xref: /dragonfly/sys/kern/vfs_nlookup.c (revision c03f08f3)
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 1.22 2007/05/09 05:12:45 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     nd->nl_flags = 0;	/* clear remaining flags (just clear everything) */
229 }
230 
231 void
232 nlookup_zero(struct nlookupdata *nd)
233 {
234     bzero(nd, sizeof(struct nlookupdata));
235 }
236 
237 /*
238  * Simple all-in-one nlookup.  Returns a locked namecache structure or NULL
239  * if an error occured.
240  *
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.
244  */
245 struct nchandle
246 nlookup_simple(const char *str, enum uio_seg seg,
247 	       int niflags, int *error)
248 {
249     struct nlookupdata nd;
250     struct nchandle nch;
251 
252     *error = nlookup_init(&nd, str, seg, niflags);
253     if (*error == 0) {
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 */
257 	    } else {
258 		    cache_zero(&nch);
259 	    }
260 	    nlookup_done(&nd);
261     } else {
262 	    cache_zero(&nch);
263     }
264     return(nch);
265 }
266 
267 /*
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.
271  *
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.
275  *
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.
280  */
281 int
282 nlookup(struct nlookupdata *nd)
283 {
284     struct nlcomponent nlc;
285     struct nchandle nch;
286     struct mount *mp;
287     int wasdotordotdot;
288     char *ptr;
289     char *xptr;
290     int error;
291     int len;
292 
293 #ifdef KTRACE
294     if (KTRPOINT(nd->nl_td, KTR_NAMEI))
295 	ktrnamei(nd->nl_td->td_proc, nd->nl_path);
296 #endif
297     bzero(&nlc, sizeof(nlc));
298 
299     /*
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.
304      */
305     nd->nl_loopcnt = 0;
306     if (nd->nl_flags & NLC_NCPISLOCKED) {
307 	nd->nl_flags &= ~NLC_NCPISLOCKED;
308 	cache_unlock(&nd->nl_nch);
309     }
310     ptr = nd->nl_path;
311 
312     /*
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.
315      */
316     for (;;) {
317 	/*
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 '/'.
322 	 */
323 	if (*ptr == '/') {
324 	    do {
325 		++ptr;
326 	    } while (*ptr == '/');
327 	    cache_copy(&nd->nl_rootnch, &nch);
328 	    cache_drop(&nd->nl_nch);
329 	    nd->nl_nch = nch;
330 	    if (*ptr == 0) {
331 		cache_lock(&nd->nl_nch);
332 		nd->nl_flags |= NLC_NCPISLOCKED;
333 		error = 0;
334 		break;
335 	    }
336 	    continue;
337 	}
338 
339 	/*
340 	 * Check directory search permissions.
341 	 */
342 	if ((error = naccess(&nd->nl_nch, VEXEC, nd->nl_cred)) != 0)
343 	    break;
344 
345 	/*
346 	 * Extract the path component
347 	 */
348 	nlc.nlc_nameptr = ptr;
349 	while (*ptr && *ptr != '/')
350 	    ++ptr;
351 	nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
352 
353 	/*
354 	 * Lookup the path component in the cache, creating an unresolved
355 	 * entry if necessary.  We have to handle "." and ".." as special
356 	 * cases.
357 	 *
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.
360 	 *
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.
365 	 *
366 	 * If NLC_DELETE is set neither '.' or '..' can be the last component
367 	 * of a path.
368 	 */
369 	if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
370 	    cache_get(&nd->nl_nch, &nch);
371 	    wasdotordotdot = 1;
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
376 	    ) {
377 		/*
378 		 * ".." at the root returns the root
379 		 */
380 		cache_get(&nd->nl_nch, &nch);
381 	    } else {
382 		/*
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.
386 		 */
387 		nch = nd->nl_nch;
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);
393 	    }
394 	    wasdotordotdot = 1;
395 	} else {
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);
400 		cache_put(&nch);
401 		nch = cache_nlookup(&nd->nl_nch, &nlc);
402 	    }
403 	    wasdotordotdot = 0;
404 	}
405 	/*
406 	 * [end of subsection] ncp is locked and ref'd.  nd->nl_nch is ref'd
407 	 */
408 
409 	/*
410 	 * Resolve the namespace if necessary.  The ncp returned by
411 	 * cache_nlookup() is referenced and locked.
412 	 *
413 	 * XXX neither '.' nor '..' should return EAGAIN since they were
414 	 * previously resolved and thus cannot be newly created ncp's.
415 	 */
416 	if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
417 	    error = cache_resolve(&nch, nd->nl_cred);
418 	    KKASSERT(error != EAGAIN);
419 	} else {
420 	    error = nch.ncp->nc_error;
421 	}
422 
423 	/*
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.
427 	 *
428 	 * Also handle invalid '.' or '..' components terminating a path
429 	 * during removal.  The standard requires this and pax pretty
430 	  *stupidly depends on it.
431 	 */
432 	for (xptr = ptr; *xptr == '/'; ++xptr)
433 		;
434 	if (*xptr == 0) {
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))
438 		error = EINVAL;
439 	}
440 
441 	/*
442 	 * Early completion on error.
443 	 */
444 	if (error) {
445 	    cache_put(&nch);
446 	    break;
447 	}
448 
449 	/*
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.
453 	 */
454 	if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
455 	    (*ptr || (nd->nl_flags & NLC_FOLLOW))
456 	) {
457 	    if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
458 		error = ELOOP;
459 		cache_put(&nch);
460 		break;
461 	    }
462 	    error = nreadsymlink(nd, &nch, &nlc);
463 	    cache_put(&nch);
464 	    if (error)
465 		break;
466 
467 	    /*
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.
471 	     *
472 	     * The symlink may not be empty.
473 	     */
474 	    len = strlen(ptr);
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);
478 		break;
479 	    }
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;
485 	    ptr = nd->nl_path;
486 
487 	    /*
488 	     * Go back up to the top to resolve any initial '/'s in the
489 	     * symlink.
490 	     */
491 	    continue;
492 	}
493 
494 	/*
495 	 * If the element is a directory and we are crossing a mount point,
496 	 * Locate the mount.
497 	 */
498 	while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
499 	    (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
500 	    (mp = cache_findmount(&nch)) != NULL
501 	) {
502 	    struct vnode *tdp;
503 
504 	    cache_put(&nch);
505 	    cache_get(&mp->mnt_ncmountpt, &nch);
506 
507 	    if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
508 		while (vfs_busy(mp, 0))
509 		    ;
510 		error = VFS_ROOT(mp, &tdp);
511 		vfs_unbusy(mp);
512 		if (error)
513 		    break;
514 		cache_setvp(&nch, tdp);
515 		vput(tdp);
516 	    }
517 	}
518 	if (error) {
519 	    cache_put(&nch);
520 	    break;
521 	}
522 
523 	/*
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.
529 	 */
530 	while (*ptr == '/') {
531 	    if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
532 		!(nd->nl_flags & NLC_WILLBEDIR)
533 	    ) {
534 		break;
535 	    }
536 	    ++ptr;
537 	}
538 
539 	/*
540 	 * Continuation case: additional elements and the current
541 	 * element is a directory.
542 	 */
543 	if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
544 	    cache_drop(&nd->nl_nch);
545 	    cache_unlock(&nch);
546 	    nd->nl_nch = nch;
547 	    continue;
548 	}
549 
550 	/*
551 	 * Failure case: additional elements and the current element
552 	 * is not a directory
553 	 */
554 	if (*ptr) {
555 	    cache_put(&nch);
556 	    error = ENOTDIR;
557 	    break;
558 	}
559 
560 	/*
561 	 * Successful lookup of last element.
562 	 *
563 	 * Check directory permissions if a deletion is specified.
564 	 */
565 	if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
566 	    if ((error = naccess(&nch, VDELETE, nd->nl_cred)) != 0) {
567 		cache_put(&nch);
568 		break;
569 	    }
570 	}
571 
572 	/*
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.
576 	 */
577 	cache_drop(&nd->nl_nch);
578 	nd->nl_nch = nch;
579 	nd->nl_flags |= NLC_NCPISLOCKED;
580 	error = 0;
581 	break;
582     }
583     return(error);
584 }
585 
586 /*
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.
590  *
591  * If no error occured a locked, ref'd ncp is stored in *ncpp.
592  */
593 int
594 nlookup_mp(struct mount *mp, struct nchandle *nch)
595 {
596     struct vnode *vp;
597     int error;
598 
599     error = 0;
600     cache_get(&mp->mnt_ncmountpt, nch);
601     if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
602 	while (vfs_busy(mp, 0))
603 	    ;
604 	error = VFS_ROOT(mp, &vp);
605 	vfs_unbusy(mp);
606 	if (error) {
607 	    cache_put(nch);
608 	} else {
609 	    cache_setvp(nch, vp);
610 	    vput(vp);
611 	}
612     }
613     return(error);
614 }
615 
616 /*
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.
619  *
620  * If an error occurs no buffer will be allocated or returned in the nlc.
621  */
622 int
623 nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
624 		struct nlcomponent *nlc)
625 {
626     struct vnode *vp;
627     struct iovec aiov;
628     struct uio auio;
629     int linklen;
630     int error;
631     char *cp;
632 
633     nlc->nlc_nameptr = NULL;
634     nlc->nlc_namelen = 0;
635     if (nch->ncp->nc_vp == NULL)
636 	return(ENOENT);
637     if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
638 	return(error);
639     cp = objcache_get(namei_oc, M_WAITOK);
640     aiov.iov_base = cp;
641     aiov.iov_len = MAXPATHLEN;
642     auio.uio_iov = &aiov;
643     auio.uio_iovcnt = 1;
644     auio.uio_offset = 0;
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);
650     if (error)
651 	goto fail;
652     linklen = MAXPATHLEN - 1 - auio.uio_resid;
653     if (varsym_enable) {
654 	linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
655 	if (linklen < 0) {
656 	    error = ENAMETOOLONG;
657 	    goto fail;
658 	}
659     }
660     cp[linklen] = 0;
661     nlc->nlc_nameptr = cp;
662     nlc->nlc_namelen = linklen;
663     vput(vp);
664     return(0);
665 fail:
666     objcache_put(namei_oc, cp);
667     vput(vp);
668     return(error);
669 }
670 
671 /*
672  * Check access [XXX cache vattr!] [XXX quota]
673  *
674  * Generally check the V* access bits from sys/vnode.h.  All specified bits
675  * must pass for this function to return 0.
676  *
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.
681  *
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.
685  *
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
688  * checks.
689  */
690 int
691 naccess(struct nchandle *nch, int vmode, struct ucred *cred)
692 {
693     struct nchandle par;
694     struct vnode *vp;
695     struct vattr va;
696     int error;
697 
698     if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
699 	cache_lock(nch);
700 	cache_resolve(nch, cred);
701 	cache_unlock(nch);
702     }
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)
707 	) {
708 	    if ((par.ncp = nch->ncp->nc_parent) == NULL) {
709 		if (error != EAGAIN)
710 			error = EINVAL;
711 	    } else {
712 		par.mount = nch->mount;
713 		cache_hold(&par);
714 		error = naccess(&par, VWRITE, cred);
715 		cache_drop(&par);
716 	    }
717 	}
718 	if ((vmode & VEXCL) && nch->ncp->nc_vp != NULL)
719 	    error = EEXIST;
720     }
721     if (error == 0) {
722 	error = cache_vget(nch, cred, LK_SHARED, &vp);
723 	if (error == ENOENT) {
724 	    if (vmode & VCREATE)
725 		error = 0;
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)
731 			error = EROFS;
732 		}
733 	    }
734 	    vput(vp);
735 	    if (error == 0)
736 		error = naccess_va(&va, vmode, cred);
737 	}
738     }
739     return(error);
740 }
741 
742 /*
743  * Check the requested access against the given vattr using cred.
744  */
745 int
746 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
747 {
748     int i;
749 
750     /*
751      * Test the immutable bit for files, directories, and softlinks.
752      */
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)
756 		return (EPERM);
757 	}
758     }
759 
760     /*
761      * root gets universal access
762      */
763     if (cred->cr_uid == 0)
764 	return(0);
765 
766     /*
767      * Check owner perms, group perms, and world perms
768      */
769     vmode &= S_IRWXU;
770     if (cred->cr_uid == va->va_uid) {
771 	if ((vmode & va->va_mode) != vmode)
772 	    return(EACCES);
773 	return(0);
774     }
775 
776     vmode >>= 3;
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)
780 		return(EACCES);
781 	    return(0);
782 	}
783     }
784 
785     vmode >>= 3;
786     if ((vmode & va->va_mode) != vmode)
787 	return(EACCES);
788     return(0);
789 }
790 
791