VLNK || va->va_type == VREG) {
773 if (va->va_flags & IMMUTABLE)
779 * root gets universal access
781 if (cred->cr_uid == 0)
785 * Check owner perms, group perms, and world perms
788 if (cred->cr_uid == va->va_uid) {
789 if ((vmode & va->va_mode) != vmode)
795 for (i = 0; i < cred->cr_ngroups; ++i) {
796 if (va->va_gid == cred->cr_groups[i]) {
797 if ((vmode & va->va_mode) != vmode)
804 if ((vmode & va->va_mode) != vmode)