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 1.22 2007/05/09 05:12:45 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>
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 = objcache_get(namei_oc, M_WAITOK);
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 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch);
113 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch);
114 if (p->p_fd->fd_njdir.ncp)
115 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch);
116 nd->nl_cred = crhold(p->p_ucred);
118 cache_copy(&rootnch, &nd->nl_nch);
119 cache_copy(&nd->nl_nch, &nd->nl_rootnch);
120 cache_copy(&nd->nl_nch, &nd->nl_jailnch);
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. rootnch 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 nchandle *ncstart)
147 bzero(nd, sizeof(struct nlookupdata));
148 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
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 cache_copy(ncstart, &nd->nl_nch);
164 cache_copy(&rootnch, &nd->nl_rootnch);
165 cache_copy(&rootnch, &nd->nl_jailnch);
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)
201 if (nd->nl_nch.ncp) {
202 if (nd->nl_flags & NLC_NCPISLOCKED) {
203 nd->nl_flags &= ~NLC_NCPISLOCKED;
204 cache_unlock(&nd->nl_nch);
206 cache_drop(&nd->nl_nch);
208 if (nd->nl_rootnch.ncp)
209 cache_drop(&nd->nl_rootnch);
210 if (nd->nl_jailnch.ncp)
211 cache_drop(&nd->nl_jailnch);
212 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
213 objcache_put(namei_oc, nd->nl_path);
220 if (nd->nl_open_vp) {
221 if (nd->nl_flags & NLC_LOCKVP) {
222 vn_unlock(nd->nl_open_vp);
223 nd->nl_flags &= ~NLC_LOCKVP;
225 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
226 nd->nl_open_vp = NULL;
228 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
232 nlookup_zero(struct nlookupdata *nd)
234 bzero(nd, sizeof(struct nlookupdata));
238 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
239 * if an error occured.
241 * Note that the returned ncp is not checked for permissions, though VEXEC
242 * is checked on the directory path leading up to the result. The caller
243 * must call naccess() to check the permissions of the returned leaf.
246 nlookup_simple(const char *str, enum uio_seg seg,
247 int niflags, int *error)
249 struct nlookupdata nd;
252 *error = nlookup_init(&nd, str, seg, niflags);
254 if ((*error = nlookup(&nd)) == 0) {
255 nch = nd.nl_nch; /* keep hold ref from structure */
256 cache_zero(&nd.nl_nch); /* and NULL out */
268 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
269 * on return, even if an error occurs. If no error occurs the returned
270 * nl_nch is always referenced and locked, otherwise it may or may not be.
272 * Intermediate directory elements, including the current directory, require
273 * execute (search) permission. nlookup does not examine the access
274 * permissions on the returned element.
276 * If NLC_CREATE or NLC_DELETE is set the last directory must allow node
277 * creation (VCREATE/VDELETE), and an error code of 0 will be returned for
278 * a non-existant target. Otherwise a non-existant target will cause
279 * ENOENT to be returned.
282 nlookup(struct nlookupdata *nd)
284 struct nlcomponent nlc;
294 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
295 ktrnamei(nd->nl_td->td_proc, nd->nl_path);
297 bzero(&nlc, sizeof(nlc));
300 * Setup for the loop. The current working namecache element must
301 * be in a refd + unlocked state. This typically the case on entry except
302 * when stringing nlookup()'s along in a chain, since nlookup() always
303 * returns nl_nch in a locked state.
306 if (nd->nl_flags & NLC_NCPISLOCKED) {
307 nd->nl_flags &= ~NLC_NCPISLOCKED;
308 cache_unlock(&nd->nl_nch);
313 * Loop on the path components. At the top of the loop nd->nl_nch
314 * is ref'd and unlocked and represents our current position.
318 * Check if the root directory should replace the current
319 * directory. This is done at the start of a translation
320 * or after a symbolic link has been found. In other cases
321 * ptr will never be pointing at a '/'.
326 } while (*ptr == '/');
327 cache_copy(&nd->nl_rootnch, &nch);
328 cache_drop(&nd->nl_nch);
331 cache_lock(&nd->nl_nch);
332 nd->nl_flags |= NLC_NCPISLOCKED;
340 * Check directory search permissions.
342 if ((error = naccess(&nd->nl_nch, VEXEC, nd->nl_cred)) != 0)
346 * Extract the path component
348 nlc.nlc_nameptr = ptr;
349 while (*ptr && *ptr != '/')
351 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
354 * Lookup the path component in the cache, creating an unresolved
355 * entry if necessary. We have to handle "." and ".." as special
358 * When handling ".." we have to detect a traversal back through a
359 * mount point. If we are at the root, ".." just returns the root.
361 * This subsection returns a locked, refd 'nch' unless it errors out.
362 * The namecache topology is not allowed to be disconnected, so
363 * encountering a NULL parent will generate EINVAL. This typically
364 * occurs when a directory is removed out from under a process.
366 * If NLC_DELETE is set neither '.' or '..' can be the last component
369 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
370 cache_get(&nd->nl_nch, &nch);
372 } else if (nlc.nlc_namelen == 2 &&
373 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
374 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
375 nd->nl_nch.ncp == nd->nl_rootnch.ncp
378 * ".." at the root returns the root
380 cache_get(&nd->nl_nch, &nch);
383 * Locate the parent ncp. If we are at the root of a
384 * filesystem mount we have to skip to the mounted-on
385 * point in the underlying filesystem.
388 while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
389 nch = nch.mount->mnt_ncmounton;
390 nch.ncp = nch.ncp->nc_parent;
391 KKASSERT(nch.ncp != NULL);
392 cache_get(&nch, &nch);
396 nch = cache_nlookup(&nd->nl_nch, &nlc);
397 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
398 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
399 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
401 nch = cache_nlookup(&nd->nl_nch, &nlc);
406 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
410 * Resolve the namespace if necessary. The ncp returned by
411 * cache_nlookup() is referenced and locked.
413 * XXX neither '.' nor '..' should return EAGAIN since they were
414 * previously resolved and thus cannot be newly created ncp's.
416 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
417 error = cache_resolve(&nch, nd->nl_cred);
418 KKASSERT(error != EAGAIN);
420 error = nch.ncp->nc_error;
424 * Early completion. ENOENT is not an error if this is the last
425 * component and NLC_CREATE was requested. Note that ncp->nc_error
426 * is left as ENOENT in that case, which we check later on.
428 * Also handle invalid '.' or '..' components terminating a path
429 * during removal. The standard requires this and pax pretty
430 *stupidly depends on it.
432 for (xptr = ptr; *xptr == '/'; ++xptr)
435 if (error == ENOENT && (nd->nl_flags & NLC_CREATE))
436 error = naccess(&nch, VCREATE, nd->nl_cred);
437 if (error == 0 && wasdotordotdot && (nd->nl_flags & NLC_DELETE))
442 * Early completion on error.
450 * If the element is a symlink and it is either not the last
451 * element or it is the last element and we are allowed to
452 * follow symlinks, resolve the symlink.
454 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
455 (*ptr || (nd->nl_flags & NLC_FOLLOW))
457 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
462 error = nreadsymlink(nd, &nch, &nlc);
468 * Concatenate trailing path elements onto the returned symlink.
469 * Note that if the path component (ptr) is not exhausted, it
470 * will being with a '/', so we do not have to add another one.
472 * The symlink may not be empty.
475 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
476 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
477 objcache_put(namei_oc, nlc.nlc_nameptr);
480 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
481 if (nd->nl_flags & NLC_HASBUF)
482 objcache_put(namei_oc, nd->nl_path);
483 nd->nl_path = nlc.nlc_nameptr;
484 nd->nl_flags |= NLC_HASBUF;
488 * Go back up to the top to resolve any initial '/'s in the
495 * If the element is a directory and we are crossing a mount point,
498 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
499 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
500 (mp = cache_findmount(&nch)) != NULL
505 cache_get(&mp->mnt_ncmountpt, &nch);
507 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
508 while (vfs_busy(mp, 0))
510 error = VFS_ROOT(mp, &tdp);
514 cache_setvp(&nch, tdp);
524 * Skip any slashes to get to the next element. If there
525 * are any slashes at all the current element must be a
526 * directory or, in the create case, intended to become a directory.
527 * If it isn't we break without incrementing ptr and fall through
528 * to the failure case below.
530 while (*ptr == '/') {
531 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
532 !(nd->nl_flags & NLC_WILLBEDIR)
540 * Continuation case: additional elements and the current
541 * element is a directory.
543 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
544 cache_drop(&nd->nl_nch);
551 * Failure case: additional elements and the current element
561 * Successful lookup of last element.
563 * Check directory permissions if a deletion is specified.
565 if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
566 if ((error = naccess(&nch, VDELETE, nd->nl_cred)) != 0) {
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_nch);
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 nchandle *nch)
600 cache_get(&mp->mnt_ncmountpt, nch);
601 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
602 while (vfs_busy(mp, 0))
604 error = VFS_ROOT(mp, &vp);
609 cache_setvp(nch, vp);
617 * Read the contents of a symlink, allocate a path buffer out of the
618 * namei_oc and initialize the supplied nlcomponent with the result.
620 * If an error occurs no buffer will be allocated or returned in the nlc.
623 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
624 struct nlcomponent *nlc)
633 nlc->nlc_nameptr = NULL;
634 nlc->nlc_namelen = 0;
635 if (nch->ncp->nc_vp == NULL)
637 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
639 cp = objcache_get(namei_oc, M_WAITOK);
641 aiov.iov_len = MAXPATHLEN;
642 auio.uio_iov = &aiov;
645 auio.uio_rw = UIO_READ;
646 auio.uio_segflg = UIO_SYSSPACE;
647 auio.uio_td = nd->nl_td;
648 auio.uio_resid = MAXPATHLEN - 1;
649 error = VOP_READLINK(vp, &auio, nd->nl_cred);
652 linklen = MAXPATHLEN - 1 - auio.uio_resid;
654 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
656 error = ENAMETOOLONG;
661 nlc->nlc_nameptr = cp;
662 nlc->nlc_namelen = linklen;
666 objcache_put(namei_oc, cp);
672 * Check access [XXX cache vattr!] [XXX quota]
674 * Generally check the V* access bits from sys/vnode.h. All specified bits
675 * must pass for this function to return 0.
677 * If VCREATE is specified and the target ncp represents a non-existant
678 * file or dir, or if VDELETE is specified and the target exists, the parent
679 * directory is checked for VWRITE. If VEXCL is specified and the target
680 * ncp represents a positive hit, an error is returned.
682 * If VCREATE is not specified and the target does not exist (negative hit),
683 * ENOENT is returned. Note that nlookup() does not (and should not) return
684 * ENOENT for non-existant leafs.
686 * The passed ncp may or may not be locked. The caller should use a
687 * locked ncp on leaf lookups, especially for VCREATE, VDELETE, and VEXCL
691 naccess(struct nchandle *nch, int vmode, struct ucred *cred)
698 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
700 cache_resolve(nch, cred);
703 error = nch->ncp->nc_error;
704 if (vmode & (VDELETE|VCREATE|VEXCL)) {
705 if (((vmode & VCREATE) && nch->ncp->nc_vp == NULL) ||
706 ((vmode & VDELETE) && nch->ncp->nc_vp != NULL)
708 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
712 par.mount = nch->mount;
714 error = naccess(&par, VWRITE, cred);
718 if ((vmode & VEXCL) && nch->ncp->nc_vp != NULL)
722 error = cache_vget(nch, cred, LK_SHARED, &vp);
723 if (error == ENOENT) {
726 } else if (error == 0) {
727 /* XXX cache the va in the namecache or in the vnode */
728 if ((error = VOP_GETATTR(vp, &va)) == 0) {
729 if ((vmode & VWRITE) && vp->v_mount) {
730 if (vp->v_mount->mnt_flag & MNT_RDONLY)
736 error = naccess_va(&va, vmode, cred);
743 * Check the requested access against the given vattr using cred.
746 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
751 * Test the immutable bit for files, directories, and softlinks.
753 if (vmode & (VWRITE|VDELETE)) {
754 if (va->va_type == VDIR || va->va_type == VLNK || va->va_type == VREG) {
755 if (va->va_flags & IMMUTABLE)
761 * root gets universal access
763 if (cred->cr_uid == 0)
767 * Check owner perms, group perms, and world perms
770 if (cred->cr_uid == va->va_uid) {
771 if ((vmode & va->va_mode) != vmode)
777 for (i = 0; i < cred->cr_ngroups; ++i) {
778 if (va->va_gid == cred->cr_groups[i]) {
779 if ((vmode & va->va_mode) != vmode)
786 if ((vmode & va->va_mode) != vmode)