kernel - misc MPSAFe work
[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>
1db695af 68#include <sys/file.h>
690a3127
MD
69
70#ifdef KTRACE
71#include <sys/ktrace.h>
72#endif
73
74/*
75 * Initialize a nlookup() structure, early error return for copyin faults
76 * or a degenerate empty string (which is not allowed).
1f95166e
HP
77 *
78 * The first process proc0's credentials are used if the calling thread
79 * is not associated with a process context.
61f96b6f
MD
80 *
81 * MPSAFE
690a3127
MD
82 */
83int
21739618
MD
84nlookup_init(struct nlookupdata *nd,
85 const char *path, enum uio_seg seg, int flags)
690a3127
MD
86{
87 size_t pathlen;
88 struct proc *p;
89 thread_t td;
90 int error;
91
92 td = curthread;
93 p = td->td_proc;
94
95 /*
96 * note: the pathlen set by copy*str() includes the terminating \0.
97 */
98 bzero(nd, sizeof(struct nlookupdata));
70aac194 99 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
100 nd->nl_flags |= NLC_HASBUF;
101 if (seg == UIO_SYSSPACE)
102 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
103 else
104 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
105
106 /*
107 * Don't allow empty pathnames.
108 * POSIX.1 requirement: "" is not a vaild file name.
109 */
110 if (error == 0 && pathlen <= 1)
111 error = ENOENT;
112
113 if (error == 0) {
21739618 114 if (p && p->p_fd) {
28623bf9
MD
115 cache_copy(&p->p_fd->fd_ncdir, &nd->nl_nch);
116 cache_copy(&p->p_fd->fd_nrdir, &nd->nl_rootnch);
117 if (p->p_fd->fd_njdir.ncp)
118 cache_copy(&p->p_fd->fd_njdir, &nd->nl_jailnch);
21739618 119 nd->nl_cred = crhold(p->p_ucred);
690a3127 120 } else {
28623bf9
MD
121 cache_copy(&rootnch, &nd->nl_nch);
122 cache_copy(&nd->nl_nch, &nd->nl_rootnch);
123 cache_copy(&nd->nl_nch, &nd->nl_jailnch);
21739618 124 nd->nl_cred = crhold(proc0.p_ucred);
690a3127
MD
125 }
126 nd->nl_td = td;
690a3127
MD
127 nd->nl_flags |= flags;
128 } else {
129 nlookup_done(nd);
130 }
131 return(error);
132}
133
1db695af
NT
134
135/*
136 * nlookup_init() for "at" family of syscalls.
137 *
138 * Works similarly to nlookup_init() but if path is relative and fd is not
139 * AT_FDCWD, path is interpreted relative to the directory pointed to by fd.
140 * In this case, the file entry pointed to by fd is ref'ed and returned in
141 * *fpp.
142 *
143 * If the call succeeds, nlookup_done_at() must be called to clean-up the nd
144 * and release the ref to the file entry.
145 */
146int
147nlookup_init_at(struct nlookupdata *nd, struct file **fpp, int fd,
148 const char *path, enum uio_seg seg, int flags)
149{
150 struct thread *td = curthread;
151 struct proc *p = td->td_proc;
152 struct file* fp;
153 struct vnode *vp;
154 int error;
155
156 *fpp = NULL;
157
158 if ((error = nlookup_init(nd, path, seg, flags)) != 0) {
159 return (error);
160 }
161
162 if (nd->nl_path[0] != '/' && fd != AT_FDCWD) {
163 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
164 goto done;
165 vp = (struct vnode*)fp->f_data;
166 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL) {
167 fdrop(fp);
168 fp = NULL;
169 error = ENOTDIR;
170 goto done;
171 }
172 cache_drop(&nd->nl_nch);
173 cache_copy(&fp->f_nchandle, &nd->nl_nch);
174 *fpp = fp;
175 }
176
177
178done:
179 if (error)
180 nlookup_done(nd);
181 return (error);
182
183}
184
690a3127 185/*
fad57d0e 186 * This works similarly to nlookup_init() but does not assume a process
28623bf9 187 * context. rootnch is always chosen for the root directory and the cred
fad57d0e
MD
188 * and starting directory are supplied in arguments.
189 */
190int
191nlookup_init_raw(struct nlookupdata *nd,
192 const char *path, enum uio_seg seg, int flags,
28623bf9 193 struct ucred *cred, struct nchandle *ncstart)
fad57d0e
MD
194{
195 size_t pathlen;
196 thread_t td;
197 int error;
198
199 td = curthread;
200
201 bzero(nd, sizeof(struct nlookupdata));
70aac194 202 nd->nl_path = objcache_get(namei_oc, M_WAITOK);
fad57d0e
MD
203 nd->nl_flags |= NLC_HASBUF;
204 if (seg == UIO_SYSSPACE)
205 error = copystr(path, nd->nl_path, MAXPATHLEN, &pathlen);
206 else
207 error = copyinstr(path, nd->nl_path, MAXPATHLEN, &pathlen);
208
209 /*
210 * Don't allow empty pathnames.
211 * POSIX.1 requirement: "" is not a vaild file name.
212 */
213 if (error == 0 && pathlen <= 1)
214 error = ENOENT;
215
216 if (error == 0) {
28623bf9
MD
217 cache_copy(ncstart, &nd->nl_nch);
218 cache_copy(&rootnch, &nd->nl_rootnch);
219 cache_copy(&rootnch, &nd->nl_jailnch);
fad57d0e
MD
220 nd->nl_cred = crhold(cred);
221 nd->nl_td = td;
222 nd->nl_flags |= flags;
223 } else {
224 nlookup_done(nd);
225 }
226 return(error);
227}
228
229/*
1f95166e
HP
230 * Set a different credential; this credential will be used by future
231 * operations performed on nd.nl_open_vp and nlookupdata structure.
232 */
233void
234nlookup_set_cred(struct nlookupdata *nd, struct ucred *cred)
235{
236 KKASSERT(nd->nl_cred != NULL);
237
238 if (nd->nl_cred != cred) {
239 cred = crhold(cred);
240 crfree(nd->nl_cred);
241 nd->nl_cred = cred;
242 }
243}
244
245/*
21739618
MD
246 * Cleanup a nlookupdata structure after we are through with it. This may
247 * be called on any nlookupdata structure initialized with nlookup_init().
248 * Calling nlookup_done() is mandatory in all cases except where nlookup_init()
249 * returns an error, even if as a consumer you believe you have taken all
250 * dynamic elements out of the nlookupdata structure.
690a3127
MD
251 */
252void
253nlookup_done(struct nlookupdata *nd)
254{
28623bf9 255 if (nd->nl_nch.ncp) {
21739618
MD
256 if (nd->nl_flags & NLC_NCPISLOCKED) {
257 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 258 cache_unlock(&nd->nl_nch);
21739618 259 }
28623bf9 260 cache_drop(&nd->nl_nch);
690a3127 261 }
28623bf9
MD
262 if (nd->nl_rootnch.ncp)
263 cache_drop(&nd->nl_rootnch);
264 if (nd->nl_jailnch.ncp)
265 cache_drop(&nd->nl_jailnch);
690a3127 266 if ((nd->nl_flags & NLC_HASBUF) && nd->nl_path) {
70aac194 267 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
268 nd->nl_path = NULL;
269 }
270 if (nd->nl_cred) {
271 crfree(nd->nl_cred);
272 nd->nl_cred = NULL;
273 }
fad57d0e
MD
274 if (nd->nl_open_vp) {
275 if (nd->nl_flags & NLC_LOCKVP) {
a11aaa81 276 vn_unlock(nd->nl_open_vp);
fad57d0e
MD
277 nd->nl_flags &= ~NLC_LOCKVP;
278 }
87de5057 279 vn_close(nd->nl_open_vp, nd->nl_vp_fmode);
fad57d0e
MD
280 nd->nl_open_vp = NULL;
281 }
5312fa43
MD
282 if (nd->nl_dvp) {
283 vrele(nd->nl_dvp);
284 nd->nl_dvp = NULL;
285 }
fad57d0e
MD
286 nd->nl_flags = 0; /* clear remaining flags (just clear everything) */
287}
288
1db695af
NT
289/*
290 * Works similarly to nlookup_done() when nd initialized with
291 * nlookup_init_at().
292 */
293void
294nlookup_done_at(struct nlookupdata *nd, struct file *fp)
295{
296 nlookup_done(nd);
297 if (fp != NULL)
298 fdrop(fp);
299}
300
fad57d0e
MD
301void
302nlookup_zero(struct nlookupdata *nd)
303{
304 bzero(nd, sizeof(struct nlookupdata));
690a3127
MD
305}
306
307/*
21739618
MD
308 * Simple all-in-one nlookup. Returns a locked namecache structure or NULL
309 * if an error occured.
310 *
311 * Note that the returned ncp is not checked for permissions, though VEXEC
312 * is checked on the directory path leading up to the result. The caller
313 * must call naccess() to check the permissions of the returned leaf.
690a3127 314 */
28623bf9 315struct nchandle
21739618
MD
316nlookup_simple(const char *str, enum uio_seg seg,
317 int niflags, int *error)
690a3127
MD
318{
319 struct nlookupdata nd;
28623bf9 320 struct nchandle nch;
690a3127
MD
321
322 *error = nlookup_init(&nd, str, seg, niflags);
323 if (*error == 0) {
21739618 324 if ((*error = nlookup(&nd)) == 0) {
28623bf9
MD
325 nch = nd.nl_nch; /* keep hold ref from structure */
326 cache_zero(&nd.nl_nch); /* and NULL out */
21739618 327 } else {
28623bf9 328 cache_zero(&nch);
21739618 329 }
690a3127
MD
330 nlookup_done(&nd);
331 } else {
28623bf9 332 cache_zero(&nch);
690a3127 333 }
28623bf9 334 return(nch);
690a3127
MD
335}
336
337/*
338 * Do a generic nlookup. Note that the passed nd is not nlookup_done()'d
339 * on return, even if an error occurs. If no error occurs the returned
28623bf9 340 * nl_nch is always referenced and locked, otherwise it may or may not be.
21739618
MD
341 *
342 * Intermediate directory elements, including the current directory, require
343 * execute (search) permission. nlookup does not examine the access
344 * permissions on the returned element.
fad57d0e 345 *
945b476a
MD
346 * If NLC_CREATE is set the last directory must allow node creation,
347 * and an error code of 0 will be returned for a non-existant
348 * target (not ENOENT).
349 *
3a907475 350 * If NLC_RENAME_DST is set the last directory mut allow node deletion,
945b476a 351 * plus the sticky check is made, and an error code of 0 will be returned
3a907475 352 * for a non-existant target (not ENOENT).
945b476a
MD
353 *
354 * If NLC_DELETE is set the last directory mut allow node deletion,
355 * plus the sticky check is made.
5312fa43
MD
356 *
357 * If NLC_REFDVP is set nd->nl_dvp will be set to the directory vnode
358 * of the returned entry. The vnode will be referenced, but not locked,
359 * and will be released by nlookup_done() along with everything else.
690a3127
MD
360 */
361int
362nlookup(struct nlookupdata *nd)
363{
364 struct nlcomponent nlc;
28623bf9 365 struct nchandle nch;
3a907475 366 struct nchandle par;
28623bf9 367 struct mount *mp;
f4d4e93a 368 int wasdotordotdot;
690a3127 369 char *ptr;
fad57d0e 370 char *xptr;
690a3127
MD
371 int error;
372 int len;
3a907475 373 int dflags;
690a3127
MD
374
375#ifdef KTRACE
376 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
9fb04d14 377 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
690a3127
MD
378#endif
379 bzero(&nlc, sizeof(nlc));
380
381 /*
382 * Setup for the loop. The current working namecache element must
524c845c
MD
383 * be in a refd + unlocked state. This typically the case on entry except
384 * when stringing nlookup()'s along in a chain, since nlookup() always
28623bf9 385 * returns nl_nch in a locked state.
690a3127
MD
386 */
387 nd->nl_loopcnt = 0;
388 if (nd->nl_flags & NLC_NCPISLOCKED) {
389 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 390 cache_unlock(&nd->nl_nch);
690a3127 391 }
5312fa43
MD
392 if (nd->nl_dvp ) {
393 vrele(nd->nl_dvp);
394 nd->nl_dvp = NULL;
395 }
690a3127
MD
396 ptr = nd->nl_path;
397
398 /*
28623bf9 399 * Loop on the path components. At the top of the loop nd->nl_nch
524c845c 400 * is ref'd and unlocked and represents our current position.
690a3127
MD
401 */
402 for (;;) {
403 /*
404 * Check if the root directory should replace the current
405 * directory. This is done at the start of a translation
406 * or after a symbolic link has been found. In other cases
407 * ptr will never be pointing at a '/'.
408 */
409 if (*ptr == '/') {
410 do {
411 ++ptr;
412 } while (*ptr == '/');
28623bf9
MD
413 cache_copy(&nd->nl_rootnch, &nch);
414 cache_drop(&nd->nl_nch);
415 nd->nl_nch = nch;
5312fa43
MD
416
417 /*
418 * Fast-track termination. There is no parent directory of
419 * the root in the same mount from the point of view of
420 * the caller so return EPERM if NLC_REFDVP is specified.
421 * e.g. 'rmdir /' is not allowed.
422 */
21739618 423 if (*ptr == 0) {
5312fa43
MD
424 if (nd->nl_flags & NLC_REFDVP) {
425 error = EPERM;
426 } else {
427 cache_lock(&nd->nl_nch);
428 nd->nl_flags |= NLC_NCPISLOCKED;
429 error = 0;
430 }
21739618
MD
431 break;
432 }
690a3127
MD
433 continue;
434 }
435
436 /*
28623bf9 437 * Check directory search permissions.
21739618 438 */
3a907475
MD
439 dflags = 0;
440 if ((error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags)) != 0)
21739618
MD
441 break;
442
443 /*
d7c75c7a
MD
444 * Extract the path component. Path components are limited to
445 * 255 characters.
690a3127
MD
446 */
447 nlc.nlc_nameptr = ptr;
448 while (*ptr && *ptr != '/')
449 ++ptr;
450 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
d7c75c7a
MD
451 if (nlc.nlc_namelen >= 256) {
452 error = ENAMETOOLONG;
453 break;
454 }
690a3127
MD
455
456 /*
21739618
MD
457 * Lookup the path component in the cache, creating an unresolved
458 * entry if necessary. We have to handle "." and ".." as special
459 * cases.
460 *
461 * When handling ".." we have to detect a traversal back through a
28623bf9 462 * mount point. If we are at the root, ".." just returns the root.
524c845c 463 *
3a907475
MD
464 * When handling "." or ".." we also have to recalculate dflags
465 * since our dflags will be for some sub-directory instead of the
466 * parent dir.
467 *
28623bf9 468 * This subsection returns a locked, refd 'nch' unless it errors out.
fad57d0e
MD
469 * The namecache topology is not allowed to be disconnected, so
470 * encountering a NULL parent will generate EINVAL. This typically
471 * occurs when a directory is removed out from under a process.
21739618
MD
472 */
473 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
28623bf9 474 cache_get(&nd->nl_nch, &nch);
f4d4e93a 475 wasdotordotdot = 1;
21739618
MD
476 } else if (nlc.nlc_namelen == 2 &&
477 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
28623bf9
MD
478 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
479 nd->nl_nch.ncp == nd->nl_rootnch.ncp
480 ) {
481 /*
482 * ".." at the root returns the root
483 */
484 cache_get(&nd->nl_nch, &nch);
21739618 485 } else {
28623bf9
MD
486 /*
487 * Locate the parent ncp. If we are at the root of a
488 * filesystem mount we have to skip to the mounted-on
489 * point in the underlying filesystem.
490 */
491 nch = nd->nl_nch;
492 while (nch.ncp == nch.mount->mnt_ncmountpt.ncp)
e5cde29c 493 nch = nch.mount->mnt_ncmounton;
28623bf9
MD
494 nch.ncp = nch.ncp->nc_parent;
495 KKASSERT(nch.ncp != NULL);
496 cache_get(&nch, &nch);
21739618 497 }
d7c75c7a 498 wasdotordotdot = 2;
21739618 499 } else {
28623bf9
MD
500 nch = cache_nlookup(&nd->nl_nch, &nlc);
501 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
6ea70f76 502 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
28623bf9
MD
503 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
504 cache_put(&nch);
505 nch = cache_nlookup(&nd->nl_nch, &nlc);
8e005a45 506 }
f4d4e93a 507 wasdotordotdot = 0;
21739618 508 }
3a907475
MD
509
510 /*
511 * If the last component was "." or ".." our dflags no longer
512 * represents the parent directory and we have to explicitly
513 * look it up.
514 */
515 if (wasdotordotdot && error == 0) {
516 dflags = 0;
517 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
518 par.mount = nch.mount;
519 cache_hold(&par);
520 dflags = 0;
521 error = naccess(&par, 0, nd->nl_cred, &dflags);
522 cache_drop(&par);
523 }
524 }
525
524c845c 526 /*
28623bf9 527 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
524c845c 528 */
21739618
MD
529
530 /*
531 * Resolve the namespace if necessary. The ncp returned by
532 * cache_nlookup() is referenced and locked.
8e005a45
MD
533 *
534 * XXX neither '.' nor '..' should return EAGAIN since they were
535 * previously resolved and thus cannot be newly created ncp's.
690a3127 536 */
28623bf9
MD
537 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
538 error = cache_resolve(&nch, nd->nl_cred);
8e005a45 539 KKASSERT(error != EAGAIN);
690a3127 540 } else {
28623bf9 541 error = nch.ncp->nc_error;
690a3127 542 }
21739618
MD
543
544 /*
fad57d0e 545 * Early completion. ENOENT is not an error if this is the last
945b476a
MD
546 * component and NLC_CREATE or NLC_RENAME (rename target) was
547 * requested. Note that ncp->nc_error is left as ENOENT in that
548 * case, which we check later on.
549 *
f4d4e93a 550 * Also handle invalid '.' or '..' components terminating a path
945b476a
MD
551 * for a create/rename/delete. The standard requires this and pax
552 * pretty stupidly depends on it.
fad57d0e
MD
553 */
554 for (xptr = ptr; *xptr == '/'; ++xptr)
555 ;
f4d4e93a 556 if (*xptr == 0) {
3a907475
MD
557 if (error == ENOENT &&
558 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
559 ) {
560 if (nd->nl_flags & NLC_NFS_RDONLY) {
c9aef211 561 error = EROFS;
3a907475
MD
562 } else {
563 error = naccess(&nch, nd->nl_flags | dflags,
564 nd->nl_cred, NULL);
565 }
c9aef211 566 }
945b476a 567 if (error == 0 && wasdotordotdot &&
3a907475
MD
568 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
569 NLC_RENAME_SRC | NLC_RENAME_DST))) {
d7c75c7a
MD
570 /*
571 * POSIX junk
572 */
573 if (nd->nl_flags & NLC_CREATE)
574 error = EEXIST;
575 else if (nd->nl_flags & NLC_DELETE)
576 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
577 else
578 error = EINVAL;
945b476a 579 }
fad57d0e
MD
580 }
581
582 /*
583 * Early completion on error.
21739618 584 */
690a3127 585 if (error) {
28623bf9 586 cache_put(&nch);
690a3127
MD
587 break;
588 }
589
590 /*
591 * If the element is a symlink and it is either not the last
592 * element or it is the last element and we are allowed to
593 * follow symlinks, resolve the symlink.
594 */
28623bf9 595 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
690a3127
MD
596 (*ptr || (nd->nl_flags & NLC_FOLLOW))
597 ) {
598 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
21739618 599 error = ELOOP;
28623bf9 600 cache_put(&nch);
690a3127
MD
601 break;
602 }
28623bf9
MD
603 error = nreadsymlink(nd, &nch, &nlc);
604 cache_put(&nch);
21739618
MD
605 if (error)
606 break;
690a3127
MD
607
608 /*
609 * Concatenate trailing path elements onto the returned symlink.
610 * Note that if the path component (ptr) is not exhausted, it
611 * will being with a '/', so we do not have to add another one.
612 *
613 * The symlink may not be empty.
614 */
615 len = strlen(ptr);
616 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
617 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
70aac194 618 objcache_put(namei_oc, nlc.nlc_nameptr);
690a3127
MD
619 break;
620 }
621 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
622 if (nd->nl_flags & NLC_HASBUF)
70aac194 623 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
624 nd->nl_path = nlc.nlc_nameptr;
625 nd->nl_flags |= NLC_HASBUF;
626 ptr = nd->nl_path;
627
628 /*
629 * Go back up to the top to resolve any initial '/'s in the
630 * symlink.
631 */
632 continue;
633 }
634
635 /*
21739618 636 * If the element is a directory and we are crossing a mount point,
28623bf9 637 * Locate the mount.
21739618 638 */
28623bf9
MD
639 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
640 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
641 (mp = cache_findmount(&nch)) != NULL
21739618 642 ) {
21739618
MD
643 struct vnode *tdp;
644
28623bf9
MD
645 cache_put(&nch);
646 cache_get(&mp->mnt_ncmountpt, &nch);
21739618 647
28623bf9 648 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 649 while (vfs_busy(mp, 0))
21739618
MD
650 ;
651 error = VFS_ROOT(mp, &tdp);
f9642f56 652 vfs_unbusy(mp);
524c845c 653 if (error)
21739618 654 break;
28623bf9 655 cache_setvp(&nch, tdp);
21739618
MD
656 vput(tdp);
657 }
658 }
659 if (error) {
28623bf9 660 cache_put(&nch);
21739618
MD
661 break;
662 }
663
664 /*
690a3127
MD
665 * Skip any slashes to get to the next element. If there
666 * are any slashes at all the current element must be a
fad57d0e
MD
667 * directory or, in the create case, intended to become a directory.
668 * If it isn't we break without incrementing ptr and fall through
669 * to the failure case below.
690a3127
MD
670 */
671 while (*ptr == '/') {
28623bf9 672 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
fad57d0e
MD
673 !(nd->nl_flags & NLC_WILLBEDIR)
674 ) {
690a3127 675 break;
fad57d0e 676 }
690a3127
MD
677 ++ptr;
678 }
679
680 /*
681 * Continuation case: additional elements and the current
682 * element is a directory.
683 */
28623bf9
MD
684 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
685 cache_drop(&nd->nl_nch);
686 cache_unlock(&nch);
687 nd->nl_nch = nch;
690a3127
MD
688 continue;
689 }
690
691 /*
692 * Failure case: additional elements and the current element
693 * is not a directory
694 */
695 if (*ptr) {
28623bf9 696 cache_put(&nch);
690a3127
MD
697 error = ENOTDIR;
698 break;
699 }
700
701 /*
fad57d0e
MD
702 * Successful lookup of last element.
703 *
3a907475
MD
704 * Check permissions if the target exists. If the target does not
705 * exist directory permissions were already tested in the early
706 * completion code above.
945b476a 707 *
3a907475
MD
708 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
709 * if the file is marked append-only, and NLC_STICKY if the directory
710 * containing the file is sticky.
fad57d0e 711 */
3a907475
MD
712 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
713 error = naccess(&nch, nd->nl_flags | dflags,
714 nd->nl_cred, NULL);
945b476a 715 if (error) {
28623bf9 716 cache_put(&nch);
fad57d0e
MD
717 break;
718 }
719 }
720
721 /*
3a907475 722 * Termination: no more elements.
5312fa43
MD
723 *
724 * If NLC_REFDVP is set acquire a referenced parent dvp.
690a3127 725 */
5312fa43
MD
726 if (nd->nl_flags & NLC_REFDVP) {
727 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
728 if (error) {
729 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
730 cache_put(&nch);
731 break;
732 }
733 }
28623bf9
MD
734 cache_drop(&nd->nl_nch);
735 nd->nl_nch = nch;
690a3127
MD
736 nd->nl_flags |= NLC_NCPISLOCKED;
737 error = 0;
738 break;
739 }
3a907475
MD
740
741 /*
742 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
743 * (ncp->nc_error will be ENOENT), but we will still return an error
744 * code of 0.
745 */
690a3127
MD
746 return(error);
747}
748
749/*
21739618
MD
750 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
751 * of continuity in the namecache tree by connecting the ncp related to the
752 * vnode under the mount to the ncp related to the mount's root vnode.
753 *
754 * If no error occured a locked, ref'd ncp is stored in *ncpp.
755 */
756int
28623bf9 757nlookup_mp(struct mount *mp, struct nchandle *nch)
21739618 758{
21739618
MD
759 struct vnode *vp;
760 int error;
761
762 error = 0;
28623bf9
MD
763 cache_get(&mp->mnt_ncmountpt, nch);
764 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 765 while (vfs_busy(mp, 0))
21739618
MD
766 ;
767 error = VFS_ROOT(mp, &vp);
f9642f56 768 vfs_unbusy(mp);
21739618 769 if (error) {
28623bf9 770 cache_put(nch);
21739618 771 } else {
28623bf9 772 cache_setvp(nch, vp);
21739618
MD
773 vput(vp);
774 }
775 }
21739618
MD
776 return(error);
777}
778
779/*
690a3127 780 * Read the contents of a symlink, allocate a path buffer out of the
70aac194 781 * namei_oc and initialize the supplied nlcomponent with the result.
690a3127
MD
782 *
783 * If an error occurs no buffer will be allocated or returned in the nlc.
784 */
785int
28623bf9 786nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
690a3127
MD
787 struct nlcomponent *nlc)
788{
21739618 789 struct vnode *vp;
690a3127
MD
790 struct iovec aiov;
791 struct uio auio;
792 int linklen;
793 int error;
794 char *cp;
795
796 nlc->nlc_nameptr = NULL;
797 nlc->nlc_namelen = 0;
28623bf9 798 if (nch->ncp->nc_vp == NULL)
690a3127 799 return(ENOENT);
28623bf9 800 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
690a3127 801 return(error);
70aac194 802 cp = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
803 aiov.iov_base = cp;
804 aiov.iov_len = MAXPATHLEN;
805 auio.uio_iov = &aiov;
806 auio.uio_iovcnt = 1;
807 auio.uio_offset = 0;
808 auio.uio_rw = UIO_READ;
809 auio.uio_segflg = UIO_SYSSPACE;
810 auio.uio_td = nd->nl_td;
811 auio.uio_resid = MAXPATHLEN - 1;
21739618 812 error = VOP_READLINK(vp, &auio, nd->nl_cred);
690a3127
MD
813 if (error)
814 goto fail;
21739618 815 linklen = MAXPATHLEN - 1 - auio.uio_resid;
690a3127
MD
816 if (varsym_enable) {
817 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
818 if (linklen < 0) {
819 error = ENAMETOOLONG;
820 goto fail;
821 }
822 }
823 cp[linklen] = 0;
824 nlc->nlc_nameptr = cp;
825 nlc->nlc_namelen = linklen;
21739618 826 vput(vp);
690a3127
MD
827 return(0);
828fail:
70aac194 829 objcache_put(namei_oc, cp);
21739618 830 vput(vp);
690a3127
MD
831 return(error);
832}
833
21739618
MD
834/*
835 * Check access [XXX cache vattr!] [XXX quota]
836 *
3a907475
MD
837 * Generally check the NLC_* access bits. All specified bits must pass
838 * for this function to return 0.
21739618 839 *
3a907475
MD
840 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
841 * access, otherwise it must exist. No error is returned in this case.
945b476a 842 *
3a907475 843 * The file must not exist if NLC_EXCL is specified.
21739618 844 *
3a907475
MD
845 * Directory permissions in general are tested for NLC_CREATE if the file
846 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
847 * whether the file exists or not.
21739618 848 *
3a907475
MD
849 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
850 * the latter is only tested if the target exists.
dae8d54f 851 *
21739618 852 * The passed ncp may or may not be locked. The caller should use a
3a907475
MD
853 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
854 * NLC_DELETE, and NLC_EXCL checks.
21739618
MD
855 */
856int
3a907475 857naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
21739618 858{
28623bf9 859 struct nchandle par;
21739618
MD
860 struct vnode *vp;
861 struct vattr va;
862 int error;
dae8d54f 863 int sticky;
21739618 864
28623bf9
MD
865 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
866 cache_lock(nch);
867 cache_resolve(nch, cred);
868 cache_unlock(nch);
21739618 869 }
28623bf9 870 error = nch->ncp->nc_error;
dae8d54f 871
945b476a 872 /*
3a907475
MD
873 * Directory permissions checks. Silently ignore ENOENT if these
874 * tests pass. It isn't an error.
945b476a 875 */
3a907475
MD
876 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
877 if (((nflags & NLC_CREATE) && nch->ncp->nc_vp == NULL) ||
878 ((nflags & NLC_DELETE) && nch->ncp->nc_vp != NULL) ||
879 ((nflags & NLC_RENAME_SRC) && nch->ncp->nc_vp != NULL) ||
880 (nflags & NLC_RENAME_DST)
21739618 881 ) {
28623bf9 882 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
8e005a45 883 if (error != EAGAIN)
fad57d0e 884 error = EINVAL;
3a907475 885 } else if (error == 0 || error == ENOENT) {
28623bf9
MD
886 par.mount = nch->mount;
887 cache_hold(&par);
3a907475
MD
888 sticky = 0;
889 error = naccess(&par, NLC_WRITE, cred, NULL);
28623bf9 890 cache_drop(&par);
8e005a45 891 }
21739618 892 }
21739618 893 }
3a907475
MD
894
895 /*
896 * NLC_EXCL check. Target file must not exist.
897 */
898 if (error == 0 && (nflags & NLC_EXCL) && nch->ncp->nc_vp != NULL)
899 error = EEXIST;
900
901 /*
902 * Get the vnode attributes so we can do the rest of our checks.
903 *
904 * NOTE: We only call naccess_va() if the target exists.
905 */
21739618 906 if (error == 0) {
28623bf9 907 error = cache_vget(nch, cred, LK_SHARED, &vp);
21739618 908 if (error == ENOENT) {
3a907475
MD
909 /*
910 * Silently zero-out ENOENT if creating or renaming
911 * (rename target). It isn't an error.
912 */
913 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
21739618
MD
914 error = 0;
915 } else if (error == 0) {
3a907475
MD
916 /*
917 * Get the vnode attributes and check for illegal O_TRUNC
918 * requests and read-only mounts.
919 *
920 * NOTE: You can still open devices on read-only mounts for
921 * writing.
922 *
923 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
924 * check on the parent directory above.
925 *
926 * XXX cache the va in the namecache or in the vnode
927 */
928 error = VOP_GETATTR(vp, &va);
929 if (error == 0 && (nflags & NLC_TRUNCATE)) {
930 switch(va.va_type) {
931 case VREG:
932 case VDATABASE:
933 case VCHR:
934 case VBLK:
c9c08c19 935 case VFIFO:
3a907475 936 break;
d7c75c7a
MD
937 case VDIR:
938 error = EISDIR;
939 break;
3a907475
MD
940 default:
941 error = EINVAL;
942 break;
943 }
944 }
945 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
946 (vp->v_mount->mnt_flag & MNT_RDONLY)
947 ) {
948 switch(va.va_type) {
949 case VDIR:
950 case VLNK:
951 case VREG:
952 case VDATABASE:
953 error = EROFS;
954 break;
955 default:
956 break;
21739618
MD
957 }
958 }
959 vput(vp);
dae8d54f 960
3a907475
MD
961 /*
962 * Check permissions based on file attributes. The passed
963 * flags (*nflagsp) are modified with feedback based on
964 * special attributes and requirements.
965 */
dae8d54f
MD
966 if (error == 0) {
967 /*
3a907475 968 * Adjust the returned (*nflagsp) if non-NULL.
dae8d54f 969 */
3a907475
MD
970 if (nflagsp) {
971 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
972 *nflagsp |= NLC_STICKY;
973 if (va.va_flags & APPEND)
974 *nflagsp |= NLC_APPENDONLY;
975 if (va.va_flags & IMMUTABLE)
976 *nflagsp |= NLC_IMMUTABLE;
977 }
dae8d54f
MD
978
979 /*
980 * Process general access.
981 */
3a907475 982 error = naccess_va(&va, nflags, cred);
dae8d54f 983 }
21739618
MD
984 }
985 }
986 return(error);
987}
988
989/*
990 * Check the requested access against the given vattr using cred.
991 */
992int
3a907475 993naccess_va(struct vattr *va, int nflags, struct ucred *cred)
21739618
MD
994{
995 int i;
3a907475 996 int vmode;
21739618
MD
997
998 /*
3a907475
MD
999 * Test the immutable bit. Creations, deletions, renames (source
1000 * or destination) are not allowed. chown/chmod/other is also not
1001 * allowed but is handled by SETATTR. Hardlinks to the immutable
1002 * file are allowed.
1003 *
1004 * If the directory is set to immutable then creations, deletions,
1005 * renames (source or dest) and hardlinks to files within the directory
1006 * are not allowed, and regular files opened through the directory may
1007 * not be written to or truncated (unless a special device).
945b476a 1008 *
3a907475
MD
1009 * NOTE! New hardlinks to immutable files work but new hardlinks to
1010 * files, immutable or not, sitting inside an immutable directory are
1011 * not allowed. As always if the file is hardlinked via some other
1012 * path additional hardlinks may be possible even if the file is marked
1013 * immutable. The sysop needs to create a closure by checking the hard
1014 * link count. Once closure is achieved you are good, and security
1015 * scripts should check link counts anyway.
1016 *
1017 * Writes and truncations are only allowed on special devices.
21739618 1018 */
3a907475
MD
1019 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1020 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1021 return (EPERM);
1022 if (nflags & (NLC_CREATE | NLC_DELETE |
1023 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1024 return (EPERM);
1025 }
1026 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1027 switch(va->va_type) {
1028 case VDIR:
d7c75c7a 1029 return (EISDIR);
3a907475
MD
1030 case VLNK:
1031 case VREG:
1032 case VDATABASE:
1033 return (EPERM);
1034 default:
1035 break;
1036 }
1037 }
1038 }
1039
1040 /*
1041 * Test the no-unlink and append-only bits for opens, rename targets,
1042 * and deletions. These bits are not tested for creations or
1043 * rename sources.
1044 *
1045 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1046 * If you do not wish this you must also set NOUNLINK.
1047 *
1048 * If the governing directory is marked APPEND-only it implies
1049 * NOUNLINK for all entries in the directory.
1050 */
1051 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1052 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1053 ) {
1054 return (EPERM);
1055 }
1056
1057 /*
1058 * A file marked append-only may not be deleted but can be renamed.
1059 */
1060 if ((va->va_flags & APPEND) &&
1061 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1062 ) {
1063 return (EPERM);
1064 }
1065
1066 /*
1067 * A file marked append-only which is opened for writing must also
1068 * be opened O_APPEND.
1069 */
1070 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1071 if (nflags & NLC_TRUNCATE)
1072 return (EPERM);
1073 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1074 if ((nflags & NLC_APPEND) == 0)
21739618
MD
1075 return (EPERM);
1076 }
1077 }
1078
1079 /*
1080 * root gets universal access
1081 */
1082 if (cred->cr_uid == 0)
1083 return(0);
1084
1085 /*
dae8d54f
MD
1086 * Check owner perms.
1087 *
3a907475 1088 * If NLC_OWN is set the owner of the file is allowed no matter when
dae8d54f 1089 * the owner-mode bits say (utimes).
21739618 1090 */
3a907475
MD
1091 vmode = 0;
1092 if (nflags & NLC_READ)
1093 vmode |= S_IRUSR;
1094 if (nflags & NLC_WRITE)
1095 vmode |= S_IWUSR;
1096 if (nflags & NLC_EXEC)
1097 vmode |= S_IXUSR;
1098
21739618 1099 if (cred->cr_uid == va->va_uid) {
3a907475 1100 if ((nflags & NLC_OWN) == 0) {
ee89633d
MD
1101 if ((vmode & va->va_mode) != vmode)
1102 return(EACCES);
1103 }
21739618
MD
1104 return(0);
1105 }
dae8d54f
MD
1106
1107 /*
3a907475
MD
1108 * If NLC_STICKY is set only the owner may delete or rename a file.
1109 * This bit is typically set on /tmp.
dae8d54f 1110 *
3a907475
MD
1111 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1112 * the specific delete or rename case. For deletions and renames we
1113 * usually just care about directory permissions, not file permissions.
dae8d54f 1114 */
3a907475
MD
1115 if ((nflags & NLC_STICKY) &&
1116 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
dae8d54f 1117 return(EACCES);
3a907475 1118 }
dae8d54f
MD
1119
1120 /*
1121 * Check group perms
1122 */
21739618
MD
1123 vmode >>= 3;
1124 for (i = 0; i < cred->cr_ngroups; ++i) {
1125 if (va->va_gid == cred->cr_groups[i]) {
1126 if ((vmode & va->va_mode) != vmode)
1127 return(EACCES);
1128 return(0);
1129 }
1130 }
1131
dae8d54f
MD
1132 /*
1133 * Check world perms
1134 */
21739618
MD
1135 vmode >>= 3;
1136 if ((vmode & va->va_mode) != vmode)
1137 return(EACCES);
1138 return(0);
1139}
1140