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 1.9 2004/12/27 20:30:03 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).
78 nlookup_init(struct nlookupdata *nd,
79 const char *path, enum uio_seg seg, int flags)
90 * note: the pathlen set by copy*str() includes the terminating \0.
92 bzero(nd, sizeof(struct nlookupdata));
93 nd->nl_path = zalloc(namei_zone);
94 nd->nl_flags |= NLC_HASBUF;
95 if (seg == UIO_SYSSPACE)
96 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
98 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
101 * Don't allow empty pathnames.
102 * POSIX.1 requirement: "" is not a vaild file name.
104 if (error == 0 && pathlen <= 1)
109 nd->nl_ncp = cache_hold(p->p_fd->fd_ncdir);
110 nd->nl_rootncp = cache_hold(p->p_fd->fd_nrdir);
111 if (p->p_fd->fd_njdir)
112 nd->nl_jailncp = cache_hold(p->p_fd->fd_njdir);
113 nd->nl_cred = crhold(p->p_ucred);
115 nd->nl_ncp = cache_hold(rootncp);
116 nd->nl_rootncp = cache_hold(nd->nl_ncp);
117 nd->nl_jailncp = cache_hold(nd->nl_ncp);
118 nd->nl_cred = crhold(proc0.p_ucred);
121 nd->nl_flags |= flags;
129 * This works similarly to nlookup_init() but does not assume a process
130 * context. rootncp is always chosen for the root directory and the cred
131 * and starting directory are supplied in arguments.
134 nlookup_init_raw(struct nlookupdata *nd,
135 const char *path, enum uio_seg seg, int flags,
136 struct ucred *cred, struct namecache *ncstart)
144 bzero(nd, sizeof(struct nlookupdata));
145 nd->nl_path = zalloc(namei_zone);
146 nd->nl_flags |= NLC_HASBUF;
147 if (seg == UIO_SYSSPACE)
148 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
150 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
153 * Don't allow empty pathnames.
154 * POSIX.1 requirement: "" is not a vaild file name.
156 if (error == 0 && pathlen <= 1)
160 nd->nl_ncp = cache_hold(ncstart);
161 nd->nl_rootncp = cache_hold(rootncp);
162 nd->nl_jailncp = cache_hold(rootncp);
163 nd->nl_cred = crhold(cred);
165 nd->nl_flags |= flags;
173 * Cleanup a nlookupdata structure after we are through with it. This may
174 * be called on any nlookupdata structure initialized with nlookup_init().
175 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
176 * returns an error, even if as a consumer you believe you have taken all
177 * dynamic elements out of the nlookupdata structure.
180 nlookup_done(struct nlookupdata *nd)
183 if (nd->nl_flags & NLC_NCPISLOCKED) {
184 nd->nl_flags &= ~NLC_NCPISLOCKED;
185 cache_unlock(nd->nl_ncp);
187 cache_drop(nd->nl_ncp);
190 if (nd->nl_rootncp) {
191 cache_drop(nd->nl_rootncp);
192 nd->nl_rootncp = NULL;
194 if (nd->nl_jailncp) {
195 cache_drop(nd->nl_jailncp);
196 nd->nl_jailncp = NULL;
198 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
199 zfree(namei_zone, nd->nl_path);
206 if (nd->nl_open_vp) {
207 if (nd->nl_flags & NLC_LOCKVP) {
208 VOP_UNLOCK(nd->nl_open_vp, 0, nd->nl_td);
209 nd->nl_flags &= ~NLC_LOCKVP;
211 vn_close(nd->nl_open_vp, nd->nl_vp_fmode, nd->nl_td);
212 nd->nl_open_vp = NULL;
214 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
218 nlookup_zero(struct nlookupdata *nd)
220 bzero(nd, sizeof(struct nlookupdata));
224 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
225 * if an error occured.
227 * Note that the returned ncp is not checked for permissions, though VEXEC
228 * is checked on the directory path leading up to the result. The caller
229 * must call naccess() to check the permissions of the returned leaf.
232 nlookup_simple(const char *str, enum uio_seg seg,
233 int niflags, int *error)
235 struct nlookupdata nd;
236 struct namecache *ncp;
238 *error = nlookup_init(&nd, str, seg, niflags);
240 if ((*error = nlookup(&nd)) == 0) {
241 ncp = nd.nl_ncp; /* keep hold ref from structure */
242 nd.nl_ncp = NULL; /* and NULL out */
254 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
255 * on return, even if an error occurs. If no error occurs the returned
256 * nl_ncp is always referenced and locked, otherwise it may or may not be.
258 * Intermediate directory elements, including the current directory, require
259 * execute (search) permission. nlookup does not examine the access
260 * permissions on the returned element.
262 * If NLC_CREATE or NLC_DELETE is set the last directory must allow node
263 * creation (VCREATE/VDELETE), and an error code of 0 will be returned for
264 * a non-existant target. Otherwise a non-existant target will cause
265 * ENOENT to be returned.
268 nlookup(struct nlookupdata *nd)
270 struct nlcomponent nlc;
271 struct namecache *ncp;
279 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
280 ktrnamei(nd->nl_td->td_proc->p_tracep, nd->nl_path);
282 bzero(&nlc, sizeof(nlc));
285 * Setup for the loop. The current working namecache element must
286 * be in a refd + unlocked state. This typically the case on entry except
287 * when stringing nlookup()'s along in a chain, since nlookup() always
288 * returns nl_ncp in a locked state.
291 if (nd->nl_flags & NLC_NCPISLOCKED) {
292 nd->nl_flags &= ~NLC_NCPISLOCKED;
293 cache_unlock(nd->nl_ncp);
298 * Loop on the path components. At the top of the loop nd->nl_ncp
299 * is ref'd and unlocked and represents our current position.
303 * Check if the root directory should replace the current
304 * directory. This is done at the start of a translation
305 * or after a symbolic link has been found. In other cases
306 * ptr will never be pointing at a '/'.
311 } while (*ptr == '/');
312 ncp = cache_hold(nd->nl_rootncp);
313 cache_drop(nd->nl_ncp);
316 cache_lock(nd->nl_ncp);
317 nd->nl_flags |= NLC_NCPISLOCKED;
325 * Check directory search permissions
327 if ((error = naccess(nd->nl_ncp, VEXEC, nd->nl_cred)) != 0)
331 * Extract the path component
333 nlc.nlc_nameptr = ptr;
334 while (*ptr && *ptr != '/')
336 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
339 * Lookup the path component in the cache, creating an unresolved
340 * entry if necessary. We have to handle "." and ".." as special
343 * When handling ".." we have to detect a traversal back through a
344 * mount point and skip the mount-under node. If we are at the root
345 * ".." just returns the root.
347 * This subsection returns a locked, refd 'ncp' unless it errors out.
348 * The namecache topology is not allowed to be disconnected, so
349 * encountering a NULL parent will generate EINVAL. This typically
350 * occurs when a directory is removed out from under a process.
352 * If NLC_DELETE is set neither '.' or '..' can be the last component
355 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
356 ncp = cache_get(nd->nl_ncp);
358 } else if (nlc.nlc_namelen == 2 &&
359 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
361 if (ncp == nd->nl_rootncp) {
362 ncp = cache_get(ncp);
364 while ((ncp->nc_flag & NCF_MOUNTPT) && ncp != nd->nl_rootncp) {
365 if (ncp->nc_parent->nc_flag & NCF_DESTROYED)
367 ncp = ncp->nc_parent; /* get to underlying node */
368 KKASSERT(ncp != NULL && 1);
370 if (ncp != nd->nl_rootncp) {
371 if (ncp->nc_parent->nc_flag & NCF_DESTROYED) {
375 ncp = ncp->nc_parent;
381 ncp = cache_get(ncp);
385 ncp = cache_nlookup(nd->nl_ncp, &nlc);
386 while ((error = cache_resolve(ncp, nd->nl_cred)) == EAGAIN) {
387 printf("[diagnostic] nlookup: relookup %*.*s\n",
388 ncp->nc_nlen, ncp->nc_nlen, ncp->nc_name);
390 ncp = cache_nlookup(nd->nl_ncp, &nlc);
395 * [end of subsection] ncp is locked and ref'd. nd->nl_ncp is ref'd
399 * Resolve the namespace if necessary. The ncp returned by
400 * cache_nlookup() is referenced and locked.
402 * XXX neither '.' nor '..' should return EAGAIN since they were
403 * previously resolved and thus cannot be newly created ncp's.
405 if (ncp->nc_flag & NCF_UNRESOLVED) {
406 error = cache_resolve(ncp, nd->nl_cred);
407 KKASSERT(error != EAGAIN);
409 error = ncp->nc_error;
413 * Early completion. ENOENT is not an error if this is the last
414 * component and NLC_CREATE was requested. Note that ncp->nc_error
415 * is left as ENOENT in that case, which we check later on.
417 * Also handle invalid '.' or '..' components terminating a path
418 * during removal. The standard requires this and pax pretty
419 *stupidly depends on it.
421 for (xptr = ptr; *xptr == '/'; ++xptr)
424 if (error == ENOENT && (nd->nl_flags & NLC_CREATE))
425 error = naccess(ncp, VCREATE, nd->nl_cred);
426 if (error == 0 && wasdotordotdot && (nd->nl_flags & NLC_DELETE))
431 * Early completion on error.
439 * If the element is a symlink and it is either not the last
440 * element or it is the last element and we are allowed to
441 * follow symlinks, resolve the symlink.
443 if ((ncp->nc_flag & NCF_ISSYMLINK) &&
444 (*ptr || (nd->nl_flags & NLC_FOLLOW))
446 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
451 error = nreadsymlink(nd, ncp, &nlc);
457 * Concatenate trailing path elements onto the returned symlink.
458 * Note that if the path component (ptr) is not exhausted, it
459 * will being with a '/', so we do not have to add another one.
461 * The symlink may not be empty.
464 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
465 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
466 zfree(namei_zone, nlc.nlc_nameptr);
469 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
470 if (nd->nl_flags & NLC_HASBUF)
471 zfree(namei_zone, nd->nl_path);
472 nd->nl_path = nlc.nlc_nameptr;
473 nd->nl_flags |= NLC_HASBUF;
477 * Go back up to the top to resolve any initial '/'s in the
484 * If the element is a directory and we are crossing a mount point,
485 * retrieve the root of the mounted filesystem from mnt_ncp and
486 * resolve it if necessary.
488 * XXX mnt_ncp should really be resolved in the mount code.
489 * NOTE! the normal nresolve() code cannot resolve mount point ncp's!
493 while ((ncp->nc_flag & NCF_ISDIR) && ncp->nc_vp->v_mountedhere &&
494 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0
499 mp = ncp->nc_vp->v_mountedhere;
501 ncp = cache_get(mp->mnt_ncp);
503 if (ncp->nc_flag & NCF_UNRESOLVED) {
504 while (vfs_busy(mp, 0, NULL, nd->nl_td))
506 error = VFS_ROOT(mp, &tdp);
507 vfs_unbusy(mp, nd->nl_td);
510 cache_setvp(ncp, tdp);
520 * Skip any slashes to get to the next element. If there
521 * are any slashes at all the current element must be a
522 * directory or, in the create case, intended to become a directory.
523 * If it isn't we break without incrementing ptr and fall through
524 * to the failure case below.
526 while (*ptr == '/') {
527 if ((ncp->nc_flag & NCF_ISDIR) == 0 &&
528 !(nd->nl_flags & NLC_WILLBEDIR)
536 * Continuation case: additional elements and the current
537 * element is a directory.
539 if (*ptr && (ncp->nc_flag & NCF_ISDIR)) {
540 cache_drop(nd->nl_ncp);
547 * Failure case: additional elements and the current element
557 * Successful lookup of last element.
559 * Check directory permissions if a deletion is specified.
561 if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
562 if ((error = naccess(ncp, VDELETE, nd->nl_cred)) != 0) {
569 * XXX vnode canvmio (test in mmap(), read(), and write())
573 * Termination: no more elements. If NLC_CREATE was set the
574 * ncp may represent a negative hit (ncp->nc_error will be ENOENT),
575 * but we still return an error code of 0.
577 cache_drop(nd->nl_ncp);
579 nd->nl_flags |= NLC_NCPISLOCKED;
587 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
588 * of continuity in the namecache tree by connecting the ncp related to the
589 * vnode under the mount to the ncp related to the mount's root vnode.
591 * If no error occured a locked, ref'd ncp is stored in *ncpp.
594 nlookup_mp(struct mount *mp, struct namecache **ncpp)
596 struct namecache *ncp;
603 if (ncp->nc_flag & NCF_UNRESOLVED) {
604 while (vfs_busy(mp, 0, NULL, curthread))
606 error = VFS_ROOT(mp, &vp);
607 vfs_unbusy(mp, curthread);
612 cache_setvp(ncp, vp);
621 * Read the contents of a symlink, allocate a path buffer out of the
622 * namei_zone and initialize the supplied nlcomponent with the result.
624 * If an error occurs no buffer will be allocated or returned in the nlc.
627 nreadsymlink(struct nlookupdata *nd, struct namecache *ncp,
628 struct nlcomponent *nlc)
637 nlc->nlc_nameptr = NULL;
638 nlc->nlc_namelen = 0;
639 if (ncp->nc_vp == NULL)
641 if ((error = cache_vget(ncp, nd->nl_cred, LK_SHARED, &vp)) != 0)
643 cp = zalloc(namei_zone);
645 aiov.iov_len = MAXPATHLEN;
646 auio.uio_iov = &aiov;
649 auio.uio_rw = UIO_READ;
650 auio.uio_segflg = UIO_SYSSPACE;
651 auio.uio_td = nd->nl_td;
652 auio.uio_resid = MAXPATHLEN - 1;
653 error = VOP_READLINK(vp, &auio, nd->nl_cred);
656 linklen = MAXPATHLEN - 1 - auio.uio_resid;
658 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
660 error = ENAMETOOLONG;
665 nlc->nlc_nameptr = cp;
666 nlc->nlc_namelen = linklen;
670 zfree(namei_zone, cp);
676 * Check access [XXX cache vattr!] [XXX quota]
678 * Generally check the V* access bits from sys/vnode.h. All specified bits
679 * must pass for this function to return 0.
681 * If VCREATE is specified and the target ncp represents a non-existant
682 * file or dir, or if VDELETE is specified and the target exists, the parent
683 * directory is checked for VWRITE. If VEXCL is specified and the target
684 * ncp represents a positive hit, an error is returned.
686 * If VCREATE is not specified and the target does not exist (negative hit),
687 * ENOENT is returned. Note that nlookup() does not (and should not) return
688 * ENOENT for non-existant leafs.
690 * The passed ncp may or may not be locked. The caller should use a
691 * locked ncp on leaf lookups, especially for VCREATE, VDELETE, and VEXCL
695 naccess(struct namecache *ncp, int vmode, struct ucred *cred)
697 struct namecache *par;
702 if (ncp->nc_flag & NCF_UNRESOLVED) {
704 cache_resolve(ncp, cred);
707 error = ncp->nc_error;
708 if (vmode & (VDELETE|VCREATE|VEXCL)) {
709 if (((vmode & VCREATE) && ncp->nc_vp == NULL) ||
710 ((vmode & VDELETE) && ncp->nc_vp != NULL)
712 if ((par = ncp->nc_parent) == NULL) {
717 error = naccess(par, VWRITE, cred);
721 if ((vmode & VEXCL) && ncp->nc_vp != NULL)
725 error = cache_vget(ncp, cred, LK_SHARED, &vp);
726 if (error == ENOENT) {
729 } else if (error == 0) {
730 /* XXX cache the va in the namecache or in the vnode */
731 if ((error = VOP_GETATTR(vp, &va, curthread)) == 0) {
732 if ((vmode & VWRITE) && vp->v_mount) {
733 if (vp->v_mount->mnt_flag & MNT_RDONLY)
739 error = naccess_va(&va, vmode, cred);
746 * Check the requested access against the given vattr using cred.
749 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
754 * Test the immutable bit for files, directories, and softlinks.
756 if (vmode & (VWRITE|VDELETE)) {
757 if (va->va_type == VDIR || va->va_type == VLNK || va->va_type == VREG) {
758 if (va->va_flags & IMMUTABLE)
764 * root gets universal access
766 if (cred->cr_uid == 0)
770 * Check owner perms, group perms, and world perms
773 if (cred->cr_uid == va->va_uid) {
774 if ((vmode & va->va_mode) != vmode)
780 for (i = 0; i < cred->cr_ngroups; ++i) {
781 if (va->va_gid == cred->cr_groups[i]) {
782 if ((vmode & va->va_mode) != vmode)
789 if ((vmode & va->va_mode) != vmode)