xref: /dragonfly/sys/kern/vfs_nlookup.c (revision cc93b0eb)
1 /*
2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
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
16  *    distribution.
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.
20  *
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
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sys/kern/vfs_nlookup.c,v 1.24 2008/05/09 17:52:17 dillon Exp $
35  */
36 /*
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.
43  *
44  * This greatly improves filesystem parallelism and is a huge simplification
45  * of the API verses the old vnode locking / namei scheme.
46  *
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
51  * these issues.
52  */
53 
54 #include "opt_ktrace.h"
55 
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>
62 #include <sys/proc.h>
63 #include <sys/namei.h>
64 #include <sys/nlookup.h>
65 #include <sys/malloc.h>
66 #include <sys/stat.h>
67 #include <sys/objcache.h>
68 
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72 
73 /*
74  * Initialize a nlookup() structure, early error return for copyin faults
75  * or a degenerate empty string (which is not allowed).
76  *
77  * The first process proc0's credentials are used if the calling thread
78  * is not associated with a process context.
79  */
80 int
81 nlookup_init(struct nlookupdata *nd,
82 	     const char *path, enum uio_seg seg, int flags)
83 {
84     size_t pathlen;
85     struct proc *p;
86     thread_t td;
87     int error;
88 
89     td = curthread;
90     p = td->td_proc;
91 
92     /*
93      * note: the pathlen set by copy*str() includes the terminating \0.
94      */
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);
100     else
101 	error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
102 
103     /*
104      * Don't allow empty pathnames.
105      * POSIX.1 requirement: "" is not a vaild file name.
106      */
107     if (error == 0 && pathlen <= 1)
108 	error = ENOENT;
109 
110     if (error == 0) {
111 	if (p && p->p_fd) {
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);
117 	} else {
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);
122 	}
123 	nd->nl_td = td;
124 	nd->nl_flags |= flags;
125     } else {
126 	nlookup_done(nd);
127     }
128     return(error);
129 }
130 
131 /*
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.
135  */
136 int
137 nlookup_init_raw(struct nlookupdata *nd,
138 	     const char *path, enum uio_seg seg, int flags,
139 	     struct ucred *cred, struct nchandle *ncstart)
140 {
141     size_t pathlen;
142     thread_t td;
143     int error;
144 
145     td = curthread;
146 
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);
152     else
153 	error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
154 
155     /*
156      * Don't allow empty pathnames.
157      * POSIX.1 requirement: "" is not a vaild file name.
158      */
159     if (error == 0 && pathlen <= 1)
160 	error = ENOENT;
161 
162     if (error == 0) {
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);
167 	nd->nl_td = td;
168 	nd->nl_flags |= flags;
169     } else {
170 	nlookup_done(nd);
171     }
172     return(error);
173 }
174 
175 /*
176  * Set a different credential; this credential will be used by future
177  * operations performed on nd.nl_open_vp and nlookupdata structure.
178  */
179 void
180 nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
181 {
182 	KKASSERT(nd->nl_cred != NULL);
183 
184 	if (nd->nl_cred != cred) {
185 		cred = crhold(cred);
186 		crfree(nd->nl_cred);
187 		nd->nl_cred = cred;
188 	}
189 }
190 
191 /*
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.
197  */
198 void
199 nlookup_done(struct nlookupdata *nd)
200 {
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);
205 	}
206 	cache_drop(&nd->nl_nch);
207     }
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);
214 	nd->nl_path = NULL;
215     }
216     if (nd->nl_cred) {
217 	crfree(nd->nl_cred);
218 	nd->nl_cred = NULL;
219     }
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;
224 	}
225 	vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
226 	nd->nl_open_vp = NULL;
227     }
228     if (nd->nl_dvp) {
229 	vrele(nd->nl_dvp);
230 	nd->nl_dvp = NULL;
231     }
232     nd->nl_flags = 0;	/* clear remaining flags (just clear everything) */
233 }
234 
235 void
236 nlookup_zero(struct nlookupdata *nd)
237 {
238     bzero(nd, sizeof(struct nlookupdata));
239 }
240 
241 /*
242  * Simple all-in-one nlookup.  Returns a locked namecache structure or NULL
243  * if an error occured.
244  *
245  * Note that the returned ncp is not checked for permissions, though VEXEC
246  * is checked on the directory path leading up to the result.  The caller
247  * must call naccess() to check the permissions of the returned leaf.
248  */
249 struct nchandle
250 nlookup_simple(const char *str, enum uio_seg seg,
251 	       int niflags, int *error)
252 {
253     struct nlookupdata nd;
254     struct nchandle nch;
255 
256     *error = nlookup_init(&nd, str, seg, niflags);
257     if (*error == 0) {
258 	    if ((*error = nlookup(&nd)) == 0) {
259 		    nch = nd.nl_nch;	/* keep hold ref from structure */
260 		    cache_zero(&nd.nl_nch); /* and NULL out */
261 	    } else {
262 		    cache_zero(&nch);
263 	    }
264 	    nlookup_done(&nd);
265     } else {
266 	    cache_zero(&nch);
267     }
268     return(nch);
269 }
270 
271 /*
272  * Do a generic nlookup.  Note that the passed nd is not nlookup_done()'d
273  * on return, even if an error occurs.  If no error occurs the returned
274  * nl_nch is always referenced and locked, otherwise it may or may not be.
275  *
276  * Intermediate directory elements, including the current directory, require
277  * execute (search) permission.  nlookup does not examine the access
278  * permissions on the returned element.
279  *
280  * If NLC_CREATE or NLC_DELETE is set the last directory must allow node
281  * creation (VCREATE/VDELETE), and an error code of 0 will be returned for
282  * a non-existant target.  Otherwise a non-existant target will cause
283  * ENOENT to be returned.
284  *
285  * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
286  * of the returned entry.  The vnode will be referenced, but not locked,
287  * and will be released by nlookup_done() along with everything else.
288  */
289 int
290 nlookup(struct nlookupdata *nd)
291 {
292     struct nlcomponent nlc;
293     struct nchandle nch;
294     struct mount *mp;
295     int wasdotordotdot;
296     char *ptr;
297     char *xptr;
298     int error;
299     int len;
300 
301 #ifdef KTRACE
302     if (KTRPOINT(nd->nl_td, KTR_NAMEI))
303 	ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
304 #endif
305     bzero(&nlc, sizeof(nlc));
306 
307     /*
308      * Setup for the loop.  The current working namecache element must
309      * be in a refd + unlocked state.  This typically the case on entry except
310      * when stringing nlookup()'s along in a chain, since nlookup() always
311      * returns nl_nch in a locked state.
312      */
313     nd->nl_loopcnt = 0;
314     if (nd->nl_flags & NLC_NCPISLOCKED) {
315 	nd->nl_flags &= ~NLC_NCPISLOCKED;
316 	cache_unlock(&nd->nl_nch);
317     }
318     if (nd->nl_dvp ) {
319 	vrele(nd->nl_dvp);
320 	nd->nl_dvp = NULL;
321     }
322     ptr = nd->nl_path;
323 
324     /*
325      * Loop on the path components.  At the top of the loop nd->nl_nch
326      * is ref'd and unlocked and represents our current position.
327      */
328     for (;;) {
329 	/*
330 	 * Check if the root directory should replace the current
331 	 * directory.  This is done at the start of a translation
332 	 * or after a symbolic link has been found.  In other cases
333 	 * ptr will never be pointing at a '/'.
334 	 */
335 	if (*ptr == '/') {
336 	    do {
337 		++ptr;
338 	    } while (*ptr == '/');
339 	    cache_copy(&nd->nl_rootnch, &nch);
340 	    cache_drop(&nd->nl_nch);
341 	    nd->nl_nch = nch;
342 
343 	    /*
344 	     * Fast-track termination.  There is no parent directory of
345 	     * the root in the same mount from the point of view of
346 	     * the caller so return EPERM if NLC_REFDVP is specified.
347 	     * e.g. 'rmdir /' is not allowed.
348 	     */
349 	    if (*ptr == 0) {
350 		if (nd->nl_flags & NLC_REFDVP) {
351 			error = EPERM;
352 		} else {
353 			cache_lock(&nd->nl_nch);
354 			nd->nl_flags |= NLC_NCPISLOCKED;
355 			error = 0;
356 		}
357 		break;
358 	    }
359 	    continue;
360 	}
361 
362 	/*
363 	 * Check directory search permissions.
364 	 */
365 	if ((error = naccess(&nd->nl_nch, VEXEC, nd->nl_cred)) != 0)
366 	    break;
367 
368 	/*
369 	 * Extract the path component
370 	 */
371 	nlc.nlc_nameptr = ptr;
372 	while (*ptr && *ptr != '/')
373 	    ++ptr;
374 	nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
375 
376 	/*
377 	 * Lookup the path component in the cache, creating an unresolved
378 	 * entry if necessary.  We have to handle "." and ".." as special
379 	 * cases.
380 	 *
381 	 * When handling ".." we have to detect a traversal back through a
382 	 * mount point.   If we are at the root, ".." just returns the root.
383 	 *
384 	 * This subsection returns a locked, refd 'nch' unless it errors out.
385 	 * The namecache topology is not allowed to be disconnected, so
386 	 * encountering a NULL parent will generate EINVAL.  This typically
387 	 * occurs when a directory is removed out from under a process.
388 	 *
389 	 * If NLC_DELETE is set neither '.' or '..' can be the last component
390 	 * of a path.
391 	 */
392 	if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
393 	    cache_get(&nd->nl_nch, &nch);
394 	    wasdotordotdot = 1;
395 	} else if (nlc.nlc_namelen == 2 &&
396 		   nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
397 	    if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
398 		nd->nl_nch.ncp == nd->nl_rootnch.ncp
399 	    ) {
400 		/*
401 		 * ".." at the root returns the root
402 		 */
403 		cache_get(&nd->nl_nch, &nch);
404 	    } else {
405 		/*
406 		 * Locate the parent ncp.  If we are at the root of a
407 		 * filesystem mount we have to skip to the mounted-on
408 		 * point in the underlying filesystem.
409 		 */
410 		nch = nd->nl_nch;
411 		while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
412 			nch = nch.mount->mnt_ncmounton;
413 		nch.ncp = nch.ncp->nc_parent;
414 		KKASSERT(nch.ncp != NULL);
415 		cache_get(&nch, &nch);
416 	    }
417 	    wasdotordotdot = 1;
418 	} else {
419 	    nch = cache_nlookup(&nd->nl_nch, &nlc);
420 	    while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
421 		kprintf("[diagnostic] nlookup: relookup %*.*s\n",
422 			nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
423 		cache_put(&nch);
424 		nch = cache_nlookup(&nd->nl_nch, &nlc);
425 	    }
426 	    wasdotordotdot = 0;
427 	}
428 	/*
429 	 * [end of subsection] ncp is locked and ref'd.  nd->nl_nch is ref'd
430 	 */
431 
432 	/*
433 	 * Resolve the namespace if necessary.  The ncp returned by
434 	 * cache_nlookup() is referenced and locked.
435 	 *
436 	 * XXX neither '.' nor '..' should return EAGAIN since they were
437 	 * previously resolved and thus cannot be newly created ncp's.
438 	 */
439 	if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
440 	    error = cache_resolve(&nch, nd->nl_cred);
441 	    KKASSERT(error != EAGAIN);
442 	} else {
443 	    error = nch.ncp->nc_error;
444 	}
445 
446 	/*
447 	 * Early completion.  ENOENT is not an error if this is the last
448 	 * component and NLC_CREATE was requested.  Note that ncp->nc_error
449 	 * is left as ENOENT in that case, which we check later on.
450 	 *
451 	 * Also handle invalid '.' or '..' components terminating a path
452 	 * during removal.  The standard requires this and pax pretty
453 	 * stupidly depends on it.
454 	 */
455 	for (xptr = ptr; *xptr == '/'; ++xptr)
456 		;
457 	if (*xptr == 0) {
458 	    if (error == ENOENT && (nd->nl_flags & NLC_CREATE))
459 		error = naccess(&nch, VCREATE, nd->nl_cred);
460 	    if (error == 0 && wasdotordotdot && (nd->nl_flags & NLC_DELETE))
461 		error = EINVAL;
462 	}
463 
464 	/*
465 	 * Early completion on error.
466 	 */
467 	if (error) {
468 	    cache_put(&nch);
469 	    break;
470 	}
471 
472 	/*
473 	 * If the element is a symlink and it is either not the last
474 	 * element or it is the last element and we are allowed to
475 	 * follow symlinks, resolve the symlink.
476 	 */
477 	if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
478 	    (*ptr || (nd->nl_flags & NLC_FOLLOW))
479 	) {
480 	    if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
481 		error = ELOOP;
482 		cache_put(&nch);
483 		break;
484 	    }
485 	    error = nreadsymlink(nd, &nch, &nlc);
486 	    cache_put(&nch);
487 	    if (error)
488 		break;
489 
490 	    /*
491 	     * Concatenate trailing path elements onto the returned symlink.
492 	     * Note that if the path component (ptr) is not exhausted, it
493 	     * will being with a '/', so we do not have to add another one.
494 	     *
495 	     * The symlink may not be empty.
496 	     */
497 	    len = strlen(ptr);
498 	    if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
499 		error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
500 		objcache_put(namei_oc, nlc.nlc_nameptr);
501 		break;
502 	    }
503 	    bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
504 	    if (nd->nl_flags & NLC_HASBUF)
505 		objcache_put(namei_oc, nd->nl_path);
506 	    nd->nl_path = nlc.nlc_nameptr;
507 	    nd->nl_flags |= NLC_HASBUF;
508 	    ptr = nd->nl_path;
509 
510 	    /*
511 	     * Go back up to the top to resolve any initial '/'s in the
512 	     * symlink.
513 	     */
514 	    continue;
515 	}
516 
517 	/*
518 	 * If the element is a directory and we are crossing a mount point,
519 	 * Locate the mount.
520 	 */
521 	while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
522 	    (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
523 	    (mp = cache_findmount(&nch)) != NULL
524 	) {
525 	    struct vnode *tdp;
526 
527 	    cache_put(&nch);
528 	    cache_get(&mp->mnt_ncmountpt, &nch);
529 
530 	    if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
531 		while (vfs_busy(mp, 0))
532 		    ;
533 		error = VFS_ROOT(mp, &tdp);
534 		vfs_unbusy(mp);
535 		if (error)
536 		    break;
537 		cache_setvp(&nch, tdp);
538 		vput(tdp);
539 	    }
540 	}
541 	if (error) {
542 	    cache_put(&nch);
543 	    break;
544 	}
545 
546 	/*
547 	 * Skip any slashes to get to the next element.  If there
548 	 * are any slashes at all the current element must be a
549 	 * directory or, in the create case, intended to become a directory.
550 	 * If it isn't we break without incrementing ptr and fall through
551 	 * to the failure case below.
552 	 */
553 	while (*ptr == '/') {
554 	    if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
555 		!(nd->nl_flags & NLC_WILLBEDIR)
556 	    ) {
557 		break;
558 	    }
559 	    ++ptr;
560 	}
561 
562 	/*
563 	 * Continuation case: additional elements and the current
564 	 * element is a directory.
565 	 */
566 	if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
567 	    cache_drop(&nd->nl_nch);
568 	    cache_unlock(&nch);
569 	    nd->nl_nch = nch;
570 	    continue;
571 	}
572 
573 	/*
574 	 * Failure case: additional elements and the current element
575 	 * is not a directory
576 	 */
577 	if (*ptr) {
578 	    cache_put(&nch);
579 	    error = ENOTDIR;
580 	    break;
581 	}
582 
583 	/*
584 	 * Successful lookup of last element.
585 	 *
586 	 * Check directory permissions if a deletion is specified.
587 	 */
588 	if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
589 	    if ((error = naccess(&nch, VDELETE, nd->nl_cred)) != 0) {
590 		cache_put(&nch);
591 		break;
592 	    }
593 	}
594 
595 	/*
596 	 * Termination: no more elements.  If NLC_CREATE was set the
597 	 * ncp may represent a negative hit (ncp->nc_error will be ENOENT),
598 	 * but we still return an error code of 0.
599 	 *
600 	 * If NLC_REFDVP is set acquire a referenced parent dvp.
601 	 */
602 	if (nd->nl_flags & NLC_REFDVP) {
603 		error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
604 		if (error) {
605 			kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
606 			cache_put(&nch);
607 			break;
608 		}
609 	}
610 	cache_drop(&nd->nl_nch);
611 	nd->nl_nch = nch;
612 	nd->nl_flags |= NLC_NCPISLOCKED;
613 	error = 0;
614 	break;
615     }
616     return(error);
617 }
618 
619 /*
620  * Resolve a mount point's glue ncp.  This ncp connects creates the illusion
621  * of continuity in the namecache tree by connecting the ncp related to the
622  * vnode under the mount to the ncp related to the mount's root vnode.
623  *
624  * If no error occured a locked, ref'd ncp is stored in *ncpp.
625  */
626 int
627 nlookup_mp(struct mount *mp, struct nchandle *nch)
628 {
629     struct vnode *vp;
630     int error;
631 
632     error = 0;
633     cache_get(&mp->mnt_ncmountpt, nch);
634     if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
635 	while (vfs_busy(mp, 0))
636 	    ;
637 	error = VFS_ROOT(mp, &vp);
638 	vfs_unbusy(mp);
639 	if (error) {
640 	    cache_put(nch);
641 	} else {
642 	    cache_setvp(nch, vp);
643 	    vput(vp);
644 	}
645     }
646     return(error);
647 }
648 
649 /*
650  * Read the contents of a symlink, allocate a path buffer out of the
651  * namei_oc and initialize the supplied nlcomponent with the result.
652  *
653  * If an error occurs no buffer will be allocated or returned in the nlc.
654  */
655 int
656 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
657 		struct nlcomponent *nlc)
658 {
659     struct vnode *vp;
660     struct iovec aiov;
661     struct uio auio;
662     int linklen;
663     int error;
664     char *cp;
665 
666     nlc->nlc_nameptr = NULL;
667     nlc->nlc_namelen = 0;
668     if (nch->ncp->nc_vp == NULL)
669 	return(ENOENT);
670     if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
671 	return(error);
672     cp = objcache_get(namei_oc, M_WAITOK);
673     aiov.iov_base = cp;
674     aiov.iov_len = MAXPATHLEN;
675     auio.uio_iov = &aiov;
676     auio.uio_iovcnt = 1;
677     auio.uio_offset = 0;
678     auio.uio_rw = UIO_READ;
679     auio.uio_segflg = UIO_SYSSPACE;
680     auio.uio_td = nd->nl_td;
681     auio.uio_resid = MAXPATHLEN - 1;
682     error = VOP_READLINK(vp, &auio, nd->nl_cred);
683     if (error)
684 	goto fail;
685     linklen = MAXPATHLEN - 1 - auio.uio_resid;
686     if (varsym_enable) {
687 	linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
688 	if (linklen < 0) {
689 	    error = ENAMETOOLONG;
690 	    goto fail;
691 	}
692     }
693     cp[linklen] = 0;
694     nlc->nlc_nameptr = cp;
695     nlc->nlc_namelen = linklen;
696     vput(vp);
697     return(0);
698 fail:
699     objcache_put(namei_oc, cp);
700     vput(vp);
701     return(error);
702 }
703 
704 /*
705  * Check access [XXX cache vattr!] [XXX quota]
706  *
707  * Generally check the V* access bits from sys/vnode.h.  All specified bits
708  * must pass for this function to return 0.
709  *
710  * If VCREATE is specified and the target ncp represents a non-existant
711  * file or dir, or if VDELETE is specified and the target exists, the parent
712  * directory is checked for VWRITE.  If VEXCL is specified and the target
713  * ncp represents a positive hit, an error is returned.
714  *
715  * If VCREATE is not specified and the target does not exist (negative hit),
716  * ENOENT is returned.  Note that nlookup() does not (and should not) return
717  * ENOENT for non-existant leafs.
718  *
719  * The passed ncp may or may not be locked.  The caller should use a
720  * locked ncp on leaf lookups, especially for VCREATE, VDELETE, and VEXCL
721  * checks.
722  */
723 int
724 naccess(struct nchandle *nch, int vmode, struct ucred *cred)
725 {
726     struct nchandle par;
727     struct vnode *vp;
728     struct vattr va;
729     int error;
730 
731     if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
732 	cache_lock(nch);
733 	cache_resolve(nch, cred);
734 	cache_unlock(nch);
735     }
736     error = nch->ncp->nc_error;
737     if (vmode & (VDELETE|VCREATE|VEXCL)) {
738 	if (((vmode & VCREATE) && nch->ncp->nc_vp == NULL) ||
739 	    ((vmode & VDELETE) && nch->ncp->nc_vp != NULL)
740 	) {
741 	    if ((par.ncp = nch->ncp->nc_parent) == NULL) {
742 		if (error != EAGAIN)
743 			error = EINVAL;
744 	    } else {
745 		par.mount = nch->mount;
746 		cache_hold(&par);
747 		error = naccess(&par, VWRITE, cred);
748 		cache_drop(&par);
749 	    }
750 	}
751 	if ((vmode & VEXCL) && nch->ncp->nc_vp != NULL)
752 	    error = EEXIST;
753     }
754     if (error == 0) {
755 	error = cache_vget(nch, cred, LK_SHARED, &vp);
756 	if (error == ENOENT) {
757 	    if (vmode & VCREATE)
758 		error = 0;
759 	} else if (error == 0) {
760 	    /* XXX cache the va in the namecache or in the vnode */
761 	    if ((error = VOP_GETATTR(vp, &va)) == 0) {
762 		if ((vmode & VWRITE) && vp->v_mount) {
763 		    if (vp->v_mount->mnt_flag & MNT_RDONLY)
764 			error = EROFS;
765 		}
766 	    }
767 	    vput(vp);
768 	    if (error == 0)
769 		error = naccess_va(&va, vmode, cred);
770 	}
771     }
772     return(error);
773 }
774 
775 /*
776  * Check the requested access against the given vattr using cred.
777  */
778 int
779 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
780 {
781     int i;
782 
783     /*
784      * Test the immutable bit for files, directories, and softlinks.
785      */
786     if (vmode & (VWRITE|VDELETE)) {
787 	if (va->va_type == VDIR || va->va_type == VLNK || va->va_type == VREG) {
788 	    if (va->va_flags & IMMUTABLE)
789 		return (EPERM);
790 	}
791     }
792 
793     /*
794      * root gets universal access
795      */
796     if (cred->cr_uid == 0)
797 	return(0);
798 
799     /*
800      * Check owner perms, group perms, and world perms
801      */
802     vmode &= S_IRWXU;
803     if (cred->cr_uid == va->va_uid) {
804 	if ((vmode & va->va_mode) != vmode)
805 	    return(EACCES);
806 	return(0);
807     }
808 
809     vmode >>= 3;
810     for (i = 0; i < cred->cr_ngroups; ++i) {
811 	if (va->va_gid == cred->cr_groups[i]) {
812 	    if ((vmode & va->va_mode) != vmode)
813 		return(EACCES);
814 	    return(0);
815 	}
816     }
817 
818     vmode >>= 3;
819     if ((vmode & va->va_mode) != vmode)
820 	return(EACCES);
821     return(0);
822 }
823 
824