1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #include <sys/cdefs.h>
38 #include "opt_capsicum.h"
39 #include "opt_ktrace.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/dirent.h>
44 #include <sys/kernel.h>
45 #include <sys/capsicum.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/namei.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
53 #include <sys/filedesc.h>
54 #include <sys/proc.h>
55 #include <sys/sdt.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #ifdef KTRACE
59 #include <sys/ktrace.h>
60 #endif
61 #ifdef INVARIANTS
62 #include <machine/_inttypes.h>
63 #endif
64
65 #include <security/audit/audit.h>
66 #include <security/mac/mac_framework.h>
67
68 #include <vm/uma.h>
69
70 #ifdef INVARIANTS
71 static void NDVALIDATE_impl(struct nameidata *, int);
72 #define NDVALIDATE(ndp) NDVALIDATE_impl(ndp, __LINE__)
73 #else
74 #define NDVALIDATE(ndp)
75 #endif
76
77 /*
78 * Prepare namei() to restart. Reset components to its original state and set
79 * ISRESTARTED flag which signals the underlying lookup code to change the root
80 * from ABI root to actual root and prevents a further restarts.
81 */
82 #define NDRESTART(ndp) do { \
83 NDREINIT_DBG(ndp); \
84 ndp->ni_resflags = 0; \
85 ndp->ni_cnd.cn_flags &= ~NAMEI_INTERNAL_FLAGS; \
86 ndp->ni_cnd.cn_flags |= ISRESTARTED; \
87 } while (0)
88
89 #ifdef KTRACE
90 #define NIKTRCAPFAIL(path) ktrcapfail(CAPFAIL_NAMEI, (path))
91 #else
92 #define NIKTRCAPFAIL(path)
93 #endif
94
95 #define NI_CAP_VIOLATION(ndp, path) do { \
96 NIKTRCAPFAIL(path); \
97 (ndp)->ni_lcf &= ~NI_LCF_KTR_FLAGS; \
98 } while (0)
99
100 SDT_PROVIDER_DEFINE(vfs);
101 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
102 "unsigned long", "bool");
103 SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool",
104 "struct nameidata");
105
106 /* Allocation zone for namei. */
107 uma_zone_t namei_zone;
108
109 /* Placeholder vnode for mp traversal. */
110 static struct vnode *vp_crossmp;
111
112 static int
crossmp_vop_islocked(struct vop_islocked_args * ap)113 crossmp_vop_islocked(struct vop_islocked_args *ap)
114 {
115
116 return (LK_SHARED);
117 }
118
119 static int
crossmp_vop_lock1(struct vop_lock1_args * ap)120 crossmp_vop_lock1(struct vop_lock1_args *ap)
121 {
122 struct vnode *vp;
123 struct lock *lk __diagused;
124 int flags;
125
126 vp = ap->a_vp;
127 lk = vp->v_vnlock;
128 flags = ap->a_flags;
129
130 KASSERT((flags & (LK_SHARED | LK_NOWAIT)) == (LK_SHARED | LK_NOWAIT),
131 ("%s: invalid lock request 0x%x for crossmp", __func__, flags));
132
133 if ((flags & LK_INTERLOCK) != 0)
134 VI_UNLOCK(vp);
135 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, ap->a_line);
136 return (0);
137 }
138
139 static int
crossmp_vop_unlock(struct vop_unlock_args * ap)140 crossmp_vop_unlock(struct vop_unlock_args *ap)
141 {
142 struct vnode *vp;
143 struct lock *lk __diagused;
144
145 vp = ap->a_vp;
146 lk = vp->v_vnlock;
147
148 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
149 LOCK_LINE);
150 return (0);
151 }
152
153 static struct vop_vector crossmp_vnodeops = {
154 .vop_default = &default_vnodeops,
155 .vop_islocked = crossmp_vop_islocked,
156 .vop_lock1 = crossmp_vop_lock1,
157 .vop_unlock = crossmp_vop_unlock,
158 };
159 /*
160 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
161 * gets allocated early. See nameiinit for the direct call below.
162 */
163
164 struct nameicap_tracker {
165 struct vnode *dp;
166 TAILQ_ENTRY(nameicap_tracker) nm_link;
167 };
168
169 /* Zone for cap mode tracker elements used for dotdot capability checks. */
170 MALLOC_DEFINE(M_NAMEITRACKER, "namei_tracker", "namei tracking for dotdot");
171
172 static void
nameiinit(void * dummy __unused)173 nameiinit(void *dummy __unused)
174 {
175
176 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
177 UMA_ALIGN_PTR, 0);
178 vfs_vector_op_register(&crossmp_vnodeops);
179 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
180 vp_crossmp->v_state = VSTATE_CONSTRUCTED;
181 vp_crossmp->v_irflag |= VIRF_CROSSMP;
182 }
183 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
184
185 static int lookup_cap_dotdot = 1;
186 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
187 &lookup_cap_dotdot, 0,
188 "enables \"..\" components in path lookup in capability mode");
189 static int lookup_cap_dotdot_nonlocal = 1;
190 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
191 &lookup_cap_dotdot_nonlocal, 0,
192 "enables \"..\" components in path lookup in capability mode "
193 "on non-local mount");
194
195 static void
nameicap_tracker_add(struct nameidata * ndp,struct vnode * dp)196 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
197 {
198 struct nameicap_tracker *nt;
199
200 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
201 return;
202 nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head);
203 if (nt != NULL && nt->dp == dp)
204 return;
205 nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK);
206 vhold(dp);
207 nt->dp = dp;
208 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
209 }
210
211 static void
nameicap_cleanup_from(struct nameidata * ndp,struct nameicap_tracker * first)212 nameicap_cleanup_from(struct nameidata *ndp, struct nameicap_tracker *first)
213 {
214 struct nameicap_tracker *nt, *nt1;
215
216 nt = first;
217 TAILQ_FOREACH_FROM_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
218 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
219 vdrop(nt->dp);
220 free(nt, M_NAMEITRACKER);
221 }
222 }
223
224 static void
nameicap_cleanup(struct nameidata * ndp)225 nameicap_cleanup(struct nameidata *ndp)
226 {
227 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
228 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
229 nameicap_cleanup_from(ndp, NULL);
230 }
231
232 /*
233 * For dotdot lookups in capability mode, only allow the component
234 * lookup to succeed if the resulting directory was already traversed
235 * during the operation. This catches situations where already
236 * traversed directory is moved to different parent, and then we walk
237 * over it with dotdots.
238 *
239 * Also allow to force failure of dotdot lookups for non-local
240 * filesystems, where external agents might assist local lookups to
241 * escape the compartment.
242 */
243 static int
nameicap_check_dotdot(struct nameidata * ndp,struct vnode * dp)244 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
245 {
246 struct nameicap_tracker *nt;
247 struct mount *mp;
248
249 if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf &
250 NI_LCF_STRICTREL) == 0)
251 return (0);
252 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL_KTR |
253 NI_LCF_CAP_DOTDOT_KTR)) == NI_LCF_STRICTREL_KTR))
254 NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
255 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0)
256 return (ENOTCAPABLE);
257 mp = dp->v_mount;
258 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
259 (mp->mnt_flag & MNT_LOCAL) == 0)
260 goto capfail;
261 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
262 nm_link) {
263 if (dp == nt->dp) {
264 nt = TAILQ_NEXT(nt, nm_link);
265 if (nt != NULL)
266 nameicap_cleanup_from(ndp, nt);
267 return (0);
268 }
269 }
270
271 capfail:
272 if (__predict_false((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0))
273 NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
274 return (ENOTCAPABLE);
275 }
276
277 static void
namei_cleanup_cnp(struct componentname * cnp)278 namei_cleanup_cnp(struct componentname *cnp)
279 {
280
281 uma_zfree(namei_zone, cnp->cn_pnbuf);
282 cnp->cn_pnbuf = NULL;
283 cnp->cn_nameptr = NULL;
284 }
285
286 static int
namei_handle_root(struct nameidata * ndp,struct vnode ** dpp)287 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
288 {
289 struct componentname *cnp;
290
291 cnp = &ndp->ni_cnd;
292 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL |
293 NI_LCF_STRICTREL_KTR)) != 0)) {
294 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
295 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
296 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
297 return (ENOTCAPABLE);
298 }
299 while (*(cnp->cn_nameptr) == '/') {
300 cnp->cn_nameptr++;
301 ndp->ni_pathlen--;
302 }
303 *dpp = ndp->ni_rootdir;
304 vrefact(*dpp);
305 return (0);
306 }
307
308 static int
namei_setup(struct nameidata * ndp,struct vnode ** dpp,struct pwd ** pwdp)309 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
310 {
311 struct componentname *cnp;
312 struct thread *td;
313 struct pwd *pwd;
314 int error;
315 bool startdir_used;
316
317 cnp = &ndp->ni_cnd;
318 td = curthread;
319
320 startdir_used = false;
321 *pwdp = NULL;
322 *dpp = NULL;
323
324 #ifdef CAPABILITY_MODE
325 /*
326 * In capability mode, lookups must be restricted to happen in
327 * the subtree with the root specified by the file descriptor:
328 * - The root must be real file descriptor, not the pseudo-descriptor
329 * AT_FDCWD.
330 * - The passed path must be relative and not absolute.
331 * - If lookup_cap_dotdot is disabled, path must not contain the
332 * '..' components.
333 * - If lookup_cap_dotdot is enabled, we verify that all '..'
334 * components lookups result in the directories which were
335 * previously walked by us, which prevents an escape from
336 * the relative root.
337 */
338 if ((cnp->cn_flags & NOCAPCHECK) == 0) {
339 if (CAP_TRACING(td)) {
340 ndp->ni_lcf |= NI_LCF_STRICTREL_KTR;
341 if (ndp->ni_dirfd == AT_FDCWD)
342 NI_CAP_VIOLATION(ndp, "AT_FDCWD");
343 }
344 if (IN_CAPABILITY_MODE(td)) {
345 ndp->ni_lcf |= NI_LCF_STRICTREL;
346 ndp->ni_resflags |= NIRES_STRICTREL;
347 if (ndp->ni_dirfd == AT_FDCWD)
348 return (ECAPMODE);
349 }
350 }
351 #endif
352 error = 0;
353
354 /*
355 * Get starting point for the translation.
356 */
357 pwd = pwd_hold(td);
358 /*
359 * The reference on ni_rootdir is acquired in the block below to avoid
360 * back-to-back atomics for absolute lookups.
361 */
362 namei_setup_rootdir(ndp, cnp, pwd);
363 ndp->ni_topdir = pwd->pwd_jdir;
364
365 if (cnp->cn_pnbuf[0] == '/') {
366 ndp->ni_resflags |= NIRES_ABS;
367 error = namei_handle_root(ndp, dpp);
368 } else {
369 if (ndp->ni_startdir != NULL) {
370 *dpp = ndp->ni_startdir;
371 startdir_used = true;
372 } else if (ndp->ni_dirfd == AT_FDCWD) {
373 *dpp = pwd->pwd_cdir;
374 vrefact(*dpp);
375 } else {
376 if (cnp->cn_flags & AUDITVNODE1)
377 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
378 if (cnp->cn_flags & AUDITVNODE2)
379 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
380
381 error = fgetvp_lookup(ndp, dpp);
382 }
383 if (error == 0 && (*dpp)->v_type != VDIR &&
384 (cnp->cn_pnbuf[0] != '\0' ||
385 (cnp->cn_flags & EMPTYPATH) == 0))
386 error = ENOTDIR;
387 }
388 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
389 if (cnp->cn_pnbuf[0] == '/') {
390 error = ENOTCAPABLE;
391 } else if ((ndp->ni_lcf & NI_LCF_STRICTREL) == 0) {
392 ndp->ni_lcf |= NI_LCF_STRICTREL |
393 NI_LCF_CAP_DOTDOT;
394 }
395 }
396
397 /*
398 * If we are auditing the kernel pathname, save the user pathname.
399 */
400 if (AUDITING_TD(td)) {
401 if (cnp->cn_flags & AUDITVNODE1)
402 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
403 if (cnp->cn_flags & AUDITVNODE2)
404 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
405 }
406 if (ndp->ni_startdir != NULL && !startdir_used)
407 vrele(ndp->ni_startdir);
408 if (error != 0) {
409 if (*dpp != NULL)
410 vrele(*dpp);
411 pwd_drop(pwd);
412 return (error);
413 }
414 if (lookup_cap_dotdot != 0) {
415 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
416 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT_KTR;
417 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
418 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
419 }
420 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
421 cnp->cn_flags, false);
422 *pwdp = pwd;
423 return (0);
424 }
425
426 static int
namei_getpath(struct nameidata * ndp)427 namei_getpath(struct nameidata *ndp)
428 {
429 struct componentname *cnp;
430 int error;
431
432 cnp = &ndp->ni_cnd;
433
434 /*
435 * Get a buffer for the name to be translated, and copy the
436 * name into the buffer.
437 */
438 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
439 if (ndp->ni_segflg == UIO_SYSSPACE) {
440 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
441 &ndp->ni_pathlen);
442 } else {
443 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
444 &ndp->ni_pathlen);
445 }
446
447 return (error);
448 }
449
450 static int
namei_emptypath(struct nameidata * ndp)451 namei_emptypath(struct nameidata *ndp)
452 {
453 struct componentname *cnp;
454 struct pwd *pwd;
455 struct vnode *dp;
456 int error;
457
458 cnp = &ndp->ni_cnd;
459 MPASS(*cnp->cn_pnbuf == '\0');
460 MPASS((cnp->cn_flags & EMPTYPATH) != 0);
461 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
462
463 ndp->ni_resflags |= NIRES_EMPTYPATH;
464 error = namei_setup(ndp, &dp, &pwd);
465 if (error != 0) {
466 goto errout;
467 }
468
469 /*
470 * Usecount on dp already provided by namei_setup.
471 */
472 ndp->ni_vp = dp;
473 pwd_drop(pwd);
474 NDVALIDATE(ndp);
475 if ((cnp->cn_flags & LOCKLEAF) != 0) {
476 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ?
477 LK_SHARED : LK_EXCLUSIVE);
478 if (VN_IS_DOOMED(dp)) {
479 vput(dp);
480 error = ENOENT;
481 goto errout;
482 }
483 }
484 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp);
485 return (0);
486
487 errout:
488 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
489 namei_cleanup_cnp(cnp);
490 return (error);
491 }
492
493 static int __noinline
namei_follow_link(struct nameidata * ndp)494 namei_follow_link(struct nameidata *ndp)
495 {
496 char *cp;
497 struct iovec aiov;
498 struct uio auio;
499 struct componentname *cnp;
500 struct thread *td;
501 int error, linklen;
502
503 error = 0;
504 cnp = &ndp->ni_cnd;
505 td = curthread;
506
507 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
508 error = ELOOP;
509 goto out;
510 }
511 #ifdef MAC
512 if ((cnp->cn_flags & NOMACCHECK) == 0) {
513 error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp);
514 if (error != 0)
515 goto out;
516 }
517 #endif
518 if (ndp->ni_pathlen > 1)
519 cp = uma_zalloc(namei_zone, M_WAITOK);
520 else
521 cp = cnp->cn_pnbuf;
522 aiov.iov_base = cp;
523 aiov.iov_len = MAXPATHLEN;
524 auio.uio_iov = &aiov;
525 auio.uio_iovcnt = 1;
526 auio.uio_offset = 0;
527 auio.uio_rw = UIO_READ;
528 auio.uio_segflg = UIO_SYSSPACE;
529 auio.uio_td = td;
530 auio.uio_resid = MAXPATHLEN;
531 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
532 if (error != 0) {
533 if (ndp->ni_pathlen > 1)
534 uma_zfree(namei_zone, cp);
535 goto out;
536 }
537 linklen = MAXPATHLEN - auio.uio_resid;
538 if (linklen == 0) {
539 if (ndp->ni_pathlen > 1)
540 uma_zfree(namei_zone, cp);
541 error = ENOENT;
542 goto out;
543 }
544 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
545 if (ndp->ni_pathlen > 1)
546 uma_zfree(namei_zone, cp);
547 error = ENAMETOOLONG;
548 goto out;
549 }
550 if (ndp->ni_pathlen > 1) {
551 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
552 uma_zfree(namei_zone, cnp->cn_pnbuf);
553 cnp->cn_pnbuf = cp;
554 } else
555 cnp->cn_pnbuf[linklen] = '\0';
556 ndp->ni_pathlen += linklen;
557 out:
558 return (error);
559 }
560
561 /*
562 * Convert a pathname into a pointer to a locked vnode.
563 *
564 * The FOLLOW flag is set when symbolic links are to be followed
565 * when they occur at the end of the name translation process.
566 * Symbolic links are always followed for all other pathname
567 * components other than the last.
568 *
569 * The segflg defines whether the name is to be copied from user
570 * space or kernel space.
571 *
572 * Overall outline of namei:
573 *
574 * copy in name
575 * get starting directory
576 * while (!done && !error) {
577 * call lookup to search path.
578 * if symbolic link, massage name in buffer and continue
579 * }
580 */
581 int
namei(struct nameidata * ndp)582 namei(struct nameidata *ndp)
583 {
584 struct vnode *dp; /* the directory we are searching */
585 struct componentname *cnp;
586 struct thread *td;
587 struct pwd *pwd;
588 int error;
589 enum cache_fpl_status status;
590
591 cnp = &ndp->ni_cnd;
592 td = curthread;
593 #ifdef INVARIANTS
594 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0,
595 ("%s: repeated call to namei without NDREINIT", __func__));
596 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED,
597 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags));
598 ndp->ni_debugflags |= NAMEI_DBG_CALLED;
599 if (ndp->ni_startdir != NULL)
600 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR;
601 if (cnp->cn_flags & FAILIFEXISTS) {
602 KASSERT(cnp->cn_nameiop == CREATE,
603 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop));
604 /*
605 * The limitation below is to restrict hairy corner cases.
606 */
607 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT,
608 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF",
609 __func__));
610 }
611 #endif
612 ndp->ni_cnd.cn_cred = td->td_ucred;
613 KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n",
614 __func__, ndp->ni_resflags));
615 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
616 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
617 ("namei: unexpected flags: %" PRIx64 "\n",
618 cnp->cn_flags & NAMEI_INTERNAL_FLAGS));
619 if (cnp->cn_flags & NOCACHE)
620 KASSERT(cnp->cn_nameiop != LOOKUP,
621 ("%s: NOCACHE passed with LOOKUP", __func__));
622 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
623 ndp->ni_startdir->v_type == VBAD);
624
625 restart:
626 ndp->ni_lcf = 0;
627 ndp->ni_loopcnt = 0;
628 ndp->ni_vp = NULL;
629
630 error = namei_getpath(ndp);
631 if (__predict_false(error != 0)) {
632 namei_cleanup_cnp(cnp);
633 SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
634 false, ndp);
635 return (error);
636 }
637
638 cnp->cn_nameptr = cnp->cn_pnbuf;
639
640 #ifdef KTRACE
641 if (KTRPOINT(td, KTR_NAMEI)) {
642 ktrnamei(cnp->cn_pnbuf);
643 }
644 #endif
645 TSNAMEI(curthread->td_proc->p_pid, cnp->cn_pnbuf);
646
647 /*
648 * First try looking up the target without locking any vnodes.
649 *
650 * We may need to start from scratch or pick up where it left off.
651 */
652 error = cache_fplookup(ndp, &status, &pwd);
653 switch (status) {
654 case CACHE_FPL_STATUS_UNSET:
655 __assert_unreachable();
656 break;
657 case CACHE_FPL_STATUS_HANDLED:
658 if (error == 0)
659 NDVALIDATE(ndp);
660 else if (__predict_false(pwd->pwd_adir != pwd->pwd_rdir &&
661 (cnp->cn_flags & ISRESTARTED) == 0)) {
662 namei_cleanup_cnp(cnp);
663 NDRESTART(ndp);
664 goto restart;
665 }
666 return (error);
667 case CACHE_FPL_STATUS_PARTIAL:
668 TAILQ_INIT(&ndp->ni_cap_tracker);
669 dp = ndp->ni_startdir;
670 break;
671 case CACHE_FPL_STATUS_DESTROYED:
672 ndp->ni_loopcnt = 0;
673 error = namei_getpath(ndp);
674 if (__predict_false(error != 0)) {
675 namei_cleanup_cnp(cnp);
676 return (error);
677 }
678 cnp->cn_nameptr = cnp->cn_pnbuf;
679 /* FALLTHROUGH */
680 case CACHE_FPL_STATUS_ABORTED:
681 TAILQ_INIT(&ndp->ni_cap_tracker);
682 MPASS(ndp->ni_lcf == 0);
683 if (*cnp->cn_pnbuf == '\0') {
684 if ((cnp->cn_flags & EMPTYPATH) != 0) {
685 return (namei_emptypath(ndp));
686 }
687 namei_cleanup_cnp(cnp);
688 SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
689 false, ndp);
690 return (ENOENT);
691 }
692 error = namei_setup(ndp, &dp, &pwd);
693 if (error != 0) {
694 namei_cleanup_cnp(cnp);
695 return (error);
696 }
697 break;
698 }
699
700 /*
701 * Locked lookup.
702 */
703 for (;;) {
704 ndp->ni_startdir = dp;
705 error = vfs_lookup(ndp);
706 if (error != 0) {
707 if (__predict_false(pwd->pwd_adir != pwd->pwd_rdir &&
708 error == ENOENT &&
709 (cnp->cn_flags & ISRESTARTED) == 0)) {
710 nameicap_cleanup(ndp);
711 pwd_drop(pwd);
712 namei_cleanup_cnp(cnp);
713 NDRESTART(ndp);
714 goto restart;
715 } else
716 goto out;
717 }
718
719 /*
720 * If not a symbolic link, we're done.
721 */
722 if ((cnp->cn_flags & ISSYMLINK) == 0) {
723 SDT_PROBE4(vfs, namei, lookup, return, error,
724 ndp->ni_vp, false, ndp);
725 nameicap_cleanup(ndp);
726 pwd_drop(pwd);
727 NDVALIDATE(ndp);
728 return (0);
729 }
730 error = namei_follow_link(ndp);
731 if (error != 0)
732 break;
733 vput(ndp->ni_vp);
734 dp = ndp->ni_dvp;
735 /*
736 * Check if root directory should replace current directory.
737 */
738 cnp->cn_nameptr = cnp->cn_pnbuf;
739 if (*(cnp->cn_nameptr) == '/') {
740 /*
741 * Reset the lookup to start from the real root without
742 * origin path name reloading.
743 */
744 if (__predict_false(ndp->ni_rootdir != pwd->pwd_rdir)) {
745 cnp->cn_flags |= ISRESTARTED;
746 ndp->ni_rootdir = pwd->pwd_rdir;
747 }
748 vrele(dp);
749 error = namei_handle_root(ndp, &dp);
750 if (error != 0)
751 goto out;
752 }
753 }
754 vput(ndp->ni_vp);
755 ndp->ni_vp = NULL;
756 vrele(ndp->ni_dvp);
757 out:
758 MPASS(error != 0);
759 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
760 namei_cleanup_cnp(cnp);
761 nameicap_cleanup(ndp);
762 pwd_drop(pwd);
763 return (error);
764 }
765
766 static int
enforce_lkflags(struct mount * mp,int lkflags)767 enforce_lkflags(struct mount *mp, int lkflags)
768 {
769
770 if (mp == NULL || ((lkflags & LK_SHARED) &&
771 !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
772 lkflags &= ~LK_SHARED;
773 lkflags |= LK_EXCLUSIVE;
774 }
775 lkflags |= LK_NODDLKTREAT;
776 return (lkflags);
777 }
778
779 static __inline int
needs_exclusive_leaf(struct mount * mp,int flags)780 needs_exclusive_leaf(struct mount *mp, int flags)
781 {
782
783 /*
784 * Intermediate nodes can use shared locks, we only need to
785 * force an exclusive lock for leaf nodes.
786 */
787 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
788 return (0);
789
790 /* Always use exclusive locks if LOCKSHARED isn't set. */
791 if (!(flags & LOCKSHARED))
792 return (1);
793
794 /*
795 * For lookups during open(), if the mount point supports
796 * extended shared operations, then use a shared lock for the
797 * leaf node, otherwise use an exclusive lock.
798 */
799 if ((flags & ISOPEN) != 0)
800 return (!MNT_EXTENDED_SHARED(mp));
801
802 /*
803 * Lookup requests outside of open() that specify LOCKSHARED
804 * only need a shared lock on the leaf vnode.
805 */
806 return (0);
807 }
808
809 /*
810 * Various filesystems expect to be able to copy a name component with length
811 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make
812 * sure that these are the same size.
813 */
814 _Static_assert(MAXNAMLEN == NAME_MAX,
815 "MAXNAMLEN and NAME_MAX have different values");
816
817 static int __noinline
vfs_lookup_degenerate(struct nameidata * ndp,struct vnode * dp,int wantparent)818 vfs_lookup_degenerate(struct nameidata *ndp, struct vnode *dp, int wantparent)
819 {
820 struct componentname *cnp;
821 struct mount *mp;
822 int error;
823
824 cnp = &ndp->ni_cnd;
825
826 cnp->cn_flags |= ISLASTCN;
827
828 mp = atomic_load_ptr(&dp->v_mount);
829 if (needs_exclusive_leaf(mp, cnp->cn_flags)) {
830 cnp->cn_lkflags &= ~LK_SHARED;
831 cnp->cn_lkflags |= LK_EXCLUSIVE;
832 }
833
834 vn_lock(dp, enforce_lkflags(mp, cnp->cn_lkflags | LK_RETRY));
835
836 if (dp->v_type != VDIR) {
837 error = ENOTDIR;
838 goto bad;
839 }
840 if (cnp->cn_nameiop != LOOKUP) {
841 error = EISDIR;
842 goto bad;
843 }
844 if (wantparent) {
845 ndp->ni_dvp = dp;
846 VREF(dp);
847 }
848 ndp->ni_vp = dp;
849 cnp->cn_namelen = 0;
850
851 if (cnp->cn_flags & AUDITVNODE1)
852 AUDIT_ARG_VNODE1(dp);
853 else if (cnp->cn_flags & AUDITVNODE2)
854 AUDIT_ARG_VNODE2(dp);
855
856 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
857 VOP_UNLOCK(dp);
858 return (0);
859 bad:
860 VOP_UNLOCK(dp);
861 return (error);
862 }
863
864 /*
865 * FAILIFEXISTS handling.
866 *
867 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange
868 * behaviour of leaving the vnode unlocked if the target is the same
869 * vnode as the parent.
870 */
871 static int __noinline
vfs_lookup_failifexists(struct nameidata * ndp)872 vfs_lookup_failifexists(struct nameidata *ndp)
873 {
874 struct componentname *cnp __diagused;
875
876 cnp = &ndp->ni_cnd;
877
878 MPASS((cnp->cn_flags & ISSYMLINK) == 0);
879 if (ndp->ni_vp == ndp->ni_dvp)
880 vrele(ndp->ni_dvp);
881 else
882 vput(ndp->ni_dvp);
883 vrele(ndp->ni_vp);
884 ndp->ni_dvp = NULL;
885 ndp->ni_vp = NULL;
886 NDFREE_PNBUF(ndp);
887 return (EEXIST);
888 }
889
890 static int __noinline
vfs_lookup_cross_mount(struct nameidata * ndp)891 vfs_lookup_cross_mount(struct nameidata *ndp)
892 {
893 struct componentname *cnp;
894 struct mount *mp;
895 struct vnode *dp, *tdp;
896 int error, crosslkflags;
897 bool crosslock;
898
899 cnp = &ndp->ni_cnd;
900 dp = ndp->ni_vp;
901
902 /*
903 * The vnode has been mounted on, find the root of the mounted
904 * filesystem.
905 */
906 do {
907 mp = dp->v_mountedhere;
908 ASSERT_VOP_LOCKED(dp, __func__);
909 VNPASS((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 && mp != NULL, dp);
910
911 crosslock = (dp->v_vflag & VV_CROSSLOCK) != 0;
912 crosslkflags = enforce_lkflags(mp, cnp->cn_lkflags);
913 if (__predict_false(crosslock)) {
914 /*
915 * We are going to be holding the vnode lock, which
916 * in this case is shared by the root vnode of the
917 * filesystem mounted at mp, across the call to
918 * VFS_ROOT(). Make the situation clear to the
919 * filesystem by passing LK_CANRECURSE if the
920 * lock is held exclusive, or by clearinng
921 * LK_NODDLKTREAT to allow recursion on the shared
922 * lock in the presence of an exclusive waiter.
923 */
924 if (VOP_ISLOCKED(dp) == LK_EXCLUSIVE) {
925 crosslkflags &= ~LK_SHARED;
926 crosslkflags |= LK_EXCLUSIVE | LK_CANRECURSE;
927 } else if ((crosslkflags & LK_EXCLUSIVE) != 0) {
928 error = vn_lock(dp, LK_UPGRADE);
929 if (error != 0) {
930 MPASS(error == ENOENT);
931 vrele(dp);
932 if (dp != ndp->ni_dvp)
933 vput(ndp->ni_dvp);
934 else
935 vrele(ndp->ni_dvp);
936 break;
937 }
938 if (dp->v_mountedhere != mp) {
939 /*
940 * Note that we rely on the
941 * VIRF_MOUNTPOINT loop condition to
942 * ensure we stop iterating if dp is
943 * no longer a mountpoint at all.
944 */
945 continue;
946 }
947 } else
948 crosslkflags &= ~LK_NODDLKTREAT;
949 }
950 if (vfs_busy(mp, 0) != 0)
951 continue;
952 if (__predict_true(!crosslock))
953 vput(dp);
954 if (dp != ndp->ni_dvp)
955 vput(ndp->ni_dvp);
956 else
957 vrele(ndp->ni_dvp);
958 vrefact(vp_crossmp);
959 ndp->ni_dvp = vp_crossmp;
960 error = VFS_ROOT(mp, crosslkflags, &tdp);
961 vfs_unbusy(mp);
962 if (__predict_false(crosslock))
963 vput(dp);
964 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
965 panic("vp_crossmp exclusively locked or reclaimed");
966 if (error != 0)
967 break;
968 ndp->ni_vp = dp = tdp;
969 } while ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0);
970
971 return (error);
972 }
973
974 /*
975 * Search a pathname.
976 * This is a very central and rather complicated routine.
977 *
978 * The pathname is pointed to by cn_nameptr and is of length ni_pathlen.
979 * The starting directory is taken from ni_startdir. The pathname is
980 * descended until done, or a symbolic link is encountered. The cn_flags
981 * has ISLASTCN or'ed if the path is completed or ISSYMLINK or'ed if a
982 * symbolic link needing interpretation is encountered.
983 *
984 * The cn_nameiop is LOOKUP, CREATE, RENAME, or DELETE depending on
985 * whether the name is to be looked up, created, renamed, or deleted.
986 * When CREATE, RENAME, or DELETE is specified, information usable in
987 * creating, renaming, or deleting a directory entry may be calculated.
988 * If cn_flags has LOCKPARENT or'ed into it, the parent directory is returned
989 * locked. If it has WANTPARENT or'ed into it, the parent directory is
990 * returned unlocked. Otherwise the parent directory is not returned. If
991 * the target of the pathname exists and LOCKLEAF is or'ed into the cn_flags
992 * the target is returned locked, otherwise it is returned unlocked.
993 *
994 * Overall outline of lookup:
995 *
996 * handle degenerate case where name is null string
997 *
998 * dirloop:
999 * identify next component of name at ndp->ni_cnd.cn_nameptr
1000 * handle .. special cases related to capabilities, chroot, jail
1001 * if .. and crossing mount points and on mounted filesys, find parent
1002 * call VOP_LOOKUP routine for next component name
1003 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
1004 * component vnode returned in ni_vp (if it exists), locked.
1005 * if result vnode is mounted on and crossing mount points,
1006 * find mounted on vnode
1007 * if more components of name, do next level at dirloop
1008 * if VOP_LOOKUP returns ERELOOKUP, repeat the same level at dirloop
1009 * return the answer in ni_vp, locked if LOCKLEAF set
1010 * if LOCKPARENT set, return locked parent in ni_dvp
1011 * if WANTPARENT set, return unlocked parent in ni_dvp
1012 */
1013 int
vfs_lookup(struct nameidata * ndp)1014 vfs_lookup(struct nameidata *ndp)
1015 {
1016 char *cp; /* pointer into pathname argument */
1017 char *prev_ni_next; /* saved ndp->ni_next */
1018 char *nulchar; /* location of '\0' in cn_pnbuf */
1019 char *lastchar; /* location of the last character */
1020 struct vnode *dp = NULL; /* the directory we are searching */
1021 struct vnode *tdp; /* saved dp */
1022 struct prison *pr;
1023 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
1024 int docache; /* == 0 do not cache last component */
1025 int wantparent; /* 1 => wantparent or lockparent flag */
1026 int rdonly; /* lookup read-only flag bit */
1027 int error = 0;
1028 int relookup = 0; /* do not consume the path component */
1029 struct componentname *cnp = &ndp->ni_cnd;
1030 int lkflags_save;
1031 int ni_dvp_unlocked;
1032
1033 /*
1034 * Setup: break out flag bits into variables.
1035 */
1036 ni_dvp_unlocked = 0;
1037 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
1038 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
1039 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
1040 /*
1041 * When set to zero, docache causes the last component of the
1042 * pathname to be deleted from the cache and the full lookup
1043 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
1044 * filesystems need some pre-computed values that are made
1045 * during the full lookup, for instance UFS sets dp->i_offset.
1046 *
1047 * The docache variable is set to zero when requested by the
1048 * NOCACHE flag and for all modifying operations except CREATE.
1049 */
1050 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
1051 if (cnp->cn_nameiop == DELETE ||
1052 (wantparent && cnp->cn_nameiop != CREATE &&
1053 cnp->cn_nameiop != LOOKUP))
1054 docache = 0;
1055 rdonly = cnp->cn_flags & RDONLY;
1056 cnp->cn_flags &= ~ISSYMLINK;
1057 ndp->ni_dvp = NULL;
1058
1059 cnp->cn_lkflags = LK_SHARED;
1060 dp = ndp->ni_startdir;
1061 ndp->ni_startdir = NULLVP;
1062
1063 /*
1064 * Leading slashes, if any, are supposed to be skipped by the caller.
1065 */
1066 MPASS(cnp->cn_nameptr[0] != '/');
1067
1068 /*
1069 * Check for degenerate name (e.g. / or "") which is a way of talking
1070 * about a directory, e.g. like "/." or ".".
1071 */
1072 if (__predict_false(cnp->cn_nameptr[0] == '\0')) {
1073 error = vfs_lookup_degenerate(ndp, dp, wantparent);
1074 if (error == 0)
1075 goto success_right_lock;
1076 goto bad_unlocked;
1077 }
1078
1079 /*
1080 * Nul-out trailing slashes (e.g., "foo///" -> "foo").
1081 *
1082 * This must be done before VOP_LOOKUP() because some fs's don't know
1083 * about trailing slashes. Remember if there were trailing slashes to
1084 * handle symlinks, existing non-directories and non-existing files
1085 * that won't be directories specially later.
1086 */
1087 MPASS(ndp->ni_pathlen >= 2);
1088 lastchar = &cnp->cn_nameptr[ndp->ni_pathlen - 2];
1089 if (*lastchar == '/') {
1090 while (lastchar >= cnp->cn_pnbuf) {
1091 *lastchar = '\0';
1092 lastchar--;
1093 ndp->ni_pathlen--;
1094 if (*lastchar != '/') {
1095 break;
1096 }
1097 }
1098 cnp->cn_flags |= TRAILINGSLASH;
1099 }
1100
1101 /*
1102 * We use shared locks until we hit the parent of the last cn then
1103 * we adjust based on the requesting flags.
1104 */
1105 vn_lock(dp,
1106 enforce_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
1107
1108 dirloop:
1109 /*
1110 * Search a new directory.
1111 *
1112 * The last component of the filename is left accessible via
1113 * cnp->cn_nameptr. It has to be freed with a call to NDFREE*.
1114 *
1115 * Store / as a temporary sentinel so that we only have one character
1116 * to test for. Pathnames tend to be short so this should not be
1117 * resulting in cache misses.
1118 */
1119 nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1];
1120 KASSERT(*nulchar == '\0',
1121 ("%s: expected nul at %p; string [%s]\n", __func__, nulchar,
1122 cnp->cn_pnbuf));
1123 *nulchar = '/';
1124 for (cp = cnp->cn_nameptr; *cp != '/'; cp++) {
1125 KASSERT(*cp != '\0',
1126 ("%s: encountered unexpected nul; string [%s]\n", __func__,
1127 cnp->cn_nameptr));
1128 continue;
1129 }
1130 *nulchar = '\0';
1131 cnp->cn_namelen = cp - cnp->cn_nameptr;
1132 if (__predict_false(cnp->cn_namelen > NAME_MAX)) {
1133 error = ENAMETOOLONG;
1134 goto bad;
1135 }
1136 prev_ni_pathlen = ndp->ni_pathlen;
1137 ndp->ni_pathlen -= cnp->cn_namelen;
1138 KASSERT(ndp->ni_pathlen <= PATH_MAX,
1139 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
1140 prev_ni_next = ndp->ni_next;
1141 ndp->ni_next = cp;
1142
1143 /*
1144 * Something else should be clearing this.
1145 */
1146 cnp->cn_flags &= ~(ISDOTDOT|ISLASTCN);
1147
1148 cnp->cn_flags |= MAKEENTRY;
1149 if (*cp == '\0' && docache == 0)
1150 cnp->cn_flags &= ~MAKEENTRY;
1151 if (cnp->cn_namelen == 2 &&
1152 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1153 cnp->cn_flags |= ISDOTDOT;
1154 if (*ndp->ni_next == 0) {
1155 cnp->cn_flags |= ISLASTCN;
1156
1157 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
1158 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))) {
1159 error = EINVAL;
1160 goto bad;
1161 }
1162 }
1163
1164 nameicap_tracker_add(ndp, dp);
1165
1166 /*
1167 * Make sure degenerate names don't get here, their handling was
1168 * previously found in this spot.
1169 */
1170 MPASS(cnp->cn_nameptr[0] != '\0');
1171
1172 /*
1173 * Handle "..": five special cases.
1174 * 0. If doing a capability lookup and lookup_cap_dotdot is
1175 * disabled, return ENOTCAPABLE.
1176 * 1. Return an error if this is the last component of
1177 * the name and the operation is DELETE or RENAME.
1178 * 2. If at root directory (e.g. after chroot)
1179 * or at absolute root directory
1180 * then ignore it so can't get out.
1181 * 3. If this vnode is the root of a mounted
1182 * filesystem, then replace it with the
1183 * vnode which was mounted on so we take the
1184 * .. in the other filesystem.
1185 * 4. If the vnode is the top directory of
1186 * the jail or chroot, don't let them out.
1187 * 5. If doing a capability lookup and lookup_cap_dotdot is
1188 * enabled, return ENOTCAPABLE if the lookup would escape
1189 * from the initial file descriptor directory. Checks are
1190 * done by ensuring that namei() already traversed the
1191 * result of dotdot lookup.
1192 */
1193 if (cnp->cn_flags & ISDOTDOT) {
1194 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL_KTR |
1195 NI_LCF_CAP_DOTDOT_KTR)) == NI_LCF_STRICTREL_KTR))
1196 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
1197 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL |
1198 NI_LCF_CAP_DOTDOT)) == NI_LCF_STRICTREL)) {
1199 error = ENOTCAPABLE;
1200 goto bad;
1201 }
1202 if ((cnp->cn_flags & ISLASTCN) != 0 &&
1203 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1204 error = EINVAL;
1205 goto bad;
1206 }
1207 for (;;) {
1208 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
1209 pr = pr->pr_parent)
1210 if (dp == pr->pr_root)
1211 break;
1212 bool isroot = dp == ndp->ni_rootdir ||
1213 dp == ndp->ni_topdir || dp == rootvnode ||
1214 pr != NULL;
1215 if (__predict_false(isroot && (ndp->ni_lcf &
1216 (NI_LCF_STRICTREL | NI_LCF_STRICTREL_KTR)) != 0)) {
1217 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
1218 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
1219 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0) {
1220 error = ENOTCAPABLE;
1221 goto capdotdot;
1222 }
1223 }
1224 if (isroot || ((dp->v_vflag & VV_ROOT) != 0 &&
1225 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
1226 ndp->ni_dvp = dp;
1227 ndp->ni_vp = dp;
1228 VREF(dp);
1229 goto nextname;
1230 }
1231 if ((dp->v_vflag & VV_ROOT) == 0)
1232 break;
1233 if (VN_IS_DOOMED(dp)) { /* forced unmount */
1234 error = ENOENT;
1235 goto bad;
1236 }
1237 tdp = dp;
1238 dp = dp->v_mount->mnt_vnodecovered;
1239 VREF(dp);
1240 vput(tdp);
1241 vn_lock(dp,
1242 enforce_lkflags(dp->v_mount, cnp->cn_lkflags |
1243 LK_RETRY));
1244 error = nameicap_check_dotdot(ndp, dp);
1245 if (error != 0) {
1246 capdotdot:
1247 goto bad;
1248 }
1249 }
1250 }
1251
1252 /*
1253 * We now have a segment name to search for, and a directory to search.
1254 */
1255 unionlookup:
1256 #ifdef MAC
1257 error = mac_vnode_check_lookup(cnp->cn_cred, dp, cnp);
1258 if (__predict_false(error))
1259 goto bad;
1260 #endif
1261 ndp->ni_dvp = dp;
1262 ndp->ni_vp = NULL;
1263 ASSERT_VOP_LOCKED(dp, "lookup");
1264 /*
1265 * If we have a shared lock we may need to upgrade the lock for the
1266 * last operation.
1267 */
1268 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
1269 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
1270 vn_lock(dp, LK_UPGRADE|LK_RETRY);
1271 if (VN_IS_DOOMED(dp)) {
1272 error = ENOENT;
1273 goto bad;
1274 }
1275 /*
1276 * If we're looking up the last component and we need an exclusive
1277 * lock, adjust our lkflags.
1278 */
1279 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1280 cnp->cn_lkflags = LK_EXCLUSIVE;
1281 lkflags_save = cnp->cn_lkflags;
1282 cnp->cn_lkflags = enforce_lkflags(dp->v_mount, cnp->cn_lkflags);
1283 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1284 cnp->cn_lkflags = lkflags_save;
1285 if (error != 0) {
1286 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1287 if ((error == ENOENT) &&
1288 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1289 (dp->v_mount->mnt_flag & MNT_UNION)) {
1290 tdp = dp;
1291 dp = dp->v_mount->mnt_vnodecovered;
1292 VREF(dp);
1293 vput(tdp);
1294 vn_lock(dp,
1295 enforce_lkflags(dp->v_mount, cnp->cn_lkflags |
1296 LK_RETRY));
1297 nameicap_tracker_add(ndp, dp);
1298 goto unionlookup;
1299 }
1300
1301 if (error == ERELOOKUP) {
1302 vref(dp);
1303 ndp->ni_vp = dp;
1304 error = 0;
1305 relookup = 1;
1306 goto good;
1307 }
1308
1309 if (error != EJUSTRETURN)
1310 goto bad;
1311 /*
1312 * At this point, we know we're at the end of the
1313 * pathname. If creating / renaming, we can consider
1314 * allowing the file or directory to be created / renamed,
1315 * provided we're not on a read-only filesystem.
1316 */
1317 if (rdonly) {
1318 error = EROFS;
1319 goto bad;
1320 }
1321 /* trailing slash only allowed for directories */
1322 if ((cnp->cn_flags & TRAILINGSLASH) &&
1323 !(cnp->cn_flags & WILLBEDIR)) {
1324 error = ENOENT;
1325 goto bad;
1326 }
1327 if ((cnp->cn_flags & LOCKPARENT) == 0)
1328 VOP_UNLOCK(dp);
1329 /*
1330 * We return with ni_vp NULL to indicate that the entry
1331 * doesn't currently exist, leaving a pointer to the
1332 * (possibly locked) directory vnode in ndp->ni_dvp.
1333 */
1334 goto success;
1335 }
1336
1337 good:
1338 dp = ndp->ni_vp;
1339
1340 /*
1341 * Check for symbolic link
1342 */
1343 if ((dp->v_type == VLNK) &&
1344 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1345 *ndp->ni_next == '/')) {
1346 cnp->cn_flags |= ISSYMLINK;
1347 if (VN_IS_DOOMED(dp)) {
1348 /*
1349 * We can't know whether the directory was mounted with
1350 * NOSYMFOLLOW, so we can't follow safely.
1351 */
1352 error = ENOENT;
1353 goto bad2;
1354 }
1355 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1356 error = EACCES;
1357 goto bad2;
1358 }
1359 /*
1360 * Symlink code always expects an unlocked dvp.
1361 */
1362 if (ndp->ni_dvp != ndp->ni_vp) {
1363 VOP_UNLOCK(ndp->ni_dvp);
1364 ni_dvp_unlocked = 1;
1365 }
1366 goto success;
1367 }
1368
1369 if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 &&
1370 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1371 error = vfs_lookup_cross_mount(ndp);
1372 if (error != 0)
1373 goto bad_unlocked;
1374 /*
1375 * FALLTHROUGH to nextname
1376 */
1377 dp = ndp->ni_vp;
1378 }
1379
1380 nextname:
1381 /*
1382 * Not a symbolic link that we will follow. Continue with the
1383 * next component if there is any; otherwise, we're done.
1384 */
1385 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1386 ("lookup: invalid path state."));
1387 if (relookup) {
1388 relookup = 0;
1389 ndp->ni_pathlen = prev_ni_pathlen;
1390 ndp->ni_next = prev_ni_next;
1391 if (ndp->ni_dvp != dp)
1392 vput(ndp->ni_dvp);
1393 else
1394 vrele(ndp->ni_dvp);
1395 goto dirloop;
1396 }
1397 if (cnp->cn_flags & ISDOTDOT) {
1398 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1399 if (error != 0)
1400 goto bad2;
1401 }
1402 if (*ndp->ni_next == '/') {
1403 cnp->cn_nameptr = ndp->ni_next;
1404 while (*cnp->cn_nameptr == '/') {
1405 cnp->cn_nameptr++;
1406 ndp->ni_pathlen--;
1407 }
1408 if (ndp->ni_dvp != dp)
1409 vput(ndp->ni_dvp);
1410 else
1411 vrele(ndp->ni_dvp);
1412 goto dirloop;
1413 }
1414 /*
1415 * If we're processing a path with a trailing slash,
1416 * check that the end result is a directory.
1417 */
1418 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1419 error = ENOTDIR;
1420 goto bad2;
1421 }
1422 /*
1423 * Disallow directory write attempts on read-only filesystems.
1424 */
1425 if (rdonly &&
1426 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1427 error = EROFS;
1428 goto bad2;
1429 }
1430 if (!wantparent) {
1431 ni_dvp_unlocked = 2;
1432 if (ndp->ni_dvp != dp)
1433 vput(ndp->ni_dvp);
1434 else
1435 vrele(ndp->ni_dvp);
1436 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1437 VOP_UNLOCK(ndp->ni_dvp);
1438 ni_dvp_unlocked = 1;
1439 }
1440
1441 if (cnp->cn_flags & AUDITVNODE1)
1442 AUDIT_ARG_VNODE1(dp);
1443 else if (cnp->cn_flags & AUDITVNODE2)
1444 AUDIT_ARG_VNODE2(dp);
1445
1446 if ((cnp->cn_flags & LOCKLEAF) == 0)
1447 VOP_UNLOCK(dp);
1448 success:
1449 /*
1450 * FIXME: for lookups which only cross a mount point to fetch the
1451 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
1452 * if either WANTPARENT or LOCKPARENT is set.
1453 */
1454 /*
1455 * Because of shared lookup we may have the vnode shared locked, but
1456 * the caller may want it to be exclusively locked.
1457 */
1458 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1459 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1460 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1461 if (VN_IS_DOOMED(dp)) {
1462 error = ENOENT;
1463 goto bad2;
1464 }
1465 }
1466 success_right_lock:
1467 if (ndp->ni_vp != NULL) {
1468 if ((cnp->cn_flags & ISDOTDOT) == 0)
1469 nameicap_tracker_add(ndp, ndp->ni_vp);
1470 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS)
1471 return (vfs_lookup_failifexists(ndp));
1472 }
1473 return (0);
1474
1475 bad2:
1476 if (ni_dvp_unlocked != 2) {
1477 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1478 vput(ndp->ni_dvp);
1479 else
1480 vrele(ndp->ni_dvp);
1481 }
1482 bad:
1483 vput(dp);
1484 bad_unlocked:
1485 ndp->ni_vp = NULL;
1486 return (error);
1487 }
1488
1489 /*
1490 * relookup - lookup a path name component
1491 * Used by lookup to re-acquire things.
1492 */
1493 int
vfs_relookup(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,bool refstart)1494 vfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1495 bool refstart)
1496 {
1497 struct vnode *dp = NULL; /* the directory we are searching */
1498 int rdonly; /* lookup read-only flag bit */
1499 int error = 0;
1500
1501 KASSERT(cnp->cn_flags & ISLASTCN,
1502 ("relookup: Not given last component."));
1503 /*
1504 * Setup: break out flag bits into variables.
1505 */
1506 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0,
1507 ("relookup: parent not wanted"));
1508 rdonly = cnp->cn_flags & RDONLY;
1509 cnp->cn_flags &= ~ISSYMLINK;
1510 dp = dvp;
1511 cnp->cn_lkflags = LK_EXCLUSIVE;
1512 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1513
1514 /*
1515 * Search a new directory.
1516 *
1517 * See a comment in vfs_lookup for cnp->cn_nameptr.
1518 *
1519 * Check for "" which represents the root directory after slash
1520 * removal.
1521 */
1522 if (cnp->cn_nameptr[0] == '\0') {
1523 /*
1524 * Support only LOOKUP for "/" because lookup()
1525 * can't succeed for CREATE, DELETE and RENAME.
1526 */
1527 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1528 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1529
1530 if (!(cnp->cn_flags & LOCKLEAF))
1531 VOP_UNLOCK(dp);
1532 *vpp = dp;
1533 /* XXX This should probably move to the top of function. */
1534 if (refstart)
1535 panic("lookup: SAVESTART");
1536 return (0);
1537 }
1538
1539 if (cnp->cn_flags & ISDOTDOT)
1540 panic ("relookup: lookup on dot-dot");
1541
1542 /*
1543 * We now have a segment name to search for, and a directory to search.
1544 */
1545 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1546 KASSERT(*vpp == NULL, ("leaf should be empty"));
1547 if (error != EJUSTRETURN)
1548 goto bad;
1549 /*
1550 * If creating and at end of pathname, then can consider
1551 * allowing file to be created.
1552 */
1553 if (rdonly) {
1554 error = EROFS;
1555 goto bad;
1556 }
1557 /* ASSERT(dvp == ndp->ni_startdir) */
1558 if (refstart)
1559 VREF(dvp);
1560 if ((cnp->cn_flags & LOCKPARENT) == 0)
1561 VOP_UNLOCK(dp);
1562 /*
1563 * We return with ni_vp NULL to indicate that the entry
1564 * doesn't currently exist, leaving a pointer to the
1565 * (possibly locked) directory vnode in ndp->ni_dvp.
1566 */
1567 return (0);
1568 }
1569
1570 dp = *vpp;
1571
1572 /*
1573 * Disallow directory write attempts on read-only filesystems.
1574 */
1575 if (rdonly &&
1576 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1577 if (dvp == dp)
1578 vrele(dvp);
1579 else
1580 vput(dvp);
1581 error = EROFS;
1582 goto bad;
1583 }
1584 /*
1585 * Set the parent lock/ref state to the requested state.
1586 */
1587 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp)
1588 VOP_UNLOCK(dvp);
1589 /*
1590 * Check for symbolic link
1591 */
1592 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1593 ("relookup: symlink found.\n"));
1594
1595 /* ASSERT(dvp == ndp->ni_startdir) */
1596 if (refstart)
1597 VREF(dvp);
1598
1599 if ((cnp->cn_flags & LOCKLEAF) == 0)
1600 VOP_UNLOCK(dp);
1601 return (0);
1602 bad:
1603 vput(dp);
1604 *vpp = NULL;
1605 return (error);
1606 }
1607
1608 #ifdef INVARIANTS
1609 /*
1610 * Validate the final state of ndp after the lookup.
1611 */
1612 static void
NDVALIDATE_impl(struct nameidata * ndp,int line)1613 NDVALIDATE_impl(struct nameidata *ndp, int line)
1614 {
1615 struct componentname *cnp;
1616
1617 cnp = &ndp->ni_cnd;
1618 if (cnp->cn_pnbuf == NULL)
1619 panic("%s: got no buf! called from %d", __func__, line);
1620 }
1621
1622 #endif
1623