kernel - Fix mount refs interactions and umount races
[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.
690a3127
MD
33 */
34/*
35 * nlookup() is the 'new' namei interface. Rather then return directory and
36 * leaf vnodes (in various lock states) the new interface instead deals in
37 * namecache records. Namecache records may represent both a positive or
38 * a negative hit. The namespace is locked via the namecache record instead
39 * of via the vnode, and only the leaf namecache record (representing the
40 * filename) needs to be locked.
41 *
42 * This greatly improves filesystem parallelism and is a huge simplification
43 * of the API verses the old vnode locking / namei scheme.
44 *
45 * Filesystems must actively control the caching aspects of the namecache,
46 * and since namecache pointers are used as handles they are non-optional
47 * even for filesystems which do not generally wish to cache things. It is
48 * intended that a separate cache coherency API will be constructed to handle
49 * these issues.
50 */
51
52#include "opt_ktrace.h"
53
54#include <sys/param.h>
55#include <sys/systm.h>
56#include <sys/kernel.h>
57#include <sys/vnode.h>
58#include <sys/mount.h>
59#include <sys/filedesc.h>
60#include <sys/proc.h>
61#include <sys/namei.h>
62#include <sys/nlookup.h>
63#include <sys/malloc.h>
21739618 64#include <sys/stat.h>
70aac194 65#include <sys/objcache.h>
1db695af 66#include <sys/file.h>
690a3127
MD
67
68#ifdef KTRACE
69#include <sys/ktrace.h>
70#endif
71
e9b56058 72static int naccess(struct nchandle *nch, int vmode, struct ucred *cred,
aabd5ce8 73 int *stickyp);
e9b56058 74
690a3127
MD
75/*
76 * Initialize a nlookup() structure, early error return for copyin faults
77 * or a degenerate empty string (which is not allowed).
1f95166e
HP
78 *
79 * The first process proc0's credentials are used if the calling thread
80 * is not associated with a process context.
61f96b6f
MD
81 *
82 * MPSAFE
690a3127
MD
83 */
84int
21739618
MD
85nlookup_init(struct nlookupdata *nd,
86 const char *path, enum uio_seg seg, int flags)
690a3127
MD
87{
88 size_t pathlen;
89 struct proc *p;
90 thread_t td;
91 int error;
92
93 td = curthread;
94 p = td->td_proc;
95
96 /*
97 * note: the pathlen set by copy*str() includes the terminating \0.
98 */
99 bzero(nd, sizeof(struct nlookupdata));
70aac194 100 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
101 nd->nl_flags |= NLC_HASBUF;
102 if (seg == UIO_SYSSPACE)
103 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
104 else
105 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
106
107 /*
108 * Don't allow empty pathnames.
109 * POSIX.1 requirement: "" is not a vaild file name.
110 */
111 if (error == 0 && pathlen <= 1)
112 error = ENOENT;
113
114 if (error == 0) {
21739618 115 if (p && p->p_fd) {
28623bf9
MD
116 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch);
117 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch);
118 if (p->p_fd->fd_njdir.ncp)
119 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch);
21739618 120 nd->nl_cred = crhold(p->p_ucred);
690a3127 121 } else {
28623bf9
MD
122 cache_copy(&rootnch, &nd->nl_nch);
123 cache_copy(&nd->nl_nch, &nd->nl_rootnch);
124 cache_copy(&nd->nl_nch, &nd->nl_jailnch);
21739618 125 nd->nl_cred = crhold(proc0.p_ucred);
690a3127
MD
126 }
127 nd->nl_td = td;
690a3127
MD
128 nd->nl_flags |= flags;
129 } else {
130 nlookup_done(nd);
131 }
132 return(error);
133}
134
1db695af
NT
135
136/*
137 * nlookup_init() for "at" family of syscalls.
138 *
139 * Works similarly to nlookup_init() but if path is relative and fd is not
140 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd.
141 * In this case, the file entry pointed to by fd is ref'ed and returned in
142 * *fpp.
143 *
144 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd
145 * and release the ref to the file entry.
146 */
147int
148nlookup_init_at(struct nlookupdata *nd, struct file **fpp, int fd,
149 const char *path, enum uio_seg seg, int flags)
150{
151 struct thread *td = curthread;
152 struct proc *p = td->td_proc;
153 struct file* fp;
154 struct vnode *vp;
155 int error;
156
157 *fpp = NULL;
158
159 if ((error = nlookup_init(nd, path, seg, flags)) != 0) {
160 return (error);
161 }
162
163 if (nd->nl_path[0] != '/' && fd != AT_FDCWD) {
164 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
165 goto done;
166 vp = (struct vnode*)fp->f_data;
167 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL) {
168 fdrop(fp);
169 fp = NULL;
170 error = ENOTDIR;
171 goto done;
172 }
173 cache_drop(&nd->nl_nch);
174 cache_copy(&fp->f_nchandle, &nd->nl_nch);
175 *fpp = fp;
176 }
177
178
179done:
180 if (error)
181 nlookup_done(nd);
182 return (error);
183
184}
185
690a3127 186/*
fad57d0e 187 * This works similarly to nlookup_init() but does not assume a process
28623bf9 188 * context. rootnch is always chosen for the root directory and the cred
fad57d0e
MD
189 * and starting directory are supplied in arguments.
190 */
191int
192nlookup_init_raw(struct nlookupdata *nd,
193 const char *path, enum uio_seg seg, int flags,
28623bf9 194 struct ucred *cred, struct nchandle *ncstart)
fad57d0e
MD
195{
196 size_t pathlen;
197 thread_t td;
198 int error;
199
200 td = curthread;
201
202 bzero(nd, sizeof(struct nlookupdata));
70aac194 203 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
fad57d0e
MD
204 nd->nl_flags |= NLC_HASBUF;
205 if (seg == UIO_SYSSPACE)
206 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
207 else
208 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
209
210 /*
211 * Don't allow empty pathnames.
212 * POSIX.1 requirement: "" is not a vaild file name.
213 */
214 if (error == 0 && pathlen <= 1)
215 error = ENOENT;
216
217 if (error == 0) {
28623bf9
MD
218 cache_copy(ncstart, &nd->nl_nch);
219 cache_copy(&rootnch, &nd->nl_rootnch);
220 cache_copy(&rootnch, &nd->nl_jailnch);
fad57d0e
MD
221 nd->nl_cred = crhold(cred);
222 nd->nl_td = td;
223 nd->nl_flags |= flags;
224 } else {
225 nlookup_done(nd);
226 }
227 return(error);
228}
229
230/*
46077e84
AH
231 * This works similarly to nlookup_init_raw() but does not rely
232 * on rootnch being initialized yet.
233 */
234int
235nlookup_init_root(struct nlookupdata *nd,
236 const char *path, enum uio_seg seg, int flags,
237 struct ucred *cred, struct nchandle *ncstart,
238 struct nchandle *ncroot)
239{
240 size_t pathlen;
241 thread_t td;
242 int error;
243
244 td = curthread;
245
246 bzero(nd, sizeof(struct nlookupdata));
247 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
248 nd->nl_flags |= NLC_HASBUF;
249 if (seg == UIO_SYSSPACE)
250 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
251 else
252 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
253
254 /*
255 * Don't allow empty pathnames.
256 * POSIX.1 requirement: "" is not a vaild file name.
257 */
258 if (error == 0 && pathlen <= 1)
259 error = ENOENT;
260
261 if (error == 0) {
262 cache_copy(ncstart, &nd->nl_nch);
263 cache_copy(ncroot, &nd->nl_rootnch);
264 cache_copy(ncroot, &nd->nl_jailnch);
265 nd->nl_cred = crhold(cred);
266 nd->nl_td = td;
267 nd->nl_flags |= flags;
268 } else {
269 nlookup_done(nd);
270 }
271 return(error);
272}
273
274/*
1f95166e
HP
275 * Set a different credential; this credential will be used by future
276 * operations performed on nd.nl_open_vp and nlookupdata structure.
277 */
278void
279nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
280{
281 KKASSERT(nd->nl_cred != NULL);
282
283 if (nd->nl_cred != cred) {
284 cred = crhold(cred);
285 crfree(nd->nl_cred);
286 nd->nl_cred = cred;
287 }
288}
289
290/*
21739618
MD
291 * Cleanup a nlookupdata structure after we are through with it. This may
292 * be called on any nlookupdata structure initialized with nlookup_init().
293 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
294 * returns an error, even if as a consumer you believe you have taken all
295 * dynamic elements out of the nlookupdata structure.
690a3127
MD
296 */
297void
298nlookup_done(struct nlookupdata *nd)
299{
28623bf9 300 if (nd->nl_nch.ncp) {
21739618
MD
301 if (nd->nl_flags & NLC_NCPISLOCKED) {
302 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 303 cache_unlock(&nd->nl_nch);
21739618 304 }
2247fe02 305 cache_drop(&nd->nl_nch); /* NULL's out the nch */
690a3127 306 }
28623bf9
MD
307 if (nd->nl_rootnch.ncp)
308 cache_drop(&nd->nl_rootnch);
309 if (nd->nl_jailnch.ncp)
310 cache_drop(&nd->nl_jailnch);
690a3127 311 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
70aac194 312 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
313 nd->nl_path = NULL;
314 }
315 if (nd->nl_cred) {
316 crfree(nd->nl_cred);
317 nd->nl_cred = NULL;
318 }
fad57d0e
MD
319 if (nd->nl_open_vp) {
320 if (nd->nl_flags & NLC_LOCKVP) {
a11aaa81 321 vn_unlock(nd->nl_open_vp);
fad57d0e
MD
322 nd->nl_flags &= ~NLC_LOCKVP;
323 }
87de5057 324 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
fad57d0e
MD
325 nd->nl_open_vp = NULL;
326 }
5312fa43
MD
327 if (nd->nl_dvp) {
328 vrele(nd->nl_dvp);
329 nd->nl_dvp = NULL;
330 }
fad57d0e
MD
331 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
332}
333
1db695af
NT
334/*
335 * Works similarly to nlookup_done() when nd initialized with
336 * nlookup_init_at().
337 */
338void
339nlookup_done_at(struct nlookupdata *nd, struct file *fp)
340{
341 nlookup_done(nd);
342 if (fp != NULL)
343 fdrop(fp);
344}
345
fad57d0e
MD
346void
347nlookup_zero(struct nlookupdata *nd)
348{
2247fe02 349 bzero(nd, sizeof(struct nlookupdata));
690a3127
MD
350}
351
352/*
21739618
MD
353 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
354 * if an error occured.
355 *
356 * Note that the returned ncp is not checked for permissions, though VEXEC
357 * is checked on the directory path leading up to the result. The caller
358 * must call naccess() to check the permissions of the returned leaf.
690a3127 359 */
28623bf9 360struct nchandle
21739618
MD
361nlookup_simple(const char *str, enum uio_seg seg,
362 int niflags, int *error)
690a3127
MD
363{
364 struct nlookupdata nd;
28623bf9 365 struct nchandle nch;
690a3127
MD
366
367 *error = nlookup_init(&nd, str, seg, niflags);
368 if (*error == 0) {
21739618 369 if ((*error = nlookup(&nd)) == 0) {
28623bf9
MD
370 nch = nd.nl_nch; /* keep hold ref from structure */
371 cache_zero(&nd.nl_nch); /* and NULL out */
21739618 372 } else {
28623bf9 373 cache_zero(&nch);
21739618 374 }
690a3127
MD
375 nlookup_done(&nd);
376 } else {
28623bf9 377 cache_zero(&nch);
690a3127 378 }
28623bf9 379 return(nch);
690a3127
MD
380}
381
382/*
383 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
384 * on return, even if an error occurs. If no error occurs the returned
28623bf9 385 * nl_nch is always referenced and locked, otherwise it may or may not be.
21739618
MD
386 *
387 * Intermediate directory elements, including the current directory, require
388 * execute (search) permission. nlookup does not examine the access
389 * permissions on the returned element.
fad57d0e 390 *
945b476a
MD
391 * If NLC_CREATE is set the last directory must allow node creation,
392 * and an error code of 0 will be returned for a non-existant
393 * target (not ENOENT).
394 *
3a907475 395 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
945b476a 396 * plus the sticky check is made, and an error code of 0 will be returned
3a907475 397 * for a non-existant target (not ENOENT).
945b476a
MD
398 *
399 * If NLC_DELETE is set the last directory mut allow node deletion,
400 * plus the sticky check is made.
5312fa43
MD
401 *
402 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
403 * of the returned entry. The vnode will be referenced, but not locked,
404 * and will be released by nlookup_done() along with everything else.
690a3127
MD
405 */
406int
407nlookup(struct nlookupdata *nd)
408{
5a9972aa 409 globaldata_t gd = mycpu;
690a3127 410 struct nlcomponent nlc;
28623bf9 411 struct nchandle nch;
3a907475 412 struct nchandle par;
f63911bf 413 struct nchandle nctmp;
28623bf9 414 struct mount *mp;
f4d4e93a 415 int wasdotordotdot;
690a3127 416 char *ptr;
fad57d0e 417 char *xptr;
690a3127
MD
418 int error;
419 int len;
3a907475 420 int dflags;
5a9972aa 421 int hit = 1;
690a3127
MD
422
423#ifdef KTRACE
424 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
9fb04d14 425 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
690a3127
MD
426#endif
427 bzero(&nlc, sizeof(nlc));
428
429 /*
2247fe02
MD
430 * Setup for the loop. The current working namecache element is
431 * always at least referenced. We lock it as required, but always
432 * return a locked, resolved namecache entry.
690a3127
MD
433 */
434 nd->nl_loopcnt = 0;
2247fe02 435 if (nd->nl_dvp) {
5312fa43
MD
436 vrele(nd->nl_dvp);
437 nd->nl_dvp = NULL;
438 }
690a3127
MD
439 ptr = nd->nl_path;
440
441 /*
28623bf9 442 * Loop on the path components. At the top of the loop nd->nl_nch
524c845c 443 * is ref'd and unlocked and represents our current position.
690a3127
MD
444 */
445 for (;;) {
446 /*
2247fe02
MD
447 * Make sure nl_nch is locked so we can access the vnode, resolution
448 * state, etc.
449 */
450 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) {
451 nd->nl_flags |= NLC_NCPISLOCKED;
452 cache_lock(&nd->nl_nch);
453 }
454
455 /*
690a3127
MD
456 * Check if the root directory should replace the current
457 * directory. This is done at the start of a translation
458 * or after a symbolic link has been found. In other cases
459 * ptr will never be pointing at a '/'.
460 */
461 if (*ptr == '/') {
462 do {
463 ++ptr;
464 } while (*ptr == '/');
2247fe02
MD
465 cache_get(&nd->nl_rootnch, &nch);
466 cache_put(&nd->nl_nch);
467 nd->nl_nch = nch; /* remains locked */
5312fa43
MD
468
469 /*
470 * Fast-track termination. There is no parent directory of
471 * the root in the same mount from the point of view of
1acebc70 472 * the caller so return EACCES if NLC_REFDVP is specified,
473 * and EEXIST if NLC_CREATE is also specified.
474 * e.g. 'rmdir /' or 'mkdir /' are not allowed.
5312fa43 475 */
21739618 476 if (*ptr == 0) {
2247fe02 477 if (nd->nl_flags & NLC_REFDVP)
1acebc70 478 error = (nd->nl_flags & NLC_CREATE) ? EEXIST : EACCES;
2247fe02 479 else
5312fa43 480 error = 0;
21739618
MD
481 break;
482 }
690a3127
MD
483 continue;
484 }
485
486 /*
28623bf9 487 * Check directory search permissions.
21739618 488 */
3a907475 489 dflags = 0;
aabd5ce8 490 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags);
e9b56058 491 if (error)
21739618
MD
492 break;
493
494 /*
d7c75c7a
MD
495 * Extract the path component. Path components are limited to
496 * 255 characters.
690a3127
MD
497 */
498 nlc.nlc_nameptr = ptr;
499 while (*ptr && *ptr != '/')
500 ++ptr;
501 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
d7c75c7a
MD
502 if (nlc.nlc_namelen >= 256) {
503 error = ENAMETOOLONG;
504 break;
505 }
690a3127
MD
506
507 /*
21739618
MD
508 * Lookup the path component in the cache, creating an unresolved
509 * entry if necessary. We have to handle "." and ".." as special
510 * cases.
511 *
512 * When handling ".." we have to detect a traversal back through a
28623bf9 513 * mount point. If we are at the root, ".." just returns the root.
524c845c 514 *
3a907475
MD
515 * When handling "." or ".." we also have to recalculate dflags
516 * since our dflags will be for some sub-directory instead of the
517 * parent dir.
518 *
28623bf9 519 * This subsection returns a locked, refd 'nch' unless it errors out.
fad57d0e
MD
520 * The namecache topology is not allowed to be disconnected, so
521 * encountering a NULL parent will generate EINVAL. This typically
522 * occurs when a directory is removed out from under a process.
21739618
MD
523 */
524 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
28623bf9 525 cache_get(&nd->nl_nch, &nch);
f4d4e93a 526 wasdotordotdot = 1;
21739618
MD
527 } else if (nlc.nlc_namelen == 2 &&
528 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
28623bf9
MD
529 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
530 nd->nl_nch.ncp == nd->nl_rootnch.ncp
531 ) {
532 /*
533 * ".." at the root returns the root
534 */
535 cache_get(&nd->nl_nch, &nch);
21739618 536 } else {
28623bf9
MD
537 /*
538 * Locate the parent ncp. If we are at the root of a
539 * filesystem mount we have to skip to the mounted-on
540 * point in the underlying filesystem.
f63911bf
MD
541 *
542 * Expect the parent to always be good since the
543 * mountpoint doesn't go away. XXX hack. cache_get()
544 * requires the ncp to already have a ref as a safety.
28623bf9 545 */
f63911bf
MD
546 nctmp = nd->nl_nch;
547 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
548 nctmp = nctmp.mount->mnt_ncmounton;
549 nctmp.ncp = nctmp.ncp->nc_parent;
550 KKASSERT(nctmp.ncp != NULL);
2247fe02
MD
551 cache_hold(&nctmp);
552 cache_get(&nctmp, &nch);
f63911bf 553 cache_drop(&nctmp); /* NOTE: zero's nctmp */
21739618 554 }
d7c75c7a 555 wasdotordotdot = 2;
21739618 556 } else {
2247fe02
MD
557 /*
558 * Must unlock nl_nch when traversing down the path.
559 */
560 cache_unlock(&nd->nl_nch);
561 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 562 nch = cache_nlookup(&nd->nl_nch, &nlc);
5a9972aa
MD
563 if (nch.ncp->nc_flag & NCF_UNRESOLVED)
564 hit = 0;
28623bf9 565 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
6ea70f76 566 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
28623bf9
MD
567 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
568 cache_put(&nch);
569 nch = cache_nlookup(&nd->nl_nch, &nlc);
8e005a45 570 }
f4d4e93a 571 wasdotordotdot = 0;
21739618 572 }
3a907475
MD
573
574 /*
575 * If the last component was "." or ".." our dflags no longer
576 * represents the parent directory and we have to explicitly
577 * look it up.
f63911bf
MD
578 *
579 * Expect the parent to be good since nch is locked.
3a907475
MD
580 */
581 if (wasdotordotdot && error == 0) {
582 dflags = 0;
583 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
584 par.mount = nch.mount;
585 cache_hold(&par);
2247fe02 586 cache_lock(&par);
aabd5ce8 587 error = naccess(&par, 0, nd->nl_cred, &dflags);
2247fe02 588 cache_put(&par);
3a907475
MD
589 }
590 }
2247fe02
MD
591 if (nd->nl_flags & NLC_NCPISLOCKED) {
592 cache_unlock(&nd->nl_nch);
593 nd->nl_flags &= ~NLC_NCPISLOCKED;
594 }
3a907475 595
524c845c 596 /*
2247fe02
MD
597 * [end of subsection]
598 *
599 * nch is locked and referenced.
600 * nd->nl_nch is unlocked and referenced.
601 *
602 * nl_nch must be unlocked or we could chain lock to the root
603 * if a resolve gets stuck (e.g. in NFS).
524c845c 604 */
21739618
MD
605
606 /*
607 * Resolve the namespace if necessary. The ncp returned by
608 * cache_nlookup() is referenced and locked.
8e005a45
MD
609 *
610 * XXX neither '.' nor '..' should return EAGAIN since they were
611 * previously resolved and thus cannot be newly created ncp's.
690a3127 612 */
28623bf9 613 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
5a9972aa 614 hit = 0;
28623bf9 615 error = cache_resolve(&nch, nd->nl_cred);
8e005a45 616 KKASSERT(error != EAGAIN);
690a3127 617 } else {
28623bf9 618 error = nch.ncp->nc_error;
690a3127 619 }
21739618
MD
620
621 /*
fad57d0e 622 * Early completion. ENOENT is not an error if this is the last
945b476a
MD
623 * component and NLC_CREATE or NLC_RENAME (rename target) was
624 * requested. Note that ncp->nc_error is left as ENOENT in that
625 * case, which we check later on.
626 *
f4d4e93a 627 * Also handle invalid '.' or '..' components terminating a path
945b476a
MD
628 * for a create/rename/delete. The standard requires this and pax
629 * pretty stupidly depends on it.
fad57d0e
MD
630 */
631 for (xptr = ptr; *xptr == '/'; ++xptr)
632 ;
f4d4e93a 633 if (*xptr == 0) {
3a907475
MD
634 if (error == ENOENT &&
635 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
636 ) {
637 if (nd->nl_flags & NLC_NFS_RDONLY) {
c9aef211 638 error = EROFS;
3a907475
MD
639 } else {
640 error = naccess(&nch, nd->nl_flags | dflags,
aabd5ce8 641 nd->nl_cred, NULL);
3a907475 642 }
c9aef211 643 }
945b476a 644 if (error == 0 && wasdotordotdot &&
3a907475
MD
645 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
646 NLC_RENAME_SRC | NLC_RENAME_DST))) {
d7c75c7a
MD
647 /*
648 * POSIX junk
649 */
650 if (nd->nl_flags & NLC_CREATE)
651 error = EEXIST;
652 else if (nd->nl_flags & NLC_DELETE)
653 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
654 else
655 error = EINVAL;
945b476a 656 }
fad57d0e
MD
657 }
658
659 /*
660 * Early completion on error.
21739618 661 */
690a3127 662 if (error) {
28623bf9 663 cache_put(&nch);
690a3127
MD
664 break;
665 }
666
667 /*
668 * If the element is a symlink and it is either not the last
669 * element or it is the last element and we are allowed to
670 * follow symlinks, resolve the symlink.
671 */
28623bf9 672 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
690a3127
MD
673 (*ptr || (nd->nl_flags & NLC_FOLLOW))
674 ) {
675 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
21739618 676 error = ELOOP;
28623bf9 677 cache_put(&nch);
690a3127
MD
678 break;
679 }
28623bf9
MD
680 error = nreadsymlink(nd, &nch, &nlc);
681 cache_put(&nch);
21739618
MD
682 if (error)
683 break;
690a3127
MD
684
685 /*
686 * Concatenate trailing path elements onto the returned symlink.
687 * Note that if the path component (ptr) is not exhausted, it
688 * will being with a '/', so we do not have to add another one.
689 *
690 * The symlink may not be empty.
691 */
692 len = strlen(ptr);
693 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
694 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
70aac194 695 objcache_put(namei_oc, nlc.nlc_nameptr);
690a3127
MD
696 break;
697 }
698 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
699 if (nd->nl_flags & NLC_HASBUF)
70aac194 700 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
701 nd->nl_path = nlc.nlc_nameptr;
702 nd->nl_flags |= NLC_HASBUF;
703 ptr = nd->nl_path;
704
705 /*
706 * Go back up to the top to resolve any initial '/'s in the
707 * symlink.
708 */
709 continue;
710 }
711
712 /*
21739618 713 * If the element is a directory and we are crossing a mount point,
28623bf9 714 * Locate the mount.
21739618 715 */
28623bf9
MD
716 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
717 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
718 (mp = cache_findmount(&nch)) != NULL
21739618 719 ) {
21739618
MD
720 struct vnode *tdp;
721
28623bf9
MD
722 cache_put(&nch);
723 cache_get(&mp->mnt_ncmountpt, &nch);
21739618 724
28623bf9 725 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 726 while (vfs_busy(mp, 0))
21739618
MD
727 ;
728 error = VFS_ROOT(mp, &tdp);
f9642f56 729 vfs_unbusy(mp);
412f07cd
MD
730 if (error) {
731 cache_dropmount(mp);
21739618 732 break;
412f07cd 733 }
28623bf9 734 cache_setvp(&nch, tdp);
21739618
MD
735 vput(tdp);
736 }
412f07cd 737 cache_dropmount(mp);
21739618
MD
738 }
739 if (error) {
28623bf9 740 cache_put(&nch);
21739618
MD
741 break;
742 }
743
744 /*
690a3127
MD
745 * Skip any slashes to get to the next element. If there
746 * are any slashes at all the current element must be a
fad57d0e
MD
747 * directory or, in the create case, intended to become a directory.
748 * If it isn't we break without incrementing ptr and fall through
749 * to the failure case below.
690a3127
MD
750 */
751 while (*ptr == '/') {
28623bf9 752 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
fad57d0e
MD
753 !(nd->nl_flags & NLC_WILLBEDIR)
754 ) {
690a3127 755 break;
fad57d0e 756 }
690a3127
MD
757 ++ptr;
758 }
759
760 /*
761 * Continuation case: additional elements and the current
762 * element is a directory.
763 */
28623bf9
MD
764 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
765 cache_drop(&nd->nl_nch);
766 cache_unlock(&nch);
2247fe02 767 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
28623bf9 768 nd->nl_nch = nch;
690a3127
MD
769 continue;
770 }
771
772 /*
773 * Failure case: additional elements and the current element
774 * is not a directory
775 */
776 if (*ptr) {
28623bf9 777 cache_put(&nch);
690a3127
MD
778 error = ENOTDIR;
779 break;
780 }
781
782 /*
fad57d0e
MD
783 * Successful lookup of last element.
784 *
3a907475
MD
785 * Check permissions if the target exists. If the target does not
786 * exist directory permissions were already tested in the early
787 * completion code above.
945b476a 788 *
3a907475
MD
789 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
790 * if the file is marked append-only, and NLC_STICKY if the directory
791 * containing the file is sticky.
fad57d0e 792 */
3a907475
MD
793 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
794 error = naccess(&nch, nd->nl_flags | dflags,
aabd5ce8 795 nd->nl_cred, NULL);
945b476a 796 if (error) {
28623bf9 797 cache_put(&nch);
fad57d0e
MD
798 break;
799 }
800 }
801
802 /*
3a907475 803 * Termination: no more elements.
5312fa43
MD
804 *
805 * If NLC_REFDVP is set acquire a referenced parent dvp.
690a3127 806 */
5312fa43 807 if (nd->nl_flags & NLC_REFDVP) {
2247fe02 808 cache_lock(&nd->nl_nch);
5312fa43 809 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
2247fe02 810 cache_unlock(&nd->nl_nch);
5312fa43
MD
811 if (error) {
812 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
813 cache_put(&nch);
814 break;
815 }
816 }
28623bf9
MD
817 cache_drop(&nd->nl_nch);
818 nd->nl_nch = nch;
690a3127
MD
819 nd->nl_flags |= NLC_NCPISLOCKED;
820 error = 0;
821 break;
822 }
3a907475 823
5a9972aa
MD
824 if (hit)
825 ++gd->gd_nchstats->ncs_longhits;
826 else
827 ++gd->gd_nchstats->ncs_longmiss;
828
3a907475
MD
829 /*
830 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
831 * (ncp->nc_error will be ENOENT), but we will still return an error
832 * code of 0.
833 */
690a3127
MD
834 return(error);
835}
836
837/*
21739618
MD
838 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
839 * of continuity in the namecache tree by connecting the ncp related to the
840 * vnode under the mount to the ncp related to the mount's root vnode.
841 *
842 * If no error occured a locked, ref'd ncp is stored in *ncpp.
843 */
844int
28623bf9 845nlookup_mp(struct mount *mp, struct nchandle *nch)
21739618 846{
21739618
MD
847 struct vnode *vp;
848 int error;
849
850 error = 0;
28623bf9
MD
851 cache_get(&mp->mnt_ncmountpt, nch);
852 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 853 while (vfs_busy(mp, 0))
21739618
MD
854 ;
855 error = VFS_ROOT(mp, &vp);
f9642f56 856 vfs_unbusy(mp);
21739618 857 if (error) {
28623bf9 858 cache_put(nch);
21739618 859 } else {
28623bf9 860 cache_setvp(nch, vp);
21739618
MD
861 vput(vp);
862 }
863 }
21739618
MD
864 return(error);
865}
866
867/*
690a3127 868 * Read the contents of a symlink, allocate a path buffer out of the
70aac194 869 * namei_oc and initialize the supplied nlcomponent with the result.
690a3127
MD
870 *
871 * If an error occurs no buffer will be allocated or returned in the nlc.
872 */
873int
28623bf9 874nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
690a3127
MD
875 struct nlcomponent *nlc)
876{
21739618 877 struct vnode *vp;
690a3127
MD
878 struct iovec aiov;
879 struct uio auio;
880 int linklen;
881 int error;
882 char *cp;
883
884 nlc->nlc_nameptr = NULL;
885 nlc->nlc_namelen = 0;
28623bf9 886 if (nch->ncp->nc_vp == NULL)
690a3127 887 return(ENOENT);
28623bf9 888 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
690a3127 889 return(error);
70aac194 890 cp = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
891 aiov.iov_base = cp;
892 aiov.iov_len = MAXPATHLEN;
893 auio.uio_iov = &aiov;
894 auio.uio_iovcnt = 1;
895 auio.uio_offset = 0;
896 auio.uio_rw = UIO_READ;
897 auio.uio_segflg = UIO_SYSSPACE;
898 auio.uio_td = nd->nl_td;
899 auio.uio_resid = MAXPATHLEN - 1;
21739618 900 error = VOP_READLINK(vp, &auio, nd->nl_cred);
690a3127
MD
901 if (error)
902 goto fail;
21739618 903 linklen = MAXPATHLEN - 1 - auio.uio_resid;
690a3127
MD
904 if (varsym_enable) {
905 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
906 if (linklen < 0) {
907 error = ENAMETOOLONG;
908 goto fail;
909 }
910 }
911 cp[linklen] = 0;
912 nlc->nlc_nameptr = cp;
913 nlc->nlc_namelen = linklen;
21739618 914 vput(vp);
690a3127
MD
915 return(0);
916fail:
70aac194 917 objcache_put(namei_oc, cp);
21739618 918 vput(vp);
690a3127
MD
919 return(error);
920}
921
21739618
MD
922/*
923 * Check access [XXX cache vattr!] [XXX quota]
924 *
3a907475
MD
925 * Generally check the NLC_* access bits. All specified bits must pass
926 * for this function to return 0.
21739618 927 *
3a907475
MD
928 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
929 * access, otherwise it must exist. No error is returned in this case.
945b476a 930 *
3a907475 931 * The file must not exist if NLC_EXCL is specified.
21739618 932 *
3a907475
MD
933 * Directory permissions in general are tested for NLC_CREATE if the file
934 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
935 * whether the file exists or not.
21739618 936 *
3a907475
MD
937 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
938 * the latter is only tested if the target exists.
dae8d54f 939 *
2247fe02 940 * The passed ncp must be referenced and locked.
21739618 941 */
5a3f9be0 942static int
aabd5ce8 943naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
21739618
MD
944{
945 struct vnode *vp;
946 struct vattr va;
aabd5ce8 947 struct namecache *ncp;
21739618 948 int error;
e9b56058 949 int cflags;
21739618 950
2247fe02 951 ASSERT_NCH_LOCKED(nch);
aabd5ce8
MD
952 ncp = nch->ncp;
953 if (ncp->nc_flag & NCF_UNRESOLVED) {
28623bf9 954 cache_resolve(nch, cred);
aabd5ce8 955 ncp = nch->ncp;
21739618 956 }
aabd5ce8 957 error = ncp->nc_error;
dae8d54f 958
945b476a 959 /*
3a907475
MD
960 * Directory permissions checks. Silently ignore ENOENT if these
961 * tests pass. It isn't an error.
2247fe02 962 *
aabd5ce8
MD
963 * We can safely resolve ncp->nc_parent because ncp is currently
964 * locked.
945b476a 965 */
3a907475 966 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
aabd5ce8
MD
967 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) ||
968 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) ||
969 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) ||
3a907475 970 (nflags & NLC_RENAME_DST)
21739618 971 ) {
f63911bf
MD
972 struct nchandle par;
973
aabd5ce8 974 if ((par.ncp = ncp->nc_parent) == NULL) {
8e005a45 975 if (error != EAGAIN)
fad57d0e 976 error = EINVAL;
3a907475 977 } else if (error == 0 || error == ENOENT) {
28623bf9 978 par.mount = nch->mount;
2247fe02
MD
979 cache_hold(&par);
980 cache_lock(&par);
aabd5ce8 981 error = naccess(&par, NLC_WRITE, cred, NULL);
2247fe02 982 cache_put(&par);
8e005a45 983 }
21739618 984 }
21739618 985 }
3a907475
MD
986
987 /*
988 * NLC_EXCL check. Target file must not exist.
989 */
aabd5ce8 990 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL)
3a907475
MD
991 error = EEXIST;
992
993 /*
994 * Get the vnode attributes so we can do the rest of our checks.
995 *
996 * NOTE: We only call naccess_va() if the target exists.
997 */
21739618 998 if (error == 0) {
28623bf9 999 error = cache_vget(nch, cred, LK_SHARED, &vp);
21739618 1000 if (error == ENOENT) {
3a907475
MD
1001 /*
1002 * Silently zero-out ENOENT if creating or renaming
1003 * (rename target). It isn't an error.
1004 */
1005 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
21739618
MD
1006 error = 0;
1007 } else if (error == 0) {
3a907475
MD
1008 /*
1009 * Get the vnode attributes and check for illegal O_TRUNC
1010 * requests and read-only mounts.
1011 *
1012 * NOTE: You can still open devices on read-only mounts for
1013 * writing.
1014 *
1015 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
1016 * check on the parent directory above.
1017 *
1018 * XXX cache the va in the namecache or in the vnode
1019 */
1020 error = VOP_GETATTR(vp, &va);
1021 if (error == 0 && (nflags & NLC_TRUNCATE)) {
1022 switch(va.va_type) {
1023 case VREG:
1024 case VDATABASE:
1025 case VCHR:
1026 case VBLK:
c9c08c19 1027 case VFIFO:
3a907475 1028 break;
d7c75c7a
MD
1029 case VDIR:
1030 error = EISDIR;
1031 break;
3a907475
MD
1032 default:
1033 error = EINVAL;
1034 break;
1035 }
1036 }
1037 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
1038 (vp->v_mount->mnt_flag & MNT_RDONLY)
1039 ) {
1040 switch(va.va_type) {
1041 case VDIR:
1042 case VLNK:
1043 case VREG:
1044 case VDATABASE:
1045 error = EROFS;
1046 break;
1047 default:
1048 break;
21739618
MD
1049 }
1050 }
1051 vput(vp);
dae8d54f 1052
3a907475
MD
1053 /*
1054 * Check permissions based on file attributes. The passed
1055 * flags (*nflagsp) are modified with feedback based on
1056 * special attributes and requirements.
1057 */
dae8d54f
MD
1058 if (error == 0) {
1059 /*
3a907475 1060 * Adjust the returned (*nflagsp) if non-NULL.
dae8d54f 1061 */
3a907475
MD
1062 if (nflagsp) {
1063 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
1064 *nflagsp |= NLC_STICKY;
1065 if (va.va_flags & APPEND)
1066 *nflagsp |= NLC_APPENDONLY;
1067 if (va.va_flags & IMMUTABLE)
1068 *nflagsp |= NLC_IMMUTABLE;
1069 }
dae8d54f
MD
1070
1071 /*
e9b56058 1072 * Track swapcache management flags in the namecache.
aabd5ce8
MD
1073 *
1074 * Calculate the flags based on the current vattr info
1075 * and recalculate the inherited flags from the parent
1076 * (the original cache linkage may have occurred without
1077 * getattrs and thus have stale flags).
e9b56058 1078 */
aabd5ce8 1079 cflags = 0;
e9b56058 1080 if (va.va_flags & SF_NOCACHE)
aabd5ce8 1081 cflags |= NCF_SF_NOCACHE;
e9b56058 1082 if (va.va_flags & UF_CACHE)
aabd5ce8
MD
1083 cflags |= NCF_UF_CACHE;
1084 if (ncp->nc_parent) {
1085 if (ncp->nc_parent->nc_flag &
1086 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) {
1087 cflags |= NCF_SF_PNOCACHE;
1088 }
1089 if (ncp->nc_parent->nc_flag &
1090 (NCF_UF_CACHE | NCF_UF_PCACHE)) {
1091 cflags |= NCF_UF_PCACHE;
1092 }
1093 }
1094 ncp->nc_flag &= ~(NCF_SF_NOCACHE | NCF_UF_CACHE |
1095 NCF_SF_PNOCACHE | NCF_UF_PCACHE);
1096 ncp->nc_flag |= cflags;
e9b56058
MD
1097
1098 /*
dae8d54f
MD
1099 * Process general access.
1100 */
3a907475 1101 error = naccess_va(&va, nflags, cred);
dae8d54f 1102 }
21739618
MD
1103 }
1104 }
1105 return(error);
1106}
1107
1108/*
1109 * Check the requested access against the given vattr using cred.
1110 */
1111int
3a907475 1112naccess_va(struct vattr *va, int nflags, struct ucred *cred)
21739618
MD
1113{
1114 int i;
3a907475 1115 int vmode;
21739618
MD
1116
1117 /*
3a907475
MD
1118 * Test the immutable bit. Creations, deletions, renames (source
1119 * or destination) are not allowed. chown/chmod/other is also not
1120 * allowed but is handled by SETATTR. Hardlinks to the immutable
1121 * file are allowed.
1122 *
1123 * If the directory is set to immutable then creations, deletions,
1124 * renames (source or dest) and hardlinks to files within the directory
1125 * are not allowed, and regular files opened through the directory may
1126 * not be written to or truncated (unless a special device).
945b476a 1127 *
3a907475
MD
1128 * NOTE! New hardlinks to immutable files work but new hardlinks to
1129 * files, immutable or not, sitting inside an immutable directory are
1130 * not allowed. As always if the file is hardlinked via some other
1131 * path additional hardlinks may be possible even if the file is marked
1132 * immutable. The sysop needs to create a closure by checking the hard
1133 * link count. Once closure is achieved you are good, and security
1134 * scripts should check link counts anyway.
1135 *
1136 * Writes and truncations are only allowed on special devices.
21739618 1137 */
3a907475
MD
1138 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1139 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1140 return (EPERM);
1141 if (nflags & (NLC_CREATE | NLC_DELETE |
1142 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1143 return (EPERM);
1144 }
1145 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1146 switch(va->va_type) {
1147 case VDIR:
d7c75c7a 1148 return (EISDIR);
3a907475
MD
1149 case VLNK:
1150 case VREG:
1151 case VDATABASE:
1152 return (EPERM);
1153 default:
1154 break;
1155 }
1156 }
1157 }
1158
1159 /*
1160 * Test the no-unlink and append-only bits for opens, rename targets,
1161 * and deletions. These bits are not tested for creations or
1162 * rename sources.
1163 *
1164 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1165 * If you do not wish this you must also set NOUNLINK.
1166 *
1167 * If the governing directory is marked APPEND-only it implies
1168 * NOUNLINK for all entries in the directory.
1169 */
1170 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1171 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1172 ) {
1173 return (EPERM);
1174 }
1175
1176 /*
1177 * A file marked append-only may not be deleted but can be renamed.
1178 */
1179 if ((va->va_flags & APPEND) &&
1180 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1181 ) {
1182 return (EPERM);
1183 }
1184
1185 /*
1186 * A file marked append-only which is opened for writing must also
1187 * be opened O_APPEND.
1188 */
1189 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1190 if (nflags & NLC_TRUNCATE)
1191 return (EPERM);
1192 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1193 if ((nflags & NLC_APPEND) == 0)
21739618
MD
1194 return (EPERM);
1195 }
1196 }
1197
1198 /*
1199 * root gets universal access
1200 */
1201 if (cred->cr_uid == 0)
1202 return(0);
1203
1204 /*
dae8d54f
MD
1205 * Check owner perms.
1206 *
3a907475 1207 * If NLC_OWN is set the owner of the file is allowed no matter when
dae8d54f 1208 * the owner-mode bits say (utimes).
21739618 1209 */
3a907475
MD
1210 vmode = 0;
1211 if (nflags & NLC_READ)
1212 vmode |= S_IRUSR;
1213 if (nflags & NLC_WRITE)
1214 vmode |= S_IWUSR;
1215 if (nflags & NLC_EXEC)
1216 vmode |= S_IXUSR;
1217
21739618 1218 if (cred->cr_uid == va->va_uid) {
3a907475 1219 if ((nflags & NLC_OWN) == 0) {
ee89633d
MD
1220 if ((vmode & va->va_mode) != vmode)
1221 return(EACCES);
1222 }
21739618
MD
1223 return(0);
1224 }
dae8d54f
MD
1225
1226 /*
3a907475
MD
1227 * If NLC_STICKY is set only the owner may delete or rename a file.
1228 * This bit is typically set on /tmp.
dae8d54f 1229 *
3a907475
MD
1230 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1231 * the specific delete or rename case. For deletions and renames we
1232 * usually just care about directory permissions, not file permissions.
dae8d54f 1233 */
3a907475
MD
1234 if ((nflags & NLC_STICKY) &&
1235 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
dae8d54f 1236 return(EACCES);
3a907475 1237 }
dae8d54f
MD
1238
1239 /*
1240 * Check group perms
1241 */
21739618
MD
1242 vmode >>= 3;
1243 for (i = 0; i < cred->cr_ngroups; ++i) {
1244 if (va->va_gid == cred->cr_groups[i]) {
1245 if ((vmode & va->va_mode) != vmode)
1246 return(EACCES);
1247 return(0);
1248 }
1249 }
1250
dae8d54f
MD
1251 /*
1252 * Check world perms
1253 */
21739618
MD
1254 vmode >>= 3;
1255 if ((vmode & va->va_mode) != vmode)
1256 return(EACCES);
1257 return(0);
1258}
1259