Add kernel-layer support for chflags checks, remove (most) from the VFS layer.
[dragonfly.git] / sys / kern / vfs_nlookup.c
CommitLineData
690a3127
MD
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 *
c9aef211 34 * $DragonFly: src/sys/kern/vfs_nlookup.c,v 1.25 2008/07/19 04:43:33 dillon Exp $
690a3127
MD
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>
21739618 66#include <sys/stat.h>
70aac194 67#include <sys/objcache.h>
690a3127
MD
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).
1f95166e
HP
76 *
77 * The first process proc0's credentials are used if the calling thread
78 * is not associated with a process context.
690a3127
MD
79 */
80int
21739618
MD
81nlookup_init(struct nlookupdata *nd,
82 const char *path, enum uio_seg seg, int flags)
690a3127
MD
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));
70aac194 96 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
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) {
21739618 111 if (p && p->p_fd) {
28623bf9
MD
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);
21739618 116 nd->nl_cred = crhold(p->p_ucred);
690a3127 117 } else {
28623bf9
MD
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);
21739618 121 nd->nl_cred = crhold(proc0.p_ucred);
690a3127
MD
122 }
123 nd->nl_td = td;
690a3127
MD
124 nd->nl_flags |= flags;
125 } else {
126 nlookup_done(nd);
127 }
128 return(error);
129}
130
131/*
fad57d0e 132 * This works similarly to nlookup_init() but does not assume a process
28623bf9 133 * context. rootnch is always chosen for the root directory and the cred
fad57d0e
MD
134 * and starting directory are supplied in arguments.
135 */
136int
137nlookup_init_raw(struct nlookupdata *nd,
138 const char *path, enum uio_seg seg, int flags,
28623bf9 139 struct ucred *cred, struct nchandle *ncstart)
fad57d0e
MD
140{
141 size_t pathlen;
142 thread_t td;
143 int error;
144
145 td = curthread;
146
147 bzero(nd, sizeof(struct nlookupdata));
70aac194 148 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
fad57d0e
MD
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) {
28623bf9
MD
163 cache_copy(ncstart, &nd->nl_nch);
164 cache_copy(&rootnch, &nd->nl_rootnch);
165 cache_copy(&rootnch, &nd->nl_jailnch);
fad57d0e
MD
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/*
1f95166e
HP
176 * Set a different credential; this credential will be used by future
177 * operations performed on nd.nl_open_vp and nlookupdata structure.
178 */
179void
180nlookup_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/*
21739618
MD
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.
690a3127
MD
197 */
198void
199nlookup_done(struct nlookupdata *nd)
200{
28623bf9 201 if (nd->nl_nch.ncp) {
21739618
MD
202 if (nd->nl_flags & NLC_NCPISLOCKED) {
203 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 204 cache_unlock(&nd->nl_nch);
21739618 205 }
28623bf9 206 cache_drop(&nd->nl_nch);
690a3127 207 }
28623bf9
MD
208 if (nd->nl_rootnch.ncp)
209 cache_drop(&nd->nl_rootnch);
210 if (nd->nl_jailnch.ncp)
211 cache_drop(&nd->nl_jailnch);
690a3127 212 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
70aac194 213 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
214 nd->nl_path = NULL;
215 }
216 if (nd->nl_cred) {
217 crfree(nd->nl_cred);
218 nd->nl_cred = NULL;
219 }
fad57d0e
MD
220 if (nd->nl_open_vp) {
221 if (nd->nl_flags & NLC_LOCKVP) {
a11aaa81 222 vn_unlock(nd->nl_open_vp);
fad57d0e
MD
223 nd->nl_flags &= ~NLC_LOCKVP;
224 }
87de5057 225 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
fad57d0e
MD
226 nd->nl_open_vp = NULL;
227 }
5312fa43
MD
228 if (nd->nl_dvp) {
229 vrele(nd->nl_dvp);
230 nd->nl_dvp = NULL;
231 }
fad57d0e
MD
232 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
233}
234
235void
236nlookup_zero(struct nlookupdata *nd)
237{
238 bzero(nd, sizeof(struct nlookupdata));
690a3127
MD
239}
240
241/*
21739618
MD
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.
690a3127 248 */
28623bf9 249struct nchandle
21739618
MD
250nlookup_simple(const char *str, enum uio_seg seg,
251 int niflags, int *error)
690a3127
MD
252{
253 struct nlookupdata nd;
28623bf9 254 struct nchandle nch;
690a3127
MD
255
256 *error = nlookup_init(&nd, str, seg, niflags);
257 if (*error == 0) {
21739618 258 if ((*error = nlookup(&nd)) == 0) {
28623bf9
MD
259 nch = nd.nl_nch; /* keep hold ref from structure */
260 cache_zero(&nd.nl_nch); /* and NULL out */
21739618 261 } else {
28623bf9 262 cache_zero(&nch);
21739618 263 }
690a3127
MD
264 nlookup_done(&nd);
265 } else {
28623bf9 266 cache_zero(&nch);
690a3127 267 }
28623bf9 268 return(nch);
690a3127
MD
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
28623bf9 274 * nl_nch is always referenced and locked, otherwise it may or may not be.
21739618
MD
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.
fad57d0e 279 *
945b476a
MD
280 * If NLC_CREATE is set the last directory must allow node creation,
281 * and an error code of 0 will be returned for a non-existant
282 * target (not ENOENT).
283 *
3a907475 284 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
945b476a 285 * plus the sticky check is made, and an error code of 0 will be returned
3a907475 286 * for a non-existant target (not ENOENT).
945b476a
MD
287 *
288 * If NLC_DELETE is set the last directory mut allow node deletion,
289 * plus the sticky check is made.
5312fa43
MD
290 *
291 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
292 * of the returned entry. The vnode will be referenced, but not locked,
293 * and will be released by nlookup_done() along with everything else.
690a3127
MD
294 */
295int
296nlookup(struct nlookupdata *nd)
297{
298 struct nlcomponent nlc;
28623bf9 299 struct nchandle nch;
3a907475 300 struct nchandle par;
28623bf9 301 struct mount *mp;
f4d4e93a 302 int wasdotordotdot;
690a3127 303 char *ptr;
fad57d0e 304 char *xptr;
690a3127
MD
305 int error;
306 int len;
3a907475 307 int dflags;
690a3127
MD
308
309#ifdef KTRACE
310 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
9fb04d14 311 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
690a3127
MD
312#endif
313 bzero(&nlc, sizeof(nlc));
314
315 /*
316 * Setup for the loop. The current working namecache element must
524c845c
MD
317 * be in a refd + unlocked state. This typically the case on entry except
318 * when stringing nlookup()'s along in a chain, since nlookup() always
28623bf9 319 * returns nl_nch in a locked state.
690a3127
MD
320 */
321 nd->nl_loopcnt = 0;
322 if (nd->nl_flags & NLC_NCPISLOCKED) {
323 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 324 cache_unlock(&nd->nl_nch);
690a3127 325 }
5312fa43
MD
326 if (nd->nl_dvp ) {
327 vrele(nd->nl_dvp);
328 nd->nl_dvp = NULL;
329 }
690a3127
MD
330 ptr = nd->nl_path;
331
332 /*
28623bf9 333 * Loop on the path components. At the top of the loop nd->nl_nch
524c845c 334 * is ref'd and unlocked and represents our current position.
690a3127
MD
335 */
336 for (;;) {
337 /*
338 * Check if the root directory should replace the current
339 * directory. This is done at the start of a translation
340 * or after a symbolic link has been found. In other cases
341 * ptr will never be pointing at a '/'.
342 */
343 if (*ptr == '/') {
344 do {
345 ++ptr;
346 } while (*ptr == '/');
28623bf9
MD
347 cache_copy(&nd->nl_rootnch, &nch);
348 cache_drop(&nd->nl_nch);
349 nd->nl_nch = nch;
5312fa43
MD
350
351 /*
352 * Fast-track termination. There is no parent directory of
353 * the root in the same mount from the point of view of
354 * the caller so return EPERM if NLC_REFDVP is specified.
355 * e.g. 'rmdir /' is not allowed.
356 */
21739618 357 if (*ptr == 0) {
5312fa43
MD
358 if (nd->nl_flags & NLC_REFDVP) {
359 error = EPERM;
360 } else {
361 cache_lock(&nd->nl_nch);
362 nd->nl_flags |= NLC_NCPISLOCKED;
363 error = 0;
364 }
21739618
MD
365 break;
366 }
690a3127
MD
367 continue;
368 }
369
370 /*
28623bf9 371 * Check directory search permissions.
21739618 372 */
3a907475
MD
373 dflags = 0;
374 if ((error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags)) != 0)
21739618
MD
375 break;
376
377 /*
690a3127
MD
378 * Extract the path component
379 */
380 nlc.nlc_nameptr = ptr;
381 while (*ptr && *ptr != '/')
382 ++ptr;
383 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
384
385 /*
21739618
MD
386 * Lookup the path component in the cache, creating an unresolved
387 * entry if necessary. We have to handle "." and ".." as special
388 * cases.
389 *
390 * When handling ".." we have to detect a traversal back through a
28623bf9 391 * mount point. If we are at the root, ".." just returns the root.
524c845c 392 *
3a907475
MD
393 * When handling "." or ".." we also have to recalculate dflags
394 * since our dflags will be for some sub-directory instead of the
395 * parent dir.
396 *
28623bf9 397 * This subsection returns a locked, refd 'nch' unless it errors out.
fad57d0e
MD
398 * The namecache topology is not allowed to be disconnected, so
399 * encountering a NULL parent will generate EINVAL. This typically
400 * occurs when a directory is removed out from under a process.
21739618
MD
401 */
402 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
28623bf9 403 cache_get(&nd->nl_nch, &nch);
f4d4e93a 404 wasdotordotdot = 1;
21739618
MD
405 } else if (nlc.nlc_namelen == 2 &&
406 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
28623bf9
MD
407 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
408 nd->nl_nch.ncp == nd->nl_rootnch.ncp
409 ) {
410 /*
411 * ".." at the root returns the root
412 */
413 cache_get(&nd->nl_nch, &nch);
21739618 414 } else {
28623bf9
MD
415 /*
416 * Locate the parent ncp. If we are at the root of a
417 * filesystem mount we have to skip to the mounted-on
418 * point in the underlying filesystem.
419 */
420 nch = nd->nl_nch;
421 while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
e5cde29c 422 nch = nch.mount->mnt_ncmounton;
28623bf9
MD
423 nch.ncp = nch.ncp->nc_parent;
424 KKASSERT(nch.ncp != NULL);
425 cache_get(&nch, &nch);
21739618 426 }
f4d4e93a 427 wasdotordotdot = 1;
21739618 428 } else {
28623bf9
MD
429 nch = cache_nlookup(&nd->nl_nch, &nlc);
430 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
6ea70f76 431 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
28623bf9
MD
432 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
433 cache_put(&nch);
434 nch = cache_nlookup(&nd->nl_nch, &nlc);
8e005a45 435 }
f4d4e93a 436 wasdotordotdot = 0;
21739618 437 }
3a907475
MD
438
439 /*
440 * If the last component was "." or ".." our dflags no longer
441 * represents the parent directory and we have to explicitly
442 * look it up.
443 */
444 if (wasdotordotdot && error == 0) {
445 dflags = 0;
446 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
447 par.mount = nch.mount;
448 cache_hold(&par);
449 dflags = 0;
450 error = naccess(&par, 0, nd->nl_cred, &dflags);
451 cache_drop(&par);
452 }
453 }
454
524c845c 455 /*
28623bf9 456 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
524c845c 457 */
21739618
MD
458
459 /*
460 * Resolve the namespace if necessary. The ncp returned by
461 * cache_nlookup() is referenced and locked.
8e005a45
MD
462 *
463 * XXX neither '.' nor '..' should return EAGAIN since they were
464 * previously resolved and thus cannot be newly created ncp's.
690a3127 465 */
28623bf9
MD
466 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
467 error = cache_resolve(&nch, nd->nl_cred);
8e005a45 468 KKASSERT(error != EAGAIN);
690a3127 469 } else {
28623bf9 470 error = nch.ncp->nc_error;
690a3127 471 }
21739618
MD
472
473 /*
fad57d0e 474 * Early completion. ENOENT is not an error if this is the last
945b476a
MD
475 * component and NLC_CREATE or NLC_RENAME (rename target) was
476 * requested. Note that ncp->nc_error is left as ENOENT in that
477 * case, which we check later on.
478 *
f4d4e93a 479 * Also handle invalid '.' or '..' components terminating a path
945b476a
MD
480 * for a create/rename/delete. The standard requires this and pax
481 * pretty stupidly depends on it.
fad57d0e
MD
482 */
483 for (xptr = ptr; *xptr == '/'; ++xptr)
484 ;
f4d4e93a 485 if (*xptr == 0) {
3a907475
MD
486 if (error == ENOENT &&
487 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
488 ) {
489 if (nd->nl_flags & NLC_NFS_RDONLY) {
c9aef211 490 error = EROFS;
3a907475
MD
491 } else {
492 error = naccess(&nch, nd->nl_flags | dflags,
493 nd->nl_cred, NULL);
494 }
c9aef211 495 }
945b476a 496 if (error == 0 && wasdotordotdot &&
3a907475
MD
497 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
498 NLC_RENAME_SRC | NLC_RENAME_DST))) {
f4d4e93a 499 error = EINVAL;
945b476a 500 }
fad57d0e
MD
501 }
502
503 /*
504 * Early completion on error.
21739618 505 */
690a3127 506 if (error) {
28623bf9 507 cache_put(&nch);
690a3127
MD
508 break;
509 }
510
511 /*
512 * If the element is a symlink and it is either not the last
513 * element or it is the last element and we are allowed to
514 * follow symlinks, resolve the symlink.
515 */
28623bf9 516 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
690a3127
MD
517 (*ptr || (nd->nl_flags & NLC_FOLLOW))
518 ) {
519 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
21739618 520 error = ELOOP;
28623bf9 521 cache_put(&nch);
690a3127
MD
522 break;
523 }
28623bf9
MD
524 error = nreadsymlink(nd, &nch, &nlc);
525 cache_put(&nch);
21739618
MD
526 if (error)
527 break;
690a3127
MD
528
529 /*
530 * Concatenate trailing path elements onto the returned symlink.
531 * Note that if the path component (ptr) is not exhausted, it
532 * will being with a '/', so we do not have to add another one.
533 *
534 * The symlink may not be empty.
535 */
536 len = strlen(ptr);
537 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
538 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
70aac194 539 objcache_put(namei_oc, nlc.nlc_nameptr);
690a3127
MD
540 break;
541 }
542 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
543 if (nd->nl_flags & NLC_HASBUF)
70aac194 544 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
545 nd->nl_path = nlc.nlc_nameptr;
546 nd->nl_flags |= NLC_HASBUF;
547 ptr = nd->nl_path;
548
549 /*
550 * Go back up to the top to resolve any initial '/'s in the
551 * symlink.
552 */
553 continue;
554 }
555
556 /*
21739618 557 * If the element is a directory and we are crossing a mount point,
28623bf9 558 * Locate the mount.
21739618 559 */
28623bf9
MD
560 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
561 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
562 (mp = cache_findmount(&nch)) != NULL
21739618 563 ) {
21739618
MD
564 struct vnode *tdp;
565
28623bf9
MD
566 cache_put(&nch);
567 cache_get(&mp->mnt_ncmountpt, &nch);
21739618 568
28623bf9 569 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 570 while (vfs_busy(mp, 0))
21739618
MD
571 ;
572 error = VFS_ROOT(mp, &tdp);
f9642f56 573 vfs_unbusy(mp);
524c845c 574 if (error)
21739618 575 break;
28623bf9 576 cache_setvp(&nch, tdp);
21739618
MD
577 vput(tdp);
578 }
579 }
580 if (error) {
28623bf9 581 cache_put(&nch);
21739618
MD
582 break;
583 }
584
585 /*
690a3127
MD
586 * Skip any slashes to get to the next element. If there
587 * are any slashes at all the current element must be a
fad57d0e
MD
588 * directory or, in the create case, intended to become a directory.
589 * If it isn't we break without incrementing ptr and fall through
590 * to the failure case below.
690a3127
MD
591 */
592 while (*ptr == '/') {
28623bf9 593 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
fad57d0e
MD
594 !(nd->nl_flags & NLC_WILLBEDIR)
595 ) {
690a3127 596 break;
fad57d0e 597 }
690a3127
MD
598 ++ptr;
599 }
600
601 /*
602 * Continuation case: additional elements and the current
603 * element is a directory.
604 */
28623bf9
MD
605 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
606 cache_drop(&nd->nl_nch);
607 cache_unlock(&nch);
608 nd->nl_nch = nch;
690a3127
MD
609 continue;
610 }
611
612 /*
613 * Failure case: additional elements and the current element
614 * is not a directory
615 */
616 if (*ptr) {
28623bf9 617 cache_put(&nch);
690a3127
MD
618 error = ENOTDIR;
619 break;
620 }
621
622 /*
fad57d0e
MD
623 * Successful lookup of last element.
624 *
3a907475
MD
625 * Check permissions if the target exists. If the target does not
626 * exist directory permissions were already tested in the early
627 * completion code above.
945b476a 628 *
3a907475
MD
629 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
630 * if the file is marked append-only, and NLC_STICKY if the directory
631 * containing the file is sticky.
fad57d0e 632 */
3a907475
MD
633 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
634 error = naccess(&nch, nd->nl_flags | dflags,
635 nd->nl_cred, NULL);
945b476a 636 if (error) {
28623bf9 637 cache_put(&nch);
fad57d0e
MD
638 break;
639 }
640 }
641
642 /*
3a907475 643 * Termination: no more elements.
5312fa43
MD
644 *
645 * If NLC_REFDVP is set acquire a referenced parent dvp.
690a3127 646 */
5312fa43
MD
647 if (nd->nl_flags & NLC_REFDVP) {
648 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
649 if (error) {
650 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
651 cache_put(&nch);
652 break;
653 }
654 }
28623bf9
MD
655 cache_drop(&nd->nl_nch);
656 nd->nl_nch = nch;
690a3127
MD
657 nd->nl_flags |= NLC_NCPISLOCKED;
658 error = 0;
659 break;
660 }
3a907475
MD
661
662 /*
663 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
664 * (ncp->nc_error will be ENOENT), but we will still return an error
665 * code of 0.
666 */
690a3127
MD
667 return(error);
668}
669
670/*
21739618
MD
671 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
672 * of continuity in the namecache tree by connecting the ncp related to the
673 * vnode under the mount to the ncp related to the mount's root vnode.
674 *
675 * If no error occured a locked, ref'd ncp is stored in *ncpp.
676 */
677int
28623bf9 678nlookup_mp(struct mount *mp, struct nchandle *nch)
21739618 679{
21739618
MD
680 struct vnode *vp;
681 int error;
682
683 error = 0;
28623bf9
MD
684 cache_get(&mp->mnt_ncmountpt, nch);
685 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 686 while (vfs_busy(mp, 0))
21739618
MD
687 ;
688 error = VFS_ROOT(mp, &vp);
f9642f56 689 vfs_unbusy(mp);
21739618 690 if (error) {
28623bf9 691 cache_put(nch);
21739618 692 } else {
28623bf9 693 cache_setvp(nch, vp);
21739618
MD
694 vput(vp);
695 }
696 }
21739618
MD
697 return(error);
698}
699
700/*
690a3127 701 * Read the contents of a symlink, allocate a path buffer out of the
70aac194 702 * namei_oc and initialize the supplied nlcomponent with the result.
690a3127
MD
703 *
704 * If an error occurs no buffer will be allocated or returned in the nlc.
705 */
706int
28623bf9 707nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
690a3127
MD
708 struct nlcomponent *nlc)
709{
21739618 710 struct vnode *vp;
690a3127
MD
711 struct iovec aiov;
712 struct uio auio;
713 int linklen;
714 int error;
715 char *cp;
716
717 nlc->nlc_nameptr = NULL;
718 nlc->nlc_namelen = 0;
28623bf9 719 if (nch->ncp->nc_vp == NULL)
690a3127 720 return(ENOENT);
28623bf9 721 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
690a3127 722 return(error);
70aac194 723 cp = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
724 aiov.iov_base = cp;
725 aiov.iov_len = MAXPATHLEN;
726 auio.uio_iov = &aiov;
727 auio.uio_iovcnt = 1;
728 auio.uio_offset = 0;
729 auio.uio_rw = UIO_READ;
730 auio.uio_segflg = UIO_SYSSPACE;
731 auio.uio_td = nd->nl_td;
732 auio.uio_resid = MAXPATHLEN - 1;
21739618 733 error = VOP_READLINK(vp, &auio, nd->nl_cred);
690a3127
MD
734 if (error)
735 goto fail;
21739618 736 linklen = MAXPATHLEN - 1 - auio.uio_resid;
690a3127
MD
737 if (varsym_enable) {
738 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
739 if (linklen < 0) {
740 error = ENAMETOOLONG;
741 goto fail;
742 }
743 }
744 cp[linklen] = 0;
745 nlc->nlc_nameptr = cp;
746 nlc->nlc_namelen = linklen;
21739618 747 vput(vp);
690a3127
MD
748 return(0);
749fail:
70aac194 750 objcache_put(namei_oc, cp);
21739618 751 vput(vp);
690a3127
MD
752 return(error);
753}
754
21739618
MD
755/*
756 * Check access [XXX cache vattr!] [XXX quota]
757 *
3a907475
MD
758 * Generally check the NLC_* access bits. All specified bits must pass
759 * for this function to return 0.
21739618 760 *
3a907475
MD
761 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
762 * access, otherwise it must exist. No error is returned in this case.
945b476a 763 *
3a907475 764 * The file must not exist if NLC_EXCL is specified.
21739618 765 *
3a907475
MD
766 * Directory permissions in general are tested for NLC_CREATE if the file
767 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
768 * whether the file exists or not.
21739618 769 *
3a907475
MD
770 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
771 * the latter is only tested if the target exists.
dae8d54f 772 *
21739618 773 * The passed ncp may or may not be locked. The caller should use a
3a907475
MD
774 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
775 * NLC_DELETE, and NLC_EXCL checks.
21739618
MD
776 */
777int
3a907475 778naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
21739618 779{
28623bf9 780 struct nchandle par;
21739618
MD
781 struct vnode *vp;
782 struct vattr va;
783 int error;
dae8d54f 784 int sticky;
21739618 785
28623bf9
MD
786 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
787 cache_lock(nch);
788 cache_resolve(nch, cred);
789 cache_unlock(nch);
21739618 790 }
28623bf9 791 error = nch->ncp->nc_error;
dae8d54f 792
945b476a 793 /*
3a907475
MD
794 * Directory permissions checks. Silently ignore ENOENT if these
795 * tests pass. It isn't an error.
945b476a 796 */
3a907475
MD
797 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
798 if (((nflags & NLC_CREATE) && nch->ncp->nc_vp == NULL) ||
799 ((nflags & NLC_DELETE) && nch->ncp->nc_vp != NULL) ||
800 ((nflags & NLC_RENAME_SRC) && nch->ncp->nc_vp != NULL) ||
801 (nflags & NLC_RENAME_DST)
21739618 802 ) {
28623bf9 803 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
8e005a45 804 if (error != EAGAIN)
fad57d0e 805 error = EINVAL;
3a907475 806 } else if (error == 0 || error == ENOENT) {
28623bf9
MD
807 par.mount = nch->mount;
808 cache_hold(&par);
3a907475
MD
809 sticky = 0;
810 error = naccess(&par, NLC_WRITE, cred, NULL);
28623bf9 811 cache_drop(&par);
8e005a45 812 }
21739618 813 }
21739618 814 }
3a907475
MD
815
816 /*
817 * NLC_EXCL check. Target file must not exist.
818 */
819 if (error == 0 && (nflags & NLC_EXCL) && nch->ncp->nc_vp != NULL)
820 error = EEXIST;
821
822 /*
823 * Get the vnode attributes so we can do the rest of our checks.
824 *
825 * NOTE: We only call naccess_va() if the target exists.
826 */
21739618 827 if (error == 0) {
28623bf9 828 error = cache_vget(nch, cred, LK_SHARED, &vp);
21739618 829 if (error == ENOENT) {
3a907475
MD
830 /*
831 * Silently zero-out ENOENT if creating or renaming
832 * (rename target). It isn't an error.
833 */
834 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
21739618
MD
835 error = 0;
836 } else if (error == 0) {
3a907475
MD
837 /*
838 * Get the vnode attributes and check for illegal O_TRUNC
839 * requests and read-only mounts.
840 *
841 * NOTE: You can still open devices on read-only mounts for
842 * writing.
843 *
844 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
845 * check on the parent directory above.
846 *
847 * XXX cache the va in the namecache or in the vnode
848 */
849 error = VOP_GETATTR(vp, &va);
850 if (error == 0 && (nflags & NLC_TRUNCATE)) {
851 switch(va.va_type) {
852 case VREG:
853 case VDATABASE:
854 case VCHR:
855 case VBLK:
856 break;
857 default:
858 error = EINVAL;
859 break;
860 }
861 }
862 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
863 (vp->v_mount->mnt_flag & MNT_RDONLY)
864 ) {
865 switch(va.va_type) {
866 case VDIR:
867 case VLNK:
868 case VREG:
869 case VDATABASE:
870 error = EROFS;
871 break;
872 default:
873 break;
21739618
MD
874 }
875 }
876 vput(vp);
dae8d54f 877
3a907475
MD
878 /*
879 * Check permissions based on file attributes. The passed
880 * flags (*nflagsp) are modified with feedback based on
881 * special attributes and requirements.
882 */
dae8d54f
MD
883 if (error == 0) {
884 /*
3a907475 885 * Adjust the returned (*nflagsp) if non-NULL.
dae8d54f 886 */
3a907475
MD
887 if (nflagsp) {
888 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
889 *nflagsp |= NLC_STICKY;
890 if (va.va_flags & APPEND)
891 *nflagsp |= NLC_APPENDONLY;
892 if (va.va_flags & IMMUTABLE)
893 *nflagsp |= NLC_IMMUTABLE;
894 }
dae8d54f
MD
895
896 /*
897 * Process general access.
898 */
3a907475 899 error = naccess_va(&va, nflags, cred);
dae8d54f 900 }
21739618
MD
901 }
902 }
903 return(error);
904}
905
906/*
907 * Check the requested access against the given vattr using cred.
908 */
909int
3a907475 910naccess_va(struct vattr *va, int nflags, struct ucred *cred)
21739618
MD
911{
912 int i;
3a907475 913 int vmode;
21739618
MD
914
915 /*
3a907475
MD
916 * Test the immutable bit. Creations, deletions, renames (source
917 * or destination) are not allowed. chown/chmod/other is also not
918 * allowed but is handled by SETATTR. Hardlinks to the immutable
919 * file are allowed.
920 *
921 * If the directory is set to immutable then creations, deletions,
922 * renames (source or dest) and hardlinks to files within the directory
923 * are not allowed, and regular files opened through the directory may
924 * not be written to or truncated (unless a special device).
945b476a 925 *
3a907475
MD
926 * NOTE! New hardlinks to immutable files work but new hardlinks to
927 * files, immutable or not, sitting inside an immutable directory are
928 * not allowed. As always if the file is hardlinked via some other
929 * path additional hardlinks may be possible even if the file is marked
930 * immutable. The sysop needs to create a closure by checking the hard
931 * link count. Once closure is achieved you are good, and security
932 * scripts should check link counts anyway.
933 *
934 * Writes and truncations are only allowed on special devices.
21739618 935 */
3a907475
MD
936 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
937 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
938 return (EPERM);
939 if (nflags & (NLC_CREATE | NLC_DELETE |
940 NLC_RENAME_SRC | NLC_RENAME_DST)) {
941 return (EPERM);
942 }
943 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
944 switch(va->va_type) {
945 case VDIR:
946 case VLNK:
947 case VREG:
948 case VDATABASE:
949 return (EPERM);
950 default:
951 break;
952 }
953 }
954 }
955
956 /*
957 * Test the no-unlink and append-only bits for opens, rename targets,
958 * and deletions. These bits are not tested for creations or
959 * rename sources.
960 *
961 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
962 * If you do not wish this you must also set NOUNLINK.
963 *
964 * If the governing directory is marked APPEND-only it implies
965 * NOUNLINK for all entries in the directory.
966 */
967 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
968 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
969 ) {
970 return (EPERM);
971 }
972
973 /*
974 * A file marked append-only may not be deleted but can be renamed.
975 */
976 if ((va->va_flags & APPEND) &&
977 (nflags & (NLC_DELETE | NLC_RENAME_DST))
978 ) {
979 return (EPERM);
980 }
981
982 /*
983 * A file marked append-only which is opened for writing must also
984 * be opened O_APPEND.
985 */
986 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
987 if (nflags & NLC_TRUNCATE)
988 return (EPERM);
989 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
990 if ((nflags & NLC_APPEND) == 0)
21739618
MD
991 return (EPERM);
992 }
993 }
994
995 /*
996 * root gets universal access
997 */
998 if (cred->cr_uid == 0)
999 return(0);
1000
1001 /*
dae8d54f
MD
1002 * Check owner perms.
1003 *
3a907475 1004 * If NLC_OWN is set the owner of the file is allowed no matter when
dae8d54f 1005 * the owner-mode bits say (utimes).
21739618 1006 */
3a907475
MD
1007 vmode = 0;
1008 if (nflags & NLC_READ)
1009 vmode |= S_IRUSR;
1010 if (nflags & NLC_WRITE)
1011 vmode |= S_IWUSR;
1012 if (nflags & NLC_EXEC)
1013 vmode |= S_IXUSR;
1014
21739618 1015 if (cred->cr_uid == va->va_uid) {
3a907475 1016 if ((nflags & NLC_OWN) == 0) {
ee89633d
MD
1017 if ((vmode & va->va_mode) != vmode)
1018 return(EACCES);
1019 }
21739618
MD
1020 return(0);
1021 }
dae8d54f
MD
1022
1023 /*
3a907475
MD
1024 * If NLC_STICKY is set only the owner may delete or rename a file.
1025 * This bit is typically set on /tmp.
dae8d54f 1026 *
3a907475
MD
1027 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1028 * the specific delete or rename case. For deletions and renames we
1029 * usually just care about directory permissions, not file permissions.
dae8d54f 1030 */
3a907475
MD
1031 if ((nflags & NLC_STICKY) &&
1032 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
dae8d54f 1033 return(EACCES);
3a907475 1034 }
dae8d54f
MD
1035
1036 /*
1037 * Check group perms
1038 */
21739618
MD
1039 vmode >>= 3;
1040 for (i = 0; i < cred->cr_ngroups; ++i) {
1041 if (va->va_gid == cred->cr_groups[i]) {
1042 if ((vmode & va->va_mode) != vmode)
1043 return(EACCES);
1044 return(0);
1045 }
1046 }
1047
dae8d54f
MD
1048 /*
1049 * Check world perms
1050 */
21739618
MD
1051 vmode >>= 3;
1052 if ((vmode & va->va_mode) != vmode)
1053 return(EACCES);
1054 return(0);
1055}
1056