xref: /dragonfly/sys/kern/vfs_nlookup.c (revision e1acdbad)
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 1.11 2005/04/19 17:54:42 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 <vm/vm_zone.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 = zalloc(namei_zone);
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 	    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);
117 	} else {
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);
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.  rootncp 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 namecache *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 = zalloc(namei_zone);
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 	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);
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_ncp) {
202 	if (nd->nl_flags & NLC_NCPISLOCKED) {
203 	    nd->nl_flags &= ~NLC_NCPISLOCKED;
204 	    cache_unlock(nd->nl_ncp);
205 	}
206 	cache_drop(nd->nl_ncp);
207 	nd->nl_ncp = NULL;
208     }
209     if (nd->nl_rootncp) {
210 	cache_drop(nd->nl_rootncp);
211 	nd->nl_rootncp = NULL;
212     }
213     if (nd->nl_jailncp) {
214 	cache_drop(nd->nl_jailncp);
215 	nd->nl_jailncp = NULL;
216     }
217     if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
218 	zfree(namei_zone, nd->nl_path);
219 	nd->nl_path = NULL;
220     }
221     if (nd->nl_cred) {
222 	crfree(nd->nl_cred);
223 	nd->nl_cred = NULL;
224     }
225     if (nd->nl_open_vp) {
226 	if (nd->nl_flags & NLC_LOCKVP) {
227 		VOP_UNLOCK(nd->nl_open_vp, 0, nd->nl_td);
228 		nd->nl_flags &= ~NLC_LOCKVP;
229 	}
230 	vn_close(nd->nl_open_vp, nd->nl_vp_fmode, nd->nl_td);
231 	nd->nl_open_vp = NULL;
232     }
233     nd->nl_flags = 0;	/* clear remaining flags (just clear everything) */
234 }
235 
236 void
237 nlookup_zero(struct nlookupdata *nd)
238 {
239     bzero(nd, sizeof(struct nlookupdata));
240 }
241 
242 /*
243  * Simple all-in-one nlookup.  Returns a locked namecache structure or NULL
244  * if an error occured.
245  *
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.
249  */
250 struct namecache *
251 nlookup_simple(const char *str, enum uio_seg seg,
252 	       int niflags, int *error)
253 {
254     struct nlookupdata nd;
255     struct namecache *ncp;
256 
257     *error = nlookup_init(&nd, str, seg, niflags);
258     if (*error == 0) {
259 	    if ((*error = nlookup(&nd)) == 0) {
260 		    ncp = nd.nl_ncp;	/* keep hold ref from structure */
261 		    nd.nl_ncp = NULL;	/* and NULL out */
262 	    } else {
263 		    ncp = NULL;
264 	    }
265 	    nlookup_done(&nd);
266     } else {
267 	    ncp = NULL;
268     }
269     return(ncp);
270 }
271 
272 /*
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.
276  *
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.
280  *
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.
285  */
286 int
287 nlookup(struct nlookupdata *nd)
288 {
289     struct nlcomponent nlc;
290     struct namecache *ncp;
291     int wasdotordotdot;
292     char *ptr;
293     char *xptr;
294     int error;
295     int len;
296 
297 #ifdef KTRACE
298     if (KTRPOINT(nd->nl_td, KTR_NAMEI))
299 	ktrnamei(nd->nl_td->td_proc->p_tracep, nd->nl_path);
300 #endif
301     bzero(&nlc, sizeof(nlc));
302 
303     /*
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.
308      */
309     nd->nl_loopcnt = 0;
310     if (nd->nl_flags & NLC_NCPISLOCKED) {
311 	nd->nl_flags &= ~NLC_NCPISLOCKED;
312 	cache_unlock(nd->nl_ncp);
313     }
314     ptr = nd->nl_path;
315 
316     /*
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.
319      */
320     for (;;) {
321 	/*
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 '/'.
326 	 */
327 	if (*ptr == '/') {
328 	    do {
329 		++ptr;
330 	    } while (*ptr == '/');
331 	    ncp = cache_hold(nd->nl_rootncp);
332 	    cache_drop(nd->nl_ncp);
333 	    nd->nl_ncp = ncp;
334 	    if (*ptr == 0) {
335 		cache_lock(nd->nl_ncp);
336 		nd->nl_flags |= NLC_NCPISLOCKED;
337 		error = 0;
338 		break;
339 	    }
340 	    continue;
341 	}
342 
343 	/*
344 	 * Check directory search permissions
345 	 */
346 	if ((error = naccess(nd->nl_ncp, VEXEC, nd->nl_cred)) != 0)
347 	    break;
348 
349 	/*
350 	 * Extract the path component
351 	 */
352 	nlc.nlc_nameptr = ptr;
353 	while (*ptr && *ptr != '/')
354 	    ++ptr;
355 	nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
356 
357 	/*
358 	 * Lookup the path component in the cache, creating an unresolved
359 	 * entry if necessary.  We have to handle "." and ".." as special
360 	 * cases.
361 	 *
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.
365 	 *
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.
370 	 *
371 	 * If NLC_DELETE is set neither '.' or '..' can be the last component
372 	 * of a path.
373 	 */
374 	if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
375 	    ncp = cache_get(nd->nl_ncp);
376 	    wasdotordotdot = 1;
377 	} else if (nlc.nlc_namelen == 2 &&
378 		   nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
379 	    ncp = nd->nl_ncp;
380 	    if (ncp == nd->nl_rootncp) {
381 		ncp = cache_get(ncp);
382 	    } else {
383 		while ((ncp->nc_flag & NCF_MOUNTPT) && ncp != nd->nl_rootncp) {
384 		    if (ncp->nc_parent->nc_flag & NCF_DESTROYED)
385 			break;
386 		    ncp = ncp->nc_parent;	/* get to underlying node */
387 		    KKASSERT(ncp != NULL && 1);
388 		}
389 		if (ncp != nd->nl_rootncp) {
390 			if (ncp->nc_parent->nc_flag & NCF_DESTROYED) {
391 				error = EINVAL;
392 				break;
393 			}
394 			ncp = ncp->nc_parent;
395 			if (ncp == NULL) {
396 				error = EINVAL;
397 				break;
398 			}
399 		}
400 		ncp = cache_get(ncp);
401 	    }
402 	    wasdotordotdot = 1;
403 	} else {
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);
408 		cache_put(ncp);
409 		ncp = cache_nlookup(nd->nl_ncp, &nlc);
410 	    }
411 	    wasdotordotdot = 0;
412 	}
413 	/*
414 	 * [end of subsection] ncp is locked and ref'd.  nd->nl_ncp is ref'd
415 	 */
416 
417 	/*
418 	 * Resolve the namespace if necessary.  The ncp returned by
419 	 * cache_nlookup() is referenced and locked.
420 	 *
421 	 * XXX neither '.' nor '..' should return EAGAIN since they were
422 	 * previously resolved and thus cannot be newly created ncp's.
423 	 */
424 	if (ncp->nc_flag & NCF_UNRESOLVED) {
425 	    error = cache_resolve(ncp, nd->nl_cred);
426 	    KKASSERT(error != EAGAIN);
427 	} else {
428 	    error = ncp->nc_error;
429 	}
430 
431 	/*
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.
435 	 *
436 	 * Also handle invalid '.' or '..' components terminating a path
437 	 * during removal.  The standard requires this and pax pretty
438 	  *stupidly depends on it.
439 	 */
440 	for (xptr = ptr; *xptr == '/'; ++xptr)
441 		;
442 	if (*xptr == 0) {
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))
446 		error = EINVAL;
447 	}
448 
449 	/*
450 	 * Early completion on error.
451 	 */
452 	if (error) {
453 	    cache_put(ncp);
454 	    break;
455 	}
456 
457 	/*
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.
461 	 */
462 	if ((ncp->nc_flag & NCF_ISSYMLINK) &&
463 	    (*ptr || (nd->nl_flags & NLC_FOLLOW))
464 	) {
465 	    if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
466 		error = ELOOP;
467 		cache_put(ncp);
468 		break;
469 	    }
470 	    error = nreadsymlink(nd, ncp, &nlc);
471 	    cache_put(ncp);
472 	    if (error)
473 		break;
474 
475 	    /*
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.
479 	     *
480 	     * The symlink may not be empty.
481 	     */
482 	    len = strlen(ptr);
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);
486 		break;
487 	    }
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;
493 	    ptr = nd->nl_path;
494 
495 	    /*
496 	     * Go back up to the top to resolve any initial '/'s in the
497 	     * symlink.
498 	     */
499 	    continue;
500 	}
501 
502 	/*
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.
506 	 *
507 	 * XXX mnt_ncp should really be resolved in the mount code.
508 	 * NOTE!  the normal nresolve() code cannot resolve mount point ncp's!
509 	 *
510 	 * XXX NOCROSSMOUNT
511 	 */
512 	while ((ncp->nc_flag & NCF_ISDIR) && ncp->nc_vp->v_mountedhere &&
513 		(nd->nl_flags & NLC_NOCROSSMOUNT) == 0
514 	) {
515 	    struct mount *mp;
516 	    struct vnode *tdp;
517 
518 	    mp = ncp->nc_vp->v_mountedhere;
519 	    cache_put(ncp);
520 	    ncp = cache_get(mp->mnt_ncp);
521 
522 	    if (ncp->nc_flag & NCF_UNRESOLVED) {
523 		while (vfs_busy(mp, 0, nd->nl_td))
524 		    ;
525 		error = VFS_ROOT(mp, &tdp);
526 		vfs_unbusy(mp, nd->nl_td);
527 		if (error)
528 		    break;
529 		cache_setvp(ncp, tdp);
530 		vput(tdp);
531 	    }
532 	}
533 	if (error) {
534 	    cache_put(ncp);
535 	    break;
536 	}
537 
538 	/*
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.
544 	 */
545 	while (*ptr == '/') {
546 	    if ((ncp->nc_flag & NCF_ISDIR) == 0 &&
547 		!(nd->nl_flags & NLC_WILLBEDIR)
548 	    ) {
549 		break;
550 	    }
551 	    ++ptr;
552 	}
553 
554 	/*
555 	 * Continuation case: additional elements and the current
556 	 * element is a directory.
557 	 */
558 	if (*ptr && (ncp->nc_flag & NCF_ISDIR)) {
559 	    cache_drop(nd->nl_ncp);
560 	    cache_unlock(ncp);
561 	    nd->nl_ncp = ncp;
562 	    continue;
563 	}
564 
565 	/*
566 	 * Failure case: additional elements and the current element
567 	 * is not a directory
568 	 */
569 	if (*ptr) {
570 	    cache_put(ncp);
571 	    error = ENOTDIR;
572 	    break;
573 	}
574 
575 	/*
576 	 * Successful lookup of last element.
577 	 *
578 	 * Check directory permissions if a deletion is specified.
579 	 */
580 	if (*ptr == 0 && (nd->nl_flags & NLC_DELETE)) {
581 	    if ((error = naccess(ncp, VDELETE, nd->nl_cred)) != 0) {
582 		cache_put(ncp);
583 		break;
584 	    }
585 	}
586 
587 	/*
588 	 * XXX vnode canvmio (test in mmap(), read(), and write())
589 	 */
590 
591 	/*
592 	 * Termination: no more elements.  If NLC_CREATE was set the
593 	 * ncp may represent a negative hit (ncp->nc_error will be ENOENT),
594 	 * but we still return an error code of 0.
595 	 */
596 	cache_drop(nd->nl_ncp);
597 	nd->nl_ncp = ncp;
598 	nd->nl_flags |= NLC_NCPISLOCKED;
599 	error = 0;
600 	break;
601     }
602     return(error);
603 }
604 
605 /*
606  * Resolve a mount point's glue ncp.  This ncp connects creates the illusion
607  * of continuity in the namecache tree by connecting the ncp related to the
608  * vnode under the mount to the ncp related to the mount's root vnode.
609  *
610  * If no error occured a locked, ref'd ncp is stored in *ncpp.
611  */
612 int
613 nlookup_mp(struct mount *mp, struct namecache **ncpp)
614 {
615     struct namecache *ncp;
616     struct vnode *vp;
617     int error;
618 
619     error = 0;
620     ncp = mp->mnt_ncp;
621     cache_get(ncp);
622     if (ncp->nc_flag & NCF_UNRESOLVED) {
623 	while (vfs_busy(mp, 0, curthread))
624 	    ;
625 	error = VFS_ROOT(mp, &vp);
626 	vfs_unbusy(mp, curthread);
627 	if (error) {
628 	    cache_put(ncp);
629 	    ncp = NULL;
630 	} else {
631 	    cache_setvp(ncp, vp);
632 	    vput(vp);
633 	}
634     }
635     *ncpp = ncp;
636     return(error);
637 }
638 
639 /*
640  * Read the contents of a symlink, allocate a path buffer out of the
641  * namei_zone and initialize the supplied nlcomponent with the result.
642  *
643  * If an error occurs no buffer will be allocated or returned in the nlc.
644  */
645 int
646 nreadsymlink(struct nlookupdata *nd, struct namecache *ncp,
647 		struct nlcomponent *nlc)
648 {
649     struct vnode *vp;
650     struct iovec aiov;
651     struct uio auio;
652     int linklen;
653     int error;
654     char *cp;
655 
656     nlc->nlc_nameptr = NULL;
657     nlc->nlc_namelen = 0;
658     if (ncp->nc_vp == NULL)
659 	return(ENOENT);
660     if ((error = cache_vget(ncp, nd->nl_cred, LK_SHARED, &vp)) != 0)
661 	return(error);
662     cp = zalloc(namei_zone);
663     aiov.iov_base = cp;
664     aiov.iov_len = MAXPATHLEN;
665     auio.uio_iov = &aiov;
666     auio.uio_iovcnt = 1;
667     auio.uio_offset = 0;
668     auio.uio_rw = UIO_READ;
669     auio.uio_segflg = UIO_SYSSPACE;
670     auio.uio_td = nd->nl_td;
671     auio.uio_resid = MAXPATHLEN - 1;
672     error = VOP_READLINK(vp, &auio, nd->nl_cred);
673     if (error)
674 	goto fail;
675     linklen = MAXPATHLEN - 1 - auio.uio_resid;
676     if (varsym_enable) {
677 	linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
678 	if (linklen < 0) {
679 	    error = ENAMETOOLONG;
680 	    goto fail;
681 	}
682     }
683     cp[linklen] = 0;
684     nlc->nlc_nameptr = cp;
685     nlc->nlc_namelen = linklen;
686     vput(vp);
687     return(0);
688 fail:
689     zfree(namei_zone, cp);
690     vput(vp);
691     return(error);
692 }
693 
694 /*
695  * Check access [XXX cache vattr!] [XXX quota]
696  *
697  * Generally check the V* access bits from sys/vnode.h.  All specified bits
698  * must pass for this function to return 0.
699  *
700  * If VCREATE is specified and the target ncp represents a non-existant
701  * file or dir, or if VDELETE is specified and the target exists, the parent
702  * directory is checked for VWRITE.  If VEXCL is specified and the target
703  * ncp represents a positive hit, an error is returned.
704  *
705  * If VCREATE is not specified and the target does not exist (negative hit),
706  * ENOENT is returned.  Note that nlookup() does not (and should not) return
707  * ENOENT for non-existant leafs.
708  *
709  * The passed ncp may or may not be locked.  The caller should use a
710  * locked ncp on leaf lookups, especially for VCREATE, VDELETE, and VEXCL
711  * checks.
712  */
713 int
714 naccess(struct namecache *ncp, int vmode, struct ucred *cred)
715 {
716     struct namecache *par;
717     struct vnode *vp;
718     struct vattr va;
719     int error;
720 
721     if (ncp->nc_flag & NCF_UNRESOLVED) {
722 	cache_lock(ncp);
723 	cache_resolve(ncp, cred);
724 	cache_unlock(ncp);
725     }
726     error = ncp->nc_error;
727     if (vmode & (VDELETE|VCREATE|VEXCL)) {
728 	if (((vmode & VCREATE) && ncp->nc_vp == NULL) ||
729 	    ((vmode & VDELETE) && ncp->nc_vp != NULL)
730 	) {
731 	    if ((par = ncp->nc_parent) == NULL) {
732 		if (error != EAGAIN)
733 			error = EINVAL;
734 	    } else {
735 		cache_hold(par);
736 		error = naccess(par, VWRITE, cred);
737 		cache_drop(par);
738 	    }
739 	}
740 	if ((vmode & VEXCL) && ncp->nc_vp != NULL)
741 	    error = EEXIST;
742     }
743     if (error == 0) {
744 	error = cache_vget(ncp, cred, LK_SHARED, &vp);
745 	if (error == ENOENT) {
746 	    if (vmode & VCREATE)
747 		error = 0;
748 	} else if (error == 0) {
749 	    /* XXX cache the va in the namecache or in the vnode */
750 	    if ((error = VOP_GETATTR(vp, &va, curthread)) == 0) {
751 		if ((vmode & VWRITE) && vp->v_mount) {
752 		    if (vp->v_mount->mnt_flag & MNT_RDONLY)
753 			error = EROFS;
754 		}
755 	    }
756 	    vput(vp);
757 	    if (error == 0)
758 		error = naccess_va(&va, vmode, cred);
759 	}
760     }
761     return(error);
762 }
763 
764 /*
765  * Check the requested access against the given vattr using cred.
766  */
767 int
768 naccess_va(struct vattr *va, int vmode, struct ucred *cred)
769 {
770     int i;
771 
772     /*
773      * Test the immutable bit for files, directories, and softlinks.
774      */
775     if (vmode & (VWRITE|VDELETE)) {
776 	if (va->va_type == VDIR || va->va_type == VLNK || va->va_type == VREG) {
777 	    if (va->va_flags & IMMUTABLE)
778 		return (EPERM);
779 	}
780     }
781 
782     /*
783      * root gets universal access
784      */
785     if (cred->cr_uid == 0)
786 	return(0);
787 
788     /*
789      * Check owner perms, group perms, and world perms
790      */
791     vmode &= S_IRWXU;
792     if (cred->cr_uid == va->va_uid) {
793 	if ((vmode & va->va_mode) != vmode)
794 	    return(EACCES);
795 	return(0);
796     }
797 
798     vmode >>= 3;
799     for (i = 0; i < cred->cr_ngroups; ++i) {
800 	if (va->va_gid == cred->cr_groups[i]) {
801 	    if ((vmode & va->va_mode) != vmode)
802 		return(EACCES);
803 	    return(0);
804 	}
805     }
806 
807     vmode >>= 3;
808     if ((vmode & va->va_mode) != vmode)
809 	return(EACCES);
810     return(0);
811 }
812 
813