kernel - Implement shared namecache locks
[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
fad57d0e
MD
186/*
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
46077e84
AH
230/*
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
1f95166e
HP
274/*
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
690a3127 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.
79fd1696
MD
405 *
406 * NOTE: As an optimization we attempt to obtain a shared namecache lock
407 * on any intermediate elements. On success, the returned element
408 * is ALWAYS locked exclusively.
690a3127 409 */
79fd1696
MD
410static
411int
412islastelement(const char *ptr)
413{
414 while (*ptr == '/')
415 ++ptr;
416 return (*ptr == 0);
417}
418
690a3127
MD
419int
420nlookup(struct nlookupdata *nd)
421{
5a9972aa 422 globaldata_t gd = mycpu;
690a3127 423 struct nlcomponent nlc;
28623bf9 424 struct nchandle nch;
3a907475 425 struct nchandle par;
f63911bf 426 struct nchandle nctmp;
28623bf9 427 struct mount *mp;
7222030f 428 struct vnode *hvp; /* hold to prevent recyclement */
f4d4e93a 429 int wasdotordotdot;
690a3127
MD
430 char *ptr;
431 int error;
432 int len;
3a907475 433 int dflags;
5a9972aa 434 int hit = 1;
690a3127
MD
435
436#ifdef KTRACE
437 if (KTRPOINT(nd->nl_td, KTR_NAMEI))
9fb04d14 438 ktrnamei(nd->nl_td->td_lwp, nd->nl_path);
690a3127
MD
439#endif
440 bzero(&nlc, sizeof(nlc));
441
442 /*
2247fe02
MD
443 * Setup for the loop. The current working namecache element is
444 * always at least referenced. We lock it as required, but always
445 * return a locked, resolved namecache entry.
690a3127
MD
446 */
447 nd->nl_loopcnt = 0;
2247fe02 448 if (nd->nl_dvp) {
5312fa43
MD
449 vrele(nd->nl_dvp);
450 nd->nl_dvp = NULL;
451 }
690a3127
MD
452 ptr = nd->nl_path;
453
454 /*
28623bf9 455 * Loop on the path components. At the top of the loop nd->nl_nch
524c845c 456 * is ref'd and unlocked and represents our current position.
690a3127
MD
457 */
458 for (;;) {
2247fe02
MD
459 /*
460 * Make sure nl_nch is locked so we can access the vnode, resolution
461 * state, etc.
462 */
463 if ((nd->nl_flags & NLC_NCPISLOCKED) == 0) {
464 nd->nl_flags |= NLC_NCPISLOCKED;
79fd1696 465 cache_lock_maybe_shared(&nd->nl_nch, islastelement(ptr));
2247fe02
MD
466 }
467
690a3127
MD
468 /*
469 * Check if the root directory should replace the current
470 * directory. This is done at the start of a translation
471 * or after a symbolic link has been found. In other cases
472 * ptr will never be pointing at a '/'.
473 */
474 if (*ptr == '/') {
475 do {
476 ++ptr;
477 } while (*ptr == '/');
79fd1696
MD
478 cache_unlock(&nd->nl_nch);
479 cache_get_maybe_shared(&nd->nl_rootnch, &nch,
480 islastelement(ptr));
481 cache_drop(&nd->nl_nch);
2247fe02 482 nd->nl_nch = nch; /* remains locked */
5312fa43
MD
483
484 /*
485 * Fast-track termination. There is no parent directory of
486 * the root in the same mount from the point of view of
1acebc70 487 * the caller so return EACCES if NLC_REFDVP is specified,
488 * and EEXIST if NLC_CREATE is also specified.
489 * e.g. 'rmdir /' or 'mkdir /' are not allowed.
5312fa43 490 */
21739618 491 if (*ptr == 0) {
2247fe02 492 if (nd->nl_flags & NLC_REFDVP)
1acebc70 493 error = (nd->nl_flags & NLC_CREATE) ? EEXIST : EACCES;
2247fe02 494 else
5312fa43 495 error = 0;
21739618
MD
496 break;
497 }
690a3127
MD
498 continue;
499 }
500
21739618 501 /*
79fd1696 502 * Check directory search permissions (nd->nl_nch is locked & refd)
21739618 503 */
3a907475 504 dflags = 0;
aabd5ce8 505 error = naccess(&nd->nl_nch, NLC_EXEC, nd->nl_cred, &dflags);
e9b56058 506 if (error)
21739618
MD
507 break;
508
690a3127 509 /*
d7c75c7a
MD
510 * Extract the path component. Path components are limited to
511 * 255 characters.
690a3127
MD
512 */
513 nlc.nlc_nameptr = ptr;
514 while (*ptr && *ptr != '/')
515 ++ptr;
516 nlc.nlc_namelen = ptr - nlc.nlc_nameptr;
d7c75c7a
MD
517 if (nlc.nlc_namelen >= 256) {
518 error = ENAMETOOLONG;
519 break;
520 }
690a3127
MD
521
522 /*
21739618
MD
523 * Lookup the path component in the cache, creating an unresolved
524 * entry if necessary. We have to handle "." and ".." as special
525 * cases.
526 *
527 * When handling ".." we have to detect a traversal back through a
28623bf9 528 * mount point. If we are at the root, ".." just returns the root.
524c845c 529 *
3a907475
MD
530 * When handling "." or ".." we also have to recalculate dflags
531 * since our dflags will be for some sub-directory instead of the
532 * parent dir.
533 *
79fd1696
MD
534 * This subsection returns a locked, refd 'nch' unless it errors out,
535 * and an unlocked but still ref'd nd->nl_nch.
536 *
fad57d0e
MD
537 * The namecache topology is not allowed to be disconnected, so
538 * encountering a NULL parent will generate EINVAL. This typically
539 * occurs when a directory is removed out from under a process.
79fd1696
MD
540 *
541 * WARNING! The unlocking of nd->nl_nch is sensitive code.
21739618 542 */
79fd1696
MD
543 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
544
21739618 545 if (nlc.nlc_namelen == 1 && nlc.nlc_nameptr[0] == '.') {
79fd1696
MD
546 cache_unlock(&nd->nl_nch);
547 nd->nl_flags &= ~NLC_NCPISLOCKED;
548 cache_get_maybe_shared(&nd->nl_nch, &nch, islastelement(ptr));
f4d4e93a 549 wasdotordotdot = 1;
21739618
MD
550 } else if (nlc.nlc_namelen == 2 &&
551 nlc.nlc_nameptr[0] == '.' && nlc.nlc_nameptr[1] == '.') {
28623bf9
MD
552 if (nd->nl_nch.mount == nd->nl_rootnch.mount &&
553 nd->nl_nch.ncp == nd->nl_rootnch.ncp
554 ) {
555 /*
556 * ".." at the root returns the root
557 */
79fd1696
MD
558 cache_unlock(&nd->nl_nch);
559 nd->nl_flags &= ~NLC_NCPISLOCKED;
560 cache_get_maybe_shared(&nd->nl_nch, &nch, islastelement(ptr));
21739618 561 } else {
28623bf9
MD
562 /*
563 * Locate the parent ncp. If we are at the root of a
564 * filesystem mount we have to skip to the mounted-on
565 * point in the underlying filesystem.
f63911bf
MD
566 *
567 * Expect the parent to always be good since the
568 * mountpoint doesn't go away. XXX hack. cache_get()
569 * requires the ncp to already have a ref as a safety.
28623bf9 570 */
f63911bf
MD
571 nctmp = nd->nl_nch;
572 while (nctmp.ncp == nctmp.mount->mnt_ncmountpt.ncp)
573 nctmp = nctmp.mount->mnt_ncmounton;
574 nctmp.ncp = nctmp.ncp->nc_parent;
575 KKASSERT(nctmp.ncp != NULL);
2247fe02 576 cache_hold(&nctmp);
79fd1696
MD
577 cache_unlock(&nd->nl_nch);
578 nd->nl_flags &= ~NLC_NCPISLOCKED;
579 cache_get_maybe_shared(&nctmp, &nch, islastelement(ptr));
f63911bf 580 cache_drop(&nctmp); /* NOTE: zero's nctmp */
21739618 581 }
d7c75c7a 582 wasdotordotdot = 2;
21739618 583 } else {
2247fe02 584 /*
7222030f
MD
585 * Must unlock nl_nch when traversing down the path. However,
586 * the child ncp has not yet been found/created and the parent's
587 * child list might be empty. Thus releasing the lock can
588 * allow a race whereby the parent ncp's vnode is recycled.
589 * This case can occur especially when maxvnodes is set very low.
590 *
591 * We need the parent's ncp to remain resolved for all normal
592 * filesystem activities, so we vhold() the vp during the lookup
593 * to prevent recyclement due to vnlru / maxvnodes.
8d09ad3d
MD
594 *
595 * If we race an unlink or rename the ncp might be marked
596 * DESTROYED after resolution, requiring a retry.
2247fe02 597 */
7222030f
MD
598 if ((hvp = nd->nl_nch.ncp->nc_vp) != NULL)
599 vhold(hvp);
2247fe02
MD
600 cache_unlock(&nd->nl_nch);
601 nd->nl_flags &= ~NLC_NCPISLOCKED;
79fd1696
MD
602 error = cache_nlookup_maybe_shared(&nd->nl_nch, &nlc,
603 islastelement(ptr), &nch);
604 if (error == EWOULDBLOCK) {
605 nch = cache_nlookup(&nd->nl_nch, &nlc);
606 if (nch.ncp->nc_flag & NCF_UNRESOLVED)
607 hit = 0;
608 for (;;) {
609 error = cache_resolve(&nch, nd->nl_cred);
610 if (error != EAGAIN &&
611 (nch.ncp->nc_flag & NCF_DESTROYED) == 0) {
612 break;
613 }
614 kprintf("[diagnostic] nlookup: relookup %*.*s\n",
615 nch.ncp->nc_nlen, nch.ncp->nc_nlen,
616 nch.ncp->nc_name);
617 cache_put(&nch);
618 nch = cache_nlookup(&nd->nl_nch, &nlc);
619 }
8e005a45 620 }
7222030f
MD
621 if (hvp)
622 vdrop(hvp);
f4d4e93a 623 wasdotordotdot = 0;
21739618 624 }
3a907475
MD
625
626 /*
627 * If the last component was "." or ".." our dflags no longer
628 * represents the parent directory and we have to explicitly
629 * look it up.
f63911bf
MD
630 *
631 * Expect the parent to be good since nch is locked.
3a907475
MD
632 */
633 if (wasdotordotdot && error == 0) {
634 dflags = 0;
635 if ((par.ncp = nch.ncp->nc_parent) != NULL) {
636 par.mount = nch.mount;
637 cache_hold(&par);
79fd1696 638 cache_lock_maybe_shared(&par, islastelement(ptr));
aabd5ce8 639 error = naccess(&par, 0, nd->nl_cred, &dflags);
2247fe02 640 cache_put(&par);
3a907475
MD
641 }
642 }
643
524c845c 644 /*
2247fe02
MD
645 * [end of subsection]
646 *
647 * nch is locked and referenced.
648 * nd->nl_nch is unlocked and referenced.
649 *
650 * nl_nch must be unlocked or we could chain lock to the root
651 * if a resolve gets stuck (e.g. in NFS).
524c845c 652 */
79fd1696 653 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
21739618
MD
654
655 /*
656 * Resolve the namespace if necessary. The ncp returned by
657 * cache_nlookup() is referenced and locked.
8e005a45
MD
658 *
659 * XXX neither '.' nor '..' should return EAGAIN since they were
660 * previously resolved and thus cannot be newly created ncp's.
690a3127 661 */
28623bf9 662 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
5a9972aa 663 hit = 0;
28623bf9 664 error = cache_resolve(&nch, nd->nl_cred);
8e005a45 665 KKASSERT(error != EAGAIN);
690a3127 666 } else {
28623bf9 667 error = nch.ncp->nc_error;
690a3127 668 }
21739618
MD
669
670 /*
fad57d0e 671 * Early completion. ENOENT is not an error if this is the last
945b476a
MD
672 * component and NLC_CREATE or NLC_RENAME (rename target) was
673 * requested. Note that ncp->nc_error is left as ENOENT in that
674 * case, which we check later on.
675 *
f4d4e93a 676 * Also handle invalid '.' or '..' components terminating a path
945b476a
MD
677 * for a create/rename/delete. The standard requires this and pax
678 * pretty stupidly depends on it.
fad57d0e 679 */
79fd1696 680 if (islastelement(ptr)) {
3a907475
MD
681 if (error == ENOENT &&
682 (nd->nl_flags & (NLC_CREATE | NLC_RENAME_DST))
683 ) {
684 if (nd->nl_flags & NLC_NFS_RDONLY) {
c9aef211 685 error = EROFS;
3a907475
MD
686 } else {
687 error = naccess(&nch, nd->nl_flags | dflags,
aabd5ce8 688 nd->nl_cred, NULL);
3a907475 689 }
c9aef211 690 }
945b476a 691 if (error == 0 && wasdotordotdot &&
3a907475
MD
692 (nd->nl_flags & (NLC_CREATE | NLC_DELETE |
693 NLC_RENAME_SRC | NLC_RENAME_DST))) {
d7c75c7a
MD
694 /*
695 * POSIX junk
696 */
697 if (nd->nl_flags & NLC_CREATE)
698 error = EEXIST;
699 else if (nd->nl_flags & NLC_DELETE)
700 error = (wasdotordotdot == 1) ? EINVAL : ENOTEMPTY;
701 else
702 error = EINVAL;
945b476a 703 }
fad57d0e
MD
704 }
705
706 /*
707 * Early completion on error.
21739618 708 */
690a3127 709 if (error) {
28623bf9 710 cache_put(&nch);
690a3127
MD
711 break;
712 }
713
714 /*
715 * If the element is a symlink and it is either not the last
716 * element or it is the last element and we are allowed to
717 * follow symlinks, resolve the symlink.
718 */
28623bf9 719 if ((nch.ncp->nc_flag & NCF_ISSYMLINK) &&
690a3127
MD
720 (*ptr || (nd->nl_flags & NLC_FOLLOW))
721 ) {
722 if (nd->nl_loopcnt++ >= MAXSYMLINKS) {
21739618 723 error = ELOOP;
28623bf9 724 cache_put(&nch);
690a3127
MD
725 break;
726 }
28623bf9
MD
727 error = nreadsymlink(nd, &nch, &nlc);
728 cache_put(&nch);
21739618
MD
729 if (error)
730 break;
690a3127
MD
731
732 /*
733 * Concatenate trailing path elements onto the returned symlink.
734 * Note that if the path component (ptr) is not exhausted, it
735 * will being with a '/', so we do not have to add another one.
736 *
737 * The symlink may not be empty.
738 */
739 len = strlen(ptr);
740 if (nlc.nlc_namelen == 0 || nlc.nlc_namelen + len >= MAXPATHLEN) {
741 error = nlc.nlc_namelen ? ENAMETOOLONG : ENOENT;
70aac194 742 objcache_put(namei_oc, nlc.nlc_nameptr);
690a3127
MD
743 break;
744 }
745 bcopy(ptr, nlc.nlc_nameptr + nlc.nlc_namelen, len + 1);
746 if (nd->nl_flags & NLC_HASBUF)
70aac194 747 objcache_put(namei_oc, nd->nl_path);
690a3127
MD
748 nd->nl_path = nlc.nlc_nameptr;
749 nd->nl_flags |= NLC_HASBUF;
750 ptr = nd->nl_path;
751
752 /*
753 * Go back up to the top to resolve any initial '/'s in the
754 * symlink.
755 */
756 continue;
757 }
758
21739618
MD
759 /*
760 * If the element is a directory and we are crossing a mount point,
28623bf9 761 * Locate the mount.
21739618 762 */
28623bf9
MD
763 while ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
764 (nd->nl_flags & NLC_NOCROSSMOUNT) == 0 &&
765 (mp = cache_findmount(&nch)) != NULL
21739618 766 ) {
21739618 767 struct vnode *tdp;
63f86bf7 768 int vfs_do_busy = 0;
21739618 769
63f86bf7
MD
770 /*
771 * VFS must be busied before the namecache entry is locked,
772 * but we don't want to waste time calling vfs_busy() if the
773 * mount point is already resolved.
774 */
775again:
28623bf9 776 cache_put(&nch);
63f86bf7 777 if (vfs_do_busy) {
ace53c28
MD
778 while (vfs_busy(mp, 0)) {
779 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
780 kprintf("nlookup: warning umount race avoided\n");
781 cache_dropmount(mp);
782 error = EBUSY;
783 vfs_do_busy = 0;
784 goto double_break;
785 }
786 }
63f86bf7 787 }
79fd1696
MD
788 cache_get_maybe_shared(&mp->mnt_ncmountpt, &nch,
789 islastelement(ptr));
21739618 790
28623bf9 791 if (nch.ncp->nc_flag & NCF_UNRESOLVED) {
63f86bf7
MD
792 if (vfs_do_busy == 0) {
793 vfs_do_busy = 1;
794 goto again;
795 }
21739618 796 error = VFS_ROOT(mp, &tdp);
f9642f56 797 vfs_unbusy(mp);
63f86bf7 798 vfs_do_busy = 0;
6fa06591
MD
799 if (error) {
800 cache_dropmount(mp);
21739618 801 break;
6fa06591 802 }
28623bf9 803 cache_setvp(&nch, tdp);
21739618
MD
804 vput(tdp);
805 }
63f86bf7
MD
806 if (vfs_do_busy)
807 vfs_unbusy(mp);
6fa06591 808 cache_dropmount(mp);
21739618 809 }
ace53c28 810
21739618 811 if (error) {
28623bf9 812 cache_put(&nch);
d6069a5c 813double_break:
21739618
MD
814 break;
815 }
816
690a3127
MD
817 /*
818 * Skip any slashes to get to the next element. If there
819 * are any slashes at all the current element must be a
fad57d0e
MD
820 * directory or, in the create case, intended to become a directory.
821 * If it isn't we break without incrementing ptr and fall through
822 * to the failure case below.
690a3127
MD
823 */
824 while (*ptr == '/') {
28623bf9 825 if ((nch.ncp->nc_flag & NCF_ISDIR) == 0 &&
fad57d0e
MD
826 !(nd->nl_flags & NLC_WILLBEDIR)
827 ) {
690a3127 828 break;
fad57d0e 829 }
690a3127
MD
830 ++ptr;
831 }
832
833 /*
834 * Continuation case: additional elements and the current
835 * element is a directory.
836 */
28623bf9
MD
837 if (*ptr && (nch.ncp->nc_flag & NCF_ISDIR)) {
838 cache_drop(&nd->nl_nch);
839 cache_unlock(&nch);
2247fe02 840 KKASSERT((nd->nl_flags & NLC_NCPISLOCKED) == 0);
28623bf9 841 nd->nl_nch = nch;
690a3127
MD
842 continue;
843 }
844
845 /*
846 * Failure case: additional elements and the current element
847 * is not a directory
848 */
849 if (*ptr) {
28623bf9 850 cache_put(&nch);
690a3127
MD
851 error = ENOTDIR;
852 break;
853 }
854
fad57d0e
MD
855 /*
856 * Successful lookup of last element.
857 *
3a907475
MD
858 * Check permissions if the target exists. If the target does not
859 * exist directory permissions were already tested in the early
860 * completion code above.
945b476a 861 *
3a907475
MD
862 * nd->nl_flags will be adjusted on return with NLC_APPENDONLY
863 * if the file is marked append-only, and NLC_STICKY if the directory
864 * containing the file is sticky.
fad57d0e 865 */
3a907475
MD
866 if (nch.ncp->nc_vp && (nd->nl_flags & NLC_ALLCHKS)) {
867 error = naccess(&nch, nd->nl_flags | dflags,
aabd5ce8 868 nd->nl_cred, NULL);
945b476a 869 if (error) {
28623bf9 870 cache_put(&nch);
fad57d0e
MD
871 break;
872 }
873 }
874
690a3127 875 /*
3a907475 876 * Termination: no more elements.
5312fa43
MD
877 *
878 * If NLC_REFDVP is set acquire a referenced parent dvp.
690a3127 879 */
5312fa43 880 if (nd->nl_flags & NLC_REFDVP) {
2247fe02 881 cache_lock(&nd->nl_nch);
5312fa43 882 error = cache_vref(&nd->nl_nch, nd->nl_cred, &nd->nl_dvp);
2247fe02 883 cache_unlock(&nd->nl_nch);
5312fa43
MD
884 if (error) {
885 kprintf("NLC_REFDVP: Cannot ref dvp of %p\n", nch.ncp);
886 cache_put(&nch);
887 break;
888 }
889 }
28623bf9
MD
890 cache_drop(&nd->nl_nch);
891 nd->nl_nch = nch;
690a3127
MD
892 nd->nl_flags |= NLC_NCPISLOCKED;
893 error = 0;
894 break;
895 }
3a907475 896
5a9972aa 897 if (hit)
79fd1696 898 ++gd->gd_nchstats->ncs_longhits;
5a9972aa 899 else
79fd1696
MD
900 ++gd->gd_nchstats->ncs_longmiss;
901
902 if (nd->nl_flags & NLC_NCPISLOCKED)
903 KKASSERT(cache_lockstatus(&nd->nl_nch) == LK_EXCLUSIVE);
5a9972aa 904
3a907475
MD
905 /*
906 * NOTE: If NLC_CREATE was set the ncp may represent a negative hit
907 * (ncp->nc_error will be ENOENT), but we will still return an error
908 * code of 0.
909 */
690a3127
MD
910 return(error);
911}
912
21739618
MD
913/*
914 * Resolve a mount point's glue ncp. This ncp connects creates the illusion
915 * of continuity in the namecache tree by connecting the ncp related to the
916 * vnode under the mount to the ncp related to the mount's root vnode.
917 *
918 * If no error occured a locked, ref'd ncp is stored in *ncpp.
919 */
920int
28623bf9 921nlookup_mp(struct mount *mp, struct nchandle *nch)
21739618 922{
21739618
MD
923 struct vnode *vp;
924 int error;
925
926 error = 0;
28623bf9
MD
927 cache_get(&mp->mnt_ncmountpt, nch);
928 if (nch->ncp->nc_flag & NCF_UNRESOLVED) {
f9642f56 929 while (vfs_busy(mp, 0))
21739618
MD
930 ;
931 error = VFS_ROOT(mp, &vp);
f9642f56 932 vfs_unbusy(mp);
21739618 933 if (error) {
28623bf9 934 cache_put(nch);
21739618 935 } else {
28623bf9 936 cache_setvp(nch, vp);
21739618
MD
937 vput(vp);
938 }
939 }
21739618
MD
940 return(error);
941}
942
690a3127
MD
943/*
944 * Read the contents of a symlink, allocate a path buffer out of the
70aac194 945 * namei_oc and initialize the supplied nlcomponent with the result.
690a3127
MD
946 *
947 * If an error occurs no buffer will be allocated or returned in the nlc.
948 */
949int
28623bf9 950nreadsymlink(struct nlookupdata *nd, struct nchandle *nch,
690a3127
MD
951 struct nlcomponent *nlc)
952{
21739618 953 struct vnode *vp;
690a3127
MD
954 struct iovec aiov;
955 struct uio auio;
956 int linklen;
957 int error;
958 char *cp;
959
960 nlc->nlc_nameptr = NULL;
961 nlc->nlc_namelen = 0;
28623bf9 962 if (nch->ncp->nc_vp == NULL)
690a3127 963 return(ENOENT);
28623bf9 964 if ((error = cache_vget(nch, nd->nl_cred, LK_SHARED, &vp)) != 0)
690a3127 965 return(error);
70aac194 966 cp = objcache_get(namei_oc, M_WAITOK);
690a3127
MD
967 aiov.iov_base = cp;
968 aiov.iov_len = MAXPATHLEN;
969 auio.uio_iov = &aiov;
970 auio.uio_iovcnt = 1;
971 auio.uio_offset = 0;
972 auio.uio_rw = UIO_READ;
973 auio.uio_segflg = UIO_SYSSPACE;
974 auio.uio_td = nd->nl_td;
975 auio.uio_resid = MAXPATHLEN - 1;
21739618 976 error = VOP_READLINK(vp, &auio, nd->nl_cred);
690a3127
MD
977 if (error)
978 goto fail;
21739618 979 linklen = MAXPATHLEN - 1 - auio.uio_resid;
690a3127
MD
980 if (varsym_enable) {
981 linklen = varsymreplace(cp, linklen, MAXPATHLEN - 1);
982 if (linklen < 0) {
983 error = ENAMETOOLONG;
984 goto fail;
985 }
986 }
987 cp[linklen] = 0;
988 nlc->nlc_nameptr = cp;
989 nlc->nlc_namelen = linklen;
21739618 990 vput(vp);
690a3127
MD
991 return(0);
992fail:
70aac194 993 objcache_put(namei_oc, cp);
21739618 994 vput(vp);
690a3127
MD
995 return(error);
996}
997
21739618
MD
998/*
999 * Check access [XXX cache vattr!] [XXX quota]
1000 *
3a907475
MD
1001 * Generally check the NLC_* access bits. All specified bits must pass
1002 * for this function to return 0.
21739618 1003 *
3a907475
MD
1004 * The file does not have to exist when checking NLC_CREATE or NLC_RENAME_DST
1005 * access, otherwise it must exist. No error is returned in this case.
945b476a 1006 *
3a907475 1007 * The file must not exist if NLC_EXCL is specified.
21739618 1008 *
3a907475
MD
1009 * Directory permissions in general are tested for NLC_CREATE if the file
1010 * does not exist, NLC_DELETE if the file does exist, and NLC_RENAME_DST
1011 * whether the file exists or not.
21739618 1012 *
3a907475
MD
1013 * The directory sticky bit is tested for NLC_DELETE and NLC_RENAME_DST,
1014 * the latter is only tested if the target exists.
dae8d54f 1015 *
79fd1696
MD
1016 * The passed ncp must be referenced and locked. If it is already resolved
1017 * it may be locked shared but otherwise should be locked exclusively.
21739618 1018 */
5a3f9be0 1019static int
aabd5ce8 1020naccess(struct nchandle *nch, int nflags, struct ucred *cred, int *nflagsp)
21739618
MD
1021{
1022 struct vnode *vp;
1023 struct vattr va;
aabd5ce8 1024 struct namecache *ncp;
21739618 1025 int error;
e9b56058 1026 int cflags;
21739618 1027
79fd1696
MD
1028 KKASSERT(cache_lockstatus(nch) > 0);
1029
aabd5ce8
MD
1030 ncp = nch->ncp;
1031 if (ncp->nc_flag & NCF_UNRESOLVED) {
28623bf9 1032 cache_resolve(nch, cred);
aabd5ce8 1033 ncp = nch->ncp;
21739618 1034 }
aabd5ce8 1035 error = ncp->nc_error;
dae8d54f 1036
945b476a 1037 /*
3a907475
MD
1038 * Directory permissions checks. Silently ignore ENOENT if these
1039 * tests pass. It isn't an error.
2247fe02 1040 *
aabd5ce8
MD
1041 * We can safely resolve ncp->nc_parent because ncp is currently
1042 * locked.
945b476a 1043 */
3a907475 1044 if (nflags & (NLC_CREATE | NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST)) {
aabd5ce8
MD
1045 if (((nflags & NLC_CREATE) && ncp->nc_vp == NULL) ||
1046 ((nflags & NLC_DELETE) && ncp->nc_vp != NULL) ||
1047 ((nflags & NLC_RENAME_SRC) && ncp->nc_vp != NULL) ||
3a907475 1048 (nflags & NLC_RENAME_DST)
21739618 1049 ) {
f63911bf
MD
1050 struct nchandle par;
1051
aabd5ce8 1052 if ((par.ncp = ncp->nc_parent) == NULL) {
8e005a45 1053 if (error != EAGAIN)
fad57d0e 1054 error = EINVAL;
3a907475 1055 } else if (error == 0 || error == ENOENT) {
28623bf9 1056 par.mount = nch->mount;
2247fe02 1057 cache_hold(&par);
79fd1696 1058 cache_lock_maybe_shared(&par, 0);
aabd5ce8 1059 error = naccess(&par, NLC_WRITE, cred, NULL);
2247fe02 1060 cache_put(&par);
8e005a45 1061 }
21739618 1062 }
21739618 1063 }
3a907475
MD
1064
1065 /*
1066 * NLC_EXCL check. Target file must not exist.
1067 */
aabd5ce8 1068 if (error == 0 && (nflags & NLC_EXCL) && ncp->nc_vp != NULL)
3a907475
MD
1069 error = EEXIST;
1070
1071 /*
1072 * Get the vnode attributes so we can do the rest of our checks.
1073 *
1074 * NOTE: We only call naccess_va() if the target exists.
1075 */
21739618 1076 if (error == 0) {
28623bf9 1077 error = cache_vget(nch, cred, LK_SHARED, &vp);
21739618 1078 if (error == ENOENT) {
3a907475
MD
1079 /*
1080 * Silently zero-out ENOENT if creating or renaming
1081 * (rename target). It isn't an error.
1082 */
1083 if (nflags & (NLC_CREATE | NLC_RENAME_DST))
21739618
MD
1084 error = 0;
1085 } else if (error == 0) {
3a907475
MD
1086 /*
1087 * Get the vnode attributes and check for illegal O_TRUNC
1088 * requests and read-only mounts.
1089 *
1090 * NOTE: You can still open devices on read-only mounts for
1091 * writing.
1092 *
1093 * NOTE: creates/deletes/renames are handled by the NLC_WRITE
1094 * check on the parent directory above.
1095 *
1096 * XXX cache the va in the namecache or in the vnode
1097 */
1098 error = VOP_GETATTR(vp, &va);
1099 if (error == 0 && (nflags & NLC_TRUNCATE)) {
1100 switch(va.va_type) {
1101 case VREG:
1102 case VDATABASE:
1103 case VCHR:
1104 case VBLK:
c9c08c19 1105 case VFIFO:
3a907475 1106 break;
d7c75c7a
MD
1107 case VDIR:
1108 error = EISDIR;
1109 break;
3a907475
MD
1110 default:
1111 error = EINVAL;
1112 break;
1113 }
1114 }
1115 if (error == 0 && (nflags & NLC_WRITE) && vp->v_mount &&
1116 (vp->v_mount->mnt_flag & MNT_RDONLY)
1117 ) {
1118 switch(va.va_type) {
1119 case VDIR:
1120 case VLNK:
1121 case VREG:
1122 case VDATABASE:
1123 error = EROFS;
1124 break;
1125 default:
1126 break;
21739618
MD
1127 }
1128 }
1129 vput(vp);
dae8d54f 1130
3a907475
MD
1131 /*
1132 * Check permissions based on file attributes. The passed
1133 * flags (*nflagsp) are modified with feedback based on
1134 * special attributes and requirements.
1135 */
dae8d54f
MD
1136 if (error == 0) {
1137 /*
3a907475 1138 * Adjust the returned (*nflagsp) if non-NULL.
dae8d54f 1139 */
3a907475
MD
1140 if (nflagsp) {
1141 if ((va.va_mode & VSVTX) && va.va_uid != cred->cr_uid)
1142 *nflagsp |= NLC_STICKY;
1143 if (va.va_flags & APPEND)
1144 *nflagsp |= NLC_APPENDONLY;
1145 if (va.va_flags & IMMUTABLE)
1146 *nflagsp |= NLC_IMMUTABLE;
1147 }
dae8d54f 1148
e9b56058
MD
1149 /*
1150 * Track swapcache management flags in the namecache.
aabd5ce8
MD
1151 *
1152 * Calculate the flags based on the current vattr info
1153 * and recalculate the inherited flags from the parent
1154 * (the original cache linkage may have occurred without
1155 * getattrs and thus have stale flags).
e9b56058 1156 */
aabd5ce8 1157 cflags = 0;
e9b56058 1158 if (va.va_flags & SF_NOCACHE)
aabd5ce8 1159 cflags |= NCF_SF_NOCACHE;
e9b56058 1160 if (va.va_flags & UF_CACHE)
aabd5ce8
MD
1161 cflags |= NCF_UF_CACHE;
1162 if (ncp->nc_parent) {
1163 if (ncp->nc_parent->nc_flag &
1164 (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) {
1165 cflags |= NCF_SF_PNOCACHE;
1166 }
1167 if (ncp->nc_parent->nc_flag &
1168 (NCF_UF_CACHE | NCF_UF_PCACHE)) {
1169 cflags |= NCF_UF_PCACHE;
1170 }
1171 }
79fd1696
MD
1172
1173 /*
1174 * XXX we're not supposed to update nc_flag when holding
1175 * a shared lock.
1176 */
1177 atomic_clear_short(&ncp->nc_flag,
1178 (NCF_SF_NOCACHE | NCF_UF_CACHE |
1179 NCF_SF_PNOCACHE | NCF_UF_PCACHE) & ~cflags);
1180 atomic_set_short(&ncp->nc_flag, cflags);
e9b56058 1181
dae8d54f
MD
1182 /*
1183 * Process general access.
1184 */
3a907475 1185 error = naccess_va(&va, nflags, cred);
dae8d54f 1186 }
21739618
MD
1187 }
1188 }
1189 return(error);
1190}
1191
1192/*
1193 * Check the requested access against the given vattr using cred.
1194 */
1195int
3a907475 1196naccess_va(struct vattr *va, int nflags, struct ucred *cred)
21739618
MD
1197{
1198 int i;
3a907475 1199 int vmode;
21739618
MD
1200
1201 /*
3a907475
MD
1202 * Test the immutable bit. Creations, deletions, renames (source
1203 * or destination) are not allowed. chown/chmod/other is also not
1204 * allowed but is handled by SETATTR. Hardlinks to the immutable
1205 * file are allowed.
1206 *
1207 * If the directory is set to immutable then creations, deletions,
1208 * renames (source or dest) and hardlinks to files within the directory
1209 * are not allowed, and regular files opened through the directory may
1210 * not be written to or truncated (unless a special device).
945b476a 1211 *
3a907475
MD
1212 * NOTE! New hardlinks to immutable files work but new hardlinks to
1213 * files, immutable or not, sitting inside an immutable directory are
1214 * not allowed. As always if the file is hardlinked via some other
1215 * path additional hardlinks may be possible even if the file is marked
1216 * immutable. The sysop needs to create a closure by checking the hard
1217 * link count. Once closure is achieved you are good, and security
1218 * scripts should check link counts anyway.
1219 *
1220 * Writes and truncations are only allowed on special devices.
21739618 1221 */
3a907475
MD
1222 if ((va->va_flags & IMMUTABLE) || (nflags & NLC_IMMUTABLE)) {
1223 if ((nflags & NLC_IMMUTABLE) && (nflags & NLC_HLINK))
1224 return (EPERM);
1225 if (nflags & (NLC_CREATE | NLC_DELETE |
1226 NLC_RENAME_SRC | NLC_RENAME_DST)) {
1227 return (EPERM);
1228 }
1229 if (nflags & (NLC_WRITE | NLC_TRUNCATE)) {
1230 switch(va->va_type) {
1231 case VDIR:
d7c75c7a 1232 return (EISDIR);
3a907475
MD
1233 case VLNK:
1234 case VREG:
1235 case VDATABASE:
1236 return (EPERM);
1237 default:
1238 break;
1239 }
1240 }
1241 }
1242
1243 /*
1244 * Test the no-unlink and append-only bits for opens, rename targets,
1245 * and deletions. These bits are not tested for creations or
1246 * rename sources.
1247 *
1248 * Unlike FreeBSD we allow a file with APPEND set to be renamed.
1249 * If you do not wish this you must also set NOUNLINK.
1250 *
1251 * If the governing directory is marked APPEND-only it implies
1252 * NOUNLINK for all entries in the directory.
1253 */
1254 if (((va->va_flags & NOUNLINK) || (nflags & NLC_APPENDONLY)) &&
1255 (nflags & (NLC_DELETE | NLC_RENAME_SRC | NLC_RENAME_DST))
1256 ) {
1257 return (EPERM);
1258 }
1259
1260 /*
1261 * A file marked append-only may not be deleted but can be renamed.
1262 */
1263 if ((va->va_flags & APPEND) &&
1264 (nflags & (NLC_DELETE | NLC_RENAME_DST))
1265 ) {
1266 return (EPERM);
1267 }
1268
1269 /*
1270 * A file marked append-only which is opened for writing must also
1271 * be opened O_APPEND.
1272 */
1273 if ((va->va_flags & APPEND) && (nflags & (NLC_OPEN | NLC_TRUNCATE))) {
1274 if (nflags & NLC_TRUNCATE)
1275 return (EPERM);
1276 if ((nflags & (NLC_OPEN | NLC_WRITE)) == (NLC_OPEN | NLC_WRITE)) {
1277 if ((nflags & NLC_APPEND) == 0)
21739618
MD
1278 return (EPERM);
1279 }
1280 }
1281
1282 /*
1283 * root gets universal access
1284 */
1285 if (cred->cr_uid == 0)
1286 return(0);
1287
1288 /*
dae8d54f
MD
1289 * Check owner perms.
1290 *
3a907475 1291 * If NLC_OWN is set the owner of the file is allowed no matter when
dae8d54f 1292 * the owner-mode bits say (utimes).
21739618 1293 */
3a907475
MD
1294 vmode = 0;
1295 if (nflags & NLC_READ)
1296 vmode |= S_IRUSR;
1297 if (nflags & NLC_WRITE)
1298 vmode |= S_IWUSR;
1299 if (nflags & NLC_EXEC)
1300 vmode |= S_IXUSR;
1301
21739618 1302 if (cred->cr_uid == va->va_uid) {
3a907475 1303 if ((nflags & NLC_OWN) == 0) {
ee89633d
MD
1304 if ((vmode & va->va_mode) != vmode)
1305 return(EACCES);
1306 }
21739618
MD
1307 return(0);
1308 }
dae8d54f
MD
1309
1310 /*
3a907475
MD
1311 * If NLC_STICKY is set only the owner may delete or rename a file.
1312 * This bit is typically set on /tmp.
dae8d54f 1313 *
3a907475
MD
1314 * Note that the NLC_READ/WRITE/EXEC bits are not typically set in
1315 * the specific delete or rename case. For deletions and renames we
1316 * usually just care about directory permissions, not file permissions.
dae8d54f 1317 */
3a907475
MD
1318 if ((nflags & NLC_STICKY) &&
1319 (nflags & (NLC_RENAME_SRC | NLC_RENAME_DST | NLC_DELETE))) {
dae8d54f 1320 return(EACCES);
3a907475 1321 }
dae8d54f
MD
1322
1323 /*
1324 * Check group perms
1325 */
21739618
MD
1326 vmode >>= 3;
1327 for (i = 0; i < cred->cr_ngroups; ++i) {
1328 if (va->va_gid == cred->cr_groups[i]) {
1329 if ((vmode & va->va_mode) != vmode)
1330 return(EACCES);
1331 return(0);
1332 }
1333 }
1334
dae8d54f
MD
1335 /*
1336 * Check world perms
1337 */
21739618
MD
1338 vmode >>= 3;
1339 if ((vmode & va->va_mode) != vmode)
1340 return(EACCES);
1341 return(0);
1342}
1343