kernel - namecache 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;
f63911bf 367 struct nchandle nctmp;
28623bf9 368 struct mount *mp;
f4d4e93a 369 int wasdotordotdot;
690a3127 370 char *ptr;
fad57d0e 371 char *xptr;
690a3127
MD
372 int error;
373 int len;
3a907475 374 int dflags;
690a3127
MD
375
376#ifdef KTRACE
377 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
9fb04d14 378 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
690a3127
MD
379#endif
380 bzero(&nlc, sizeof(nlc));
381
382 /*
383 * Setup for the loop. The current working namecache element must
524c845c
MD
384 * be in a refd + unlocked state. This typically the case on entry except
385 * when stringing nlookup()'s along in a chain, since nlookup() always
28623bf9 386 * returns nl_nch in a locked state.
690a3127
MD
387 */
388 nd->nl_loopcnt = 0;
389 if (nd->nl_flags & NLC_NCPISLOCKED) {
390 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 391 cache_unlock(&nd->nl_nch);
690a3127 392 }
5312fa43
MD
393 if (nd->nl_dvp ) {
394 vrele(nd->nl_dvp);
395 nd->nl_dvp = NULL;
396 }
690a3127
MD
397 ptr = nd->nl_path;
398
399 /*
28623bf9 400 * Loop on the path components. At the top of the loop nd->nl_nch
524c845c 401 * is ref'd and unlocked and represents our current position.
690a3127
MD
402 */
403 for (;;) {
404 /*
405 * Check if the root directory should replace the current
406 * directory. This is done at the start of a translation
407 * or after a symbolic link has been found. In other cases
408 * ptr will never be pointing at a '/'.
409 */
410 if (*ptr == '/') {
411 do {
412 ++ptr;
413 } while (*ptr == '/');
28623bf9
MD
414 cache_copy(&nd->nl_rootnch, &nch);
415 cache_drop(&nd->nl_nch);
416 nd->nl_nch = nch;
5312fa43
MD
417
418 /*
419 * Fast-track termination. There is no parent directory of
420 * the root in the same mount from the point of view of
421 * the caller so return EPERM if NLC_REFDVP is specified.
422 * e.g. 'rmdir /' is not allowed.
423 */
21739618 424 if (*ptr == 0) {
5312fa43
MD
425 if (nd->nl_flags & NLC_REFDVP) {
426 error = EPERM;
427 } else {
428 cache_lock(&nd->nl_nch);
429 nd->nl_flags |= NLC_NCPISLOCKED;
430 error = 0;
431 }
21739618
MD
432 break;
433 }
690a3127
MD
434 continue;
435 }
436
437 /*
28623bf9 438 * Check directory search permissions.
21739618 439 */
3a907475
MD
440 dflags = 0;
441 if ((error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags)) != 0)
21739618
MD
442 break;
443
444 /*
d7c75c7a
MD
445 * Extract the path component. Path components are limited to
446 * 255 characters.
690a3127
MD
447 */
448 nlc.nlc_nameptr = ptr;
449 while (*ptr && *ptr != '/')
450 ++ptr;
451 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
d7c75c7a
MD
452 if (nlc.nlc_namelen >= 256) {
453 error = ENAMETOOLONG;
454 break;
455 }
690a3127
MD
456
457 /*
21739618
MD
458 * Lookup the path component in the cache, creating an unresolved
459 * entry if necessary. We have to handle "." and ".." as special
460 * cases.
461 *
462 * When handling ".." we have to detect a traversal back through a
28623bf9 463 * mount point. If we are at the root, ".." just returns the root.
524c845c 464 *
3a907475
MD
465 * When handling "." or ".." we also have to recalculate dflags
466 * since our dflags will be for some sub-directory instead of the
467 * parent dir.
468 *
28623bf9 469 * This subsection returns a locked, refd 'nch' unless it errors out.
fad57d0e
MD
470 * The namecache topology is not allowed to be disconnected, so
471 * encountering a NULL parent will generate EINVAL. This typically
472 * occurs when a directory is removed out from under a process.
21739618
MD
473 */
474 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
28623bf9 475 cache_get(&nd->nl_nch, &nch);
f4d4e93a 476 wasdotordotdot = 1;
21739618
MD
477 } else if (nlc.nlc_namelen == 2 &&
478 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
28623bf9
MD
479 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
480 nd->nl_nch.ncp == nd->nl_rootnch.ncp
481 ) {
482 /*
483 * ".." at the root returns the root
484 */
485 cache_get(&nd->nl_nch, &nch);
21739618 486 } else {
28623bf9
MD
487 /*
488 * Locate the parent ncp. If we are at the root of a
489 * filesystem mount we have to skip to the mounted-on
490 * point in the underlying filesystem.
f63911bf
MD
491 *
492 * Expect the parent to always be good since the
493 * mountpoint doesn't go away. XXX hack. cache_get()
494 * requires the ncp to already have a ref as a safety.
28623bf9 495 */
f63911bf
MD
496 nctmp = nd->nl_nch;
497 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
498 nctmp = nctmp.mount->mnt_ncmounton;
499 nctmp.ncp = nctmp.ncp->nc_parent;
500 KKASSERT(nctmp.ncp != NULL);
501 cache_copy(&nctmp, &nch); /* XXX hack */
28623bf9 502 cache_get(&nch, &nch);
f63911bf 503 cache_drop(&nctmp); /* NOTE: zero's nctmp */
21739618 504 }
d7c75c7a 505 wasdotordotdot = 2;
21739618 506 } else {
28623bf9
MD
507 nch = cache_nlookup(&nd->nl_nch, &nlc);
508 while ((error = cache_resolve(&nch, nd->nl_cred)) == EAGAIN) {
6ea70f76 509 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
28623bf9
MD
510 nch.ncp->nc_nlen, nch.ncp->nc_nlen, nch.ncp->nc_name);
511 cache_put(&nch);
512 nch = cache_nlookup(&nd->nl_nch, &nlc);
8e005a45 513 }
f4d4e93a 514 wasdotordotdot = 0;
21739618 515 }
3a907475
MD
516
517 /*
518 * If the last component was "." or ".." our dflags no longer
519 * represents the parent directory and we have to explicitly
520 * look it up.
f63911bf
MD
521 *
522 * Expect the parent to be good since nch is locked.
3a907475
MD
523 */
524 if (wasdotordotdot && error == 0) {
525 dflags = 0;
526 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
527 par.mount = nch.mount;
528 cache_hold(&par);
529 dflags = 0;
530 error = naccess(&par, 0, nd->nl_cred, &dflags);
531 cache_drop(&par);
532 }
533 }
534
524c845c 535 /*
28623bf9 536 * [end of subsection] ncp is locked and ref'd. nd->nl_nch is ref'd
524c845c 537 */
21739618
MD
538
539 /*
540 * Resolve the namespace if necessary. The ncp returned by
541 * cache_nlookup() is referenced and locked.
8e005a45
MD
542 *
543 * XXX neither '.' nor '..' should return EAGAIN since they were
544 * previously resolved and thus cannot be newly created ncp's.
690a3127 545 */
28623bf9
MD
546 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
547 error = cache_resolve(&nch, nd->nl_cred);
8e005a45 548 KKASSERT(error != EAGAIN);
690a3127 549 } else {
28623bf9 550 error = nch.ncp->nc_error;
690a3127 551 }
21739618
MD
552
553 /*
fad57d0e 554 * Early completion. ENOENT is not an error if this is the last
945b476a
MD
555 * component and NLC_CREATE or NLC_RENAME (rename target) was
556 * requested. Note that ncp->nc_error is left as ENOENT in that
557 * case, which we check later on.
558 *
f4d4e93a 559 * Also handle invalid '.' or '..' components terminating a path
945b476a
MD
560 * for a create/rename/delete. The standard requires this and pax
561 * pretty stupidly depends on it.
fad57d0e
MD
562 */
563 for (xptr = ptr; *xptr == '/'; ++xptr)
564 ;
f4d4e93a 565 if (*xptr == 0) {
3a907475
MD
566 if (error == ENOENT &&
567 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
568 ) {
569 if (nd->nl_flags & NLC_NFS_RDONLY) {
c9aef211 570 error = EROFS;
3a907475
MD
571 } else {
572 error = naccess(&nch, nd->nl_flags | dflags,
573 nd->nl_cred, NULL);
574 }
c9aef211 575 }
945b476a 576 if (error == 0 && wasdotordotdot &&
3a907475
MD
577 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
578 NLC_RENAME_SRC | NLC_RENAME_DST))) {
d7c75c7a
MD
579 /*
580 * POSIX junk
581 */
582 if (nd->nl_flags & NLC_CREATE)
583 error = EEXIST;
584 else if (nd->nl_flags & NLC_DELETE)
585 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
586 else
587 error = EINVAL;
945b476a 588 }
fad57d0e
MD
589 }
590
591 /*
592 * Early completion on error.
21739618 593 */
690a3127 594 if (error) {
28623bf9 595 cache_put(&nch);
690a3127
MD
596 break;
597 }
598
599 /*
600 * If the element is a symlink and it is either not the last
601 * element or it is the last element and we are allowed to
602 * follow symlinks, resolve the symlink.
603 */
28623bf9 604 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
690a3127
MD
605 (*ptr || (nd->nl_flags & NLC_FOLLOW))
606 ) {
607 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
21739618 608 error = ELOOP;
28623bf9 609 cache_put(&nch);
690a3127
MD
610 break;
611 }
28623bf9
MD
612 error = nreadsymlink(nd, &nch, &nlc);
613 cache_put(&nch);
21739618
MD
614 if (error)
615 break;
690a3127
MD
616
617 /*
618 * Concatenate trailing path elements onto the returned symlink.
619 * Note that if the path component (ptr) is not exhausted, it
620 * will being with a '/', so we do not have to add another one.
621 *
622 * The symlink may not be empty.
623 */
624 len = strlen(ptr);
625 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
626 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
70aac194 627 objcache_put(namei_oc, nlc.nlc_nameptr);
690a3127
MD
628 break;
629 }
630 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
631 if (nd->nl_flags & NLC_HASBUF)
70aac194 632 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
633 nd->nl_path = nlc.nlc_nameptr;
634 nd->nl_flags |= NLC_HASBUF;
635 ptr = nd->nl_path;
636
637 /*
638 * Go back up to the top to resolve any initial '/'s in the
639 * symlink.
640 */
641 continue;
642 }
643
644 /*
21739618 645 * If the element is a directory and we are crossing a mount point,
28623bf9 646 * Locate the mount.
21739618 647 */
28623bf9
MD
648 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
649 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
650 (mp = cache_findmount(&nch)) != NULL
21739618 651 ) {
21739618
MD
652 struct vnode *tdp;
653
28623bf9
MD
654 cache_put(&nch);
655 cache_get(&mp->mnt_ncmountpt, &nch);
21739618 656
28623bf9 657 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 658 while (vfs_busy(mp, 0))
21739618
MD
659 ;
660 error = VFS_ROOT(mp, &tdp);
f9642f56 661 vfs_unbusy(mp);
524c845c 662 if (error)
21739618 663 break;
28623bf9 664 cache_setvp(&nch, tdp);
21739618
MD
665 vput(tdp);
666 }
667 }
668 if (error) {
28623bf9 669 cache_put(&nch);
21739618
MD
670 break;
671 }
672
673 /*
690a3127
MD
674 * Skip any slashes to get to the next element. If there
675 * are any slashes at all the current element must be a
fad57d0e
MD
676 * directory or, in the create case, intended to become a directory.
677 * If it isn't we break without incrementing ptr and fall through
678 * to the failure case below.
690a3127
MD
679 */
680 while (*ptr == '/') {
28623bf9 681 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
fad57d0e
MD
682 !(nd->nl_flags & NLC_WILLBEDIR)
683 ) {
690a3127 684 break;
fad57d0e 685 }
690a3127
MD
686 ++ptr;
687 }
688
689 /*
690 * Continuation case: additional elements and the current
691 * element is a directory.
692 */
28623bf9
MD
693 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
694 cache_drop(&nd->nl_nch);
695 cache_unlock(&nch);
696 nd->nl_nch = nch;
690a3127
MD
697 continue;
698 }
699
700 /*
701 * Failure case: additional elements and the current element
702 * is not a directory
703 */
704 if (*ptr) {
28623bf9 705 cache_put(&nch);
690a3127
MD
706 error = ENOTDIR;
707 break;
708 }
709
710 /*
fad57d0e
MD
711 * Successful lookup of last element.
712 *
3a907475
MD
713 * Check permissions if the target exists. If the target does not
714 * exist directory permissions were already tested in the early
715 * completion code above.
945b476a 716 *
3a907475
MD
717 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
718 * if the file is marked append-only, and NLC_STICKY if the directory
719 * containing the file is sticky.
fad57d0e 720 */
3a907475
MD
721 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
722 error = naccess(&nch, nd->nl_flags | dflags,
723 nd->nl_cred, NULL);
945b476a 724 if (error) {
28623bf9 725 cache_put(&nch);
fad57d0e
MD
726 break;
727 }
728 }
729
730 /*
3a907475 731 * Termination: no more elements.
5312fa43
MD
732 *
733 * If NLC_REFDVP is set acquire a referenced parent dvp.
690a3127 734 */
5312fa43
MD
735 if (nd->nl_flags & NLC_REFDVP) {
736 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
737 if (error) {
738 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
739 cache_put(&nch);
740 break;
741 }
742 }
28623bf9
MD
743 cache_drop(&nd->nl_nch);
744 nd->nl_nch = nch;
690a3127
MD
745 nd->nl_flags |= NLC_NCPISLOCKED;
746 error = 0;
747 break;
748 }
3a907475
MD
749
750 /*
751 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
752 * (ncp->nc_error will be ENOENT), but we will still return an error
753 * code of 0.
754 */
690a3127
MD
755 return(error);
756}
757
758/*
21739618
MD
759 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
760 * of continuity in the namecache tree by connecting the ncp related to the
761 * vnode under the mount to the ncp related to the mount's root vnode.
762 *
763 * If no error occured a locked, ref'd ncp is stored in *ncpp.
764 */
765int
28623bf9 766nlookup_mp(struct mount *mp, struct nchandle *nch)
21739618 767{
21739618
MD
768 struct vnode *vp;
769 int error;
770
771 error = 0;
28623bf9
MD
772 cache_get(&mp->mnt_ncmountpt, nch);
773 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 774 while (vfs_busy(mp, 0))
21739618
MD
775 ;
776 error = VFS_ROOT(mp, &vp);
f9642f56 777 vfs_unbusy(mp);
21739618 778 if (error) {
28623bf9 779 cache_put(nch);
21739618 780 } else {
28623bf9 781 cache_setvp(nch, vp);
21739618
MD
782 vput(vp);
783 }
784 }
21739618
MD
785 return(error);
786}
787
788/*
690a3127 789 * Read the contents of a symlink, allocate a path buffer out of the
70aac194 790 * namei_oc and initialize the supplied nlcomponent with the result.
690a3127
MD
791 *
792 * If an error occurs no buffer will be allocated or returned in the nlc.
793 */
794int
28623bf9 795nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
690a3127
MD
796 struct nlcomponent *nlc)
797{
21739618 798 struct vnode *vp;
690a3127
MD
799 struct iovec aiov;
800 struct uio auio;
801 int linklen;
802 int error;
803 char *cp;
804
805 nlc->nlc_nameptr = NULL;
806 nlc->nlc_namelen = 0;
28623bf9 807 if (nch->ncp->nc_vp == NULL)
690a3127 808 return(ENOENT);
28623bf9 809 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
690a3127 810 return(error);
70aac194 811 cp = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
812 aiov.iov_base = cp;
813 aiov.iov_len = MAXPATHLEN;
814 auio.uio_iov = &aiov;
815 auio.uio_iovcnt = 1;
816 auio.uio_offset = 0;
817 auio.uio_rw = UIO_READ;
818 auio.uio_segflg = UIO_SYSSPACE;
819 auio.uio_td = nd->nl_td;
820 auio.uio_resid = MAXPATHLEN - 1;
21739618 821 error = VOP_READLINK(vp, &auio, nd->nl_cred);
690a3127
MD
822 if (error)
823 goto fail;
21739618 824 linklen = MAXPATHLEN - 1 - auio.uio_resid;
690a3127
MD
825 if (varsym_enable) {
826 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
827 if (linklen < 0) {
828 error = ENAMETOOLONG;
829 goto fail;
830 }
831 }
832 cp[linklen] = 0;
833 nlc->nlc_nameptr = cp;
834 nlc->nlc_namelen = linklen;
21739618 835 vput(vp);
690a3127
MD
836 return(0);
837fail:
70aac194 838 objcache_put(namei_oc, cp);
21739618 839 vput(vp);
690a3127
MD
840 return(error);
841}
842
21739618
MD
843/*
844 * Check access [XXX cache vattr!] [XXX quota]
845 *
3a907475
MD
846 * Generally check the NLC_* access bits. All specified bits must pass
847 * for this function to return 0.
21739618 848 *
3a907475
MD
849 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
850 * access, otherwise it must exist. No error is returned in this case.
945b476a 851 *
3a907475 852 * The file must not exist if NLC_EXCL is specified.
21739618 853 *
3a907475
MD
854 * Directory permissions in general are tested for NLC_CREATE if the file
855 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
856 * whether the file exists or not.
21739618 857 *
3a907475
MD
858 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
859 * the latter is only tested if the target exists.
dae8d54f 860 *
21739618 861 * The passed ncp may or may not be locked. The caller should use a
3a907475
MD
862 * locked ncp on leaf lookups, especially for NLC_CREATE, NLC_RENAME_DST,
863 * NLC_DELETE, and NLC_EXCL checks.
21739618
MD
864 */
865int
3a907475 866naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
21739618
MD
867{
868 struct vnode *vp;
869 struct vattr va;
870 int error;
dae8d54f 871 int sticky;
21739618 872
28623bf9
MD
873 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
874 cache_lock(nch);
875 cache_resolve(nch, cred);
876 cache_unlock(nch);
21739618 877 }
28623bf9 878 error = nch->ncp->nc_error;
dae8d54f 879
945b476a 880 /*
3a907475
MD
881 * Directory permissions checks. Silently ignore ENOENT if these
882 * tests pass. It isn't an error.
945b476a 883 */
3a907475
MD
884 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
885 if (((nflags & NLC_CREATE) && nch->ncp->nc_vp == NULL) ||
886 ((nflags & NLC_DELETE) && nch->ncp->nc_vp != NULL) ||
887 ((nflags & NLC_RENAME_SRC) && nch->ncp->nc_vp != NULL) ||
888 (nflags & NLC_RENAME_DST)
21739618 889 ) {
f63911bf
MD
890 lwkt_tokref nlock;
891 struct nchandle par;
892
893 lwkt_gettoken(&nlock, &vfs_token);
28623bf9 894 if ((par.ncp = nch->ncp->nc_parent) == NULL) {
8e005a45 895 if (error != EAGAIN)
fad57d0e 896 error = EINVAL;
3a907475 897 } else if (error == 0 || error == ENOENT) {
28623bf9
MD
898 par.mount = nch->mount;
899 cache_hold(&par);
3a907475
MD
900 sticky = 0;
901 error = naccess(&par, NLC_WRITE, cred, NULL);
28623bf9 902 cache_drop(&par);
8e005a45 903 }
f63911bf 904 lwkt_reltoken(&nlock);
21739618 905 }
21739618 906 }
3a907475
MD
907
908 /*
909 * NLC_EXCL check. Target file must not exist.
910 */
911 if (error == 0 && (nflags & NLC_EXCL) && nch->ncp->nc_vp != NULL)
912 error = EEXIST;
913
914 /*
915 * Get the vnode attributes so we can do the rest of our checks.
916 *
917 * NOTE: We only call naccess_va() if the target exists.
918 */
21739618 919 if (error == 0) {
28623bf9 920 error = cache_vget(nch, cred, LK_SHARED, &vp);
21739618 921 if (error == ENOENT) {
3a907475
MD
922 /*
923 * Silently zero-out ENOENT if creating or renaming
924 * (rename target). It isn't an error.
925 */
926 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
21739618
MD
927 error = 0;
928 } else if (error == 0) {
3a907475
MD
929 /*
930 * Get the vnode attributes and check for illegal O_TRUNC
931 * requests and read-only mounts.
932 *
933 * NOTE: You can still open devices on read-only mounts for
934 * writing.
935 *
936 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
937 * check on the parent directory above.
938 *
939 * XXX cache the va in the namecache or in the vnode
940 */
941 error = VOP_GETATTR(vp, &va);
942 if (error == 0 && (nflags & NLC_TRUNCATE)) {
943 switch(va.va_type) {
944 case VREG:
945 case VDATABASE:
946 case VCHR:
947 case VBLK:
c9c08c19 948 case VFIFO:
3a907475 949 break;
d7c75c7a
MD
950 case VDIR:
951 error = EISDIR;
952 break;
3a907475
MD
953 default:
954 error = EINVAL;
955 break;
956 }
957 }
958 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
959 (vp->v_mount->mnt_flag & MNT_RDONLY)
960 ) {
961 switch(va.va_type) {
962 case VDIR:
963 case VLNK:
964 case VREG:
965 case VDATABASE:
966 error = EROFS;
967 break;
968 default:
969 break;
21739618
MD
970 }
971 }
972 vput(vp);
dae8d54f 973
3a907475
MD
974 /*
975 * Check permissions based on file attributes. The passed
976 * flags (*nflagsp) are modified with feedback based on
977 * special attributes and requirements.
978 */
dae8d54f
MD
979 if (error == 0) {
980 /*
3a907475 981 * Adjust the returned (*nflagsp) if non-NULL.
dae8d54f 982 */
3a907475
MD
983 if (nflagsp) {
984 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
985 *nflagsp |= NLC_STICKY;
986 if (va.va_flags & APPEND)
987 *nflagsp |= NLC_APPENDONLY;
988 if (va.va_flags & IMMUTABLE)
989 *nflagsp |= NLC_IMMUTABLE;
990 }
dae8d54f
MD
991
992 /*
993 * Process general access.
994 */
3a907475 995 error = naccess_va(&va, nflags, cred);
dae8d54f 996 }
21739618
MD
997 }
998 }
999 return(error);
1000}
1001
1002/*
1003 * Check the requested access against the given vattr using cred.
1004 */
1005int
3a907475 1006naccess_va(struct vattr *va, int nflags, struct ucred *cred)
21739618
MD
1007{
1008 int i;
3a907475 1009 int vmode;
21739618
MD
1010
1011 /*
3a907475
MD
1012 * Test the immutable bit. Creations, deletions, renames (source
1013 * or destination) are not allowed. chown/chmod/other is also not
1014 * allowed but is handled by SETATTR. Hardlinks to the immutable
1015 * file are allowed.
1016 *
1017 * If the directory is set to immutable then creations, deletions,
1018 * renames (source or dest) and hardlinks to files within the directory
1019 * are not allowed, and regular files opened through the directory may
1020 * not be written to or truncated (unless a special device).
945b476a 1021 *
3a907475
MD
1022 * NOTE! New hardlinks to immutable files work but new hardlinks to
1023 * files, immutable or not, sitting inside an immutable directory are
1024 * not allowed. As always if the file is hardlinked via some other
1025 * path additional hardlinks may be possible even if the file is marked
1026 * immutable. The sysop needs to create a closure by checking the hard
1027 * link count. Once closure is achieved you are good, and security
1028 * scripts should check link counts anyway.
1029 *
1030 * Writes and truncations are only allowed on special devices.
21739618 1031 */
3a907475
MD
1032 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1033 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1034 return (EPERM);
1035 if (nflags & (NLC_CREATE | NLC_DELETE |
1036 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1037 return (EPERM);
1038 }
1039 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1040 switch(va->va_type) {
1041 case VDIR:
d7c75c7a 1042 return (EISDIR);
3a907475
MD
1043 case VLNK:
1044 case VREG:
1045 case VDATABASE:
1046 return (EPERM);
1047 default:
1048 break;
1049 }
1050 }
1051 }
1052
1053 /*
1054 * Test the no-unlink and append-only bits for opens, rename targets,
1055 * and deletions. These bits are not tested for creations or
1056 * rename sources.
1057 *
1058 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1059 * If you do not wish this you must also set NOUNLINK.
1060 *
1061 * If the governing directory is marked APPEND-only it implies
1062 * NOUNLINK for all entries in the directory.
1063 */
1064 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1065 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1066 ) {
1067 return (EPERM);
1068 }
1069
1070 /*
1071 * A file marked append-only may not be deleted but can be renamed.
1072 */
1073 if ((va->va_flags & APPEND) &&
1074 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1075 ) {
1076 return (EPERM);
1077 }
1078
1079 /*
1080 * A file marked append-only which is opened for writing must also
1081 * be opened O_APPEND.
1082 */
1083 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1084 if (nflags & NLC_TRUNCATE)
1085 return (EPERM);
1086 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1087 if ((nflags & NLC_APPEND) == 0)
21739618
MD
1088 return (EPERM);
1089 }
1090 }
1091
1092 /*
1093 * root gets universal access
1094 */
1095 if (cred->cr_uid == 0)
1096 return(0);
1097
1098 /*
dae8d54f
MD
1099 * Check owner perms.
1100 *
3a907475 1101 * If NLC_OWN is set the owner of the file is allowed no matter when
dae8d54f 1102 * the owner-mode bits say (utimes).
21739618 1103 */
3a907475
MD
1104 vmode = 0;
1105 if (nflags & NLC_READ)
1106 vmode |= S_IRUSR;
1107 if (nflags & NLC_WRITE)
1108 vmode |= S_IWUSR;
1109 if (nflags & NLC_EXEC)
1110 vmode |= S_IXUSR;
1111
21739618 1112 if (cred->cr_uid == va->va_uid) {
3a907475 1113 if ((nflags & NLC_OWN) == 0) {
ee89633d
MD
1114 if ((vmode & va->va_mode) != vmode)
1115 return(EACCES);
1116 }
21739618
MD
1117 return(0);
1118 }
dae8d54f
MD
1119
1120 /*
3a907475
MD
1121 * If NLC_STICKY is set only the owner may delete or rename a file.
1122 * This bit is typically set on /tmp.
dae8d54f 1123 *
3a907475
MD
1124 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1125 * the specific delete or rename case. For deletions and renames we
1126 * usually just care about directory permissions, not file permissions.
dae8d54f 1127 */
3a907475
MD
1128 if ((nflags & NLC_STICKY) &&
1129 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
dae8d54f 1130 return(EACCES);
3a907475 1131 }
dae8d54f
MD
1132
1133 /*
1134 * Check group perms
1135 */
21739618
MD
1136 vmode >>= 3;
1137 for (i = 0; i < cred->cr_ngroups; ++i) {
1138 if (va->va_gid == cred->cr_groups[i]) {
1139 if ((vmode & va->va_mode) != vmode)
1140 return(EACCES);
1141 return(0);
1142 }
1143 }
1144
dae8d54f
MD
1145 /*
1146 * Check world perms
1147 */
21739618
MD
1148 vmode >>= 3;
1149 if ((vmode & va->va_mode) != vmode)
1150 return(EACCES);
1151 return(0);
1152}
1153