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 * Reset ndp to its original state.
79 */
80 #define NDRESET(ndp) do { \
81 NDREINIT_DBG(ndp); \
82 ndp->ni_resflags = 0; \
83 ndp->ni_cnd.cn_flags &= ~NAMEI_INTERNAL_FLAGS; \
84 } while (0)
85 /*
86 * Prepare namei() to restart. Reset components to its original state and set
87 * ISRESTARTED flag which signals the underlying lookup code to change the root
88 * from ABI root to actual root and prevents a further restarts.
89 */
90 #define NDRESTART(ndp) do { \
91 NDRESET(ndp); \
92 ndp->ni_cnd.cn_flags |= ISRESTARTED; \
93 } while (0)
94
95 #ifdef KTRACE
96 #define NIKTRCAPFAIL(path) ktrcapfail(CAPFAIL_NAMEI, (path))
97 #else
98 #define NIKTRCAPFAIL(path)
99 #endif
100
101 #define NI_CAP_VIOLATION(ndp, path) do { \
102 NIKTRCAPFAIL(path); \
103 (ndp)->ni_lcf &= ~NI_LCF_KTR_FLAGS; \
104 } while (0)
105
106 SDT_PROVIDER_DEFINE(vfs);
107 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
108 "unsigned long", "bool");
109 SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool",
110 "struct nameidata");
111
112 /* Allocation zone for namei. */
113 uma_zone_t namei_zone;
114
115 /* Placeholder vnode for mp traversal. */
116 static struct vnode *vp_crossmp;
117
118 static int
crossmp_vop_islocked(struct vop_islocked_args * ap)119 crossmp_vop_islocked(struct vop_islocked_args *ap)
120 {
121
122 return (LK_SHARED);
123 }
124
125 static int
crossmp_vop_lock1(struct vop_lock1_args * ap)126 crossmp_vop_lock1(struct vop_lock1_args *ap)
127 {
128 struct vnode *vp;
129 struct lock *lk __diagused;
130 int flags;
131
132 vp = ap->a_vp;
133 lk = vp->v_vnlock;
134 flags = ap->a_flags;
135
136 KASSERT((flags & (LK_SHARED | LK_NOWAIT)) == (LK_SHARED | LK_NOWAIT),
137 ("%s: invalid lock request 0x%x for crossmp", __func__, flags));
138
139 if ((flags & LK_INTERLOCK) != 0)
140 VI_UNLOCK(vp);
141 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, ap->a_line);
142 return (0);
143 }
144
145 static int
crossmp_vop_unlock(struct vop_unlock_args * ap)146 crossmp_vop_unlock(struct vop_unlock_args *ap)
147 {
148 struct vnode *vp;
149 struct lock *lk __diagused;
150
151 vp = ap->a_vp;
152 lk = vp->v_vnlock;
153
154 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
155 LOCK_LINE);
156 return (0);
157 }
158
159 static struct vop_vector crossmp_vnodeops = {
160 .vop_default = &default_vnodeops,
161 .vop_islocked = crossmp_vop_islocked,
162 .vop_lock1 = crossmp_vop_lock1,
163 .vop_unlock = crossmp_vop_unlock,
164 };
165 /*
166 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
167 * gets allocated early. See nameiinit for the direct call below.
168 */
169
170 struct nameicap_tracker {
171 TAILQ_ENTRY(nameicap_tracker) nm_link;
172 struct mount *mp;
173 };
174
175 /* Zone for cap mode tracker elements used for dotdot capability checks. */
176 MALLOC_DEFINE(M_NAMEITRACKER, "namei_tracker", "namei tracking for dotdot");
177
178 static void
nameiinit(void * dummy __unused)179 nameiinit(void *dummy __unused)
180 {
181
182 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
183 UMA_ALIGN_PTR, 0);
184 vfs_vector_op_register(&crossmp_vnodeops);
185 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
186 vp_crossmp->v_state = VSTATE_CONSTRUCTED;
187 vp_crossmp->v_irflag |= VIRF_CROSSMP;
188 }
189 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
190
191 static int lookup_cap_dotdot = 1;
192 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
193 &lookup_cap_dotdot, 0,
194 "enables \"..\" components in path lookup in capability mode");
195 static int lookup_cap_dotdot_nonlocal = 1;
196 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
197 &lookup_cap_dotdot_nonlocal, 0,
198 "enables \"..\" components in path lookup in capability mode "
199 "on non-local mount");
200
201 static int
nameicap_tracker_add(struct nameidata * ndp,struct vnode * dp)202 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
203 {
204 struct nameicap_tracker *nt;
205 struct mount *mp;
206 int error;
207
208 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
209 return (0);
210 mp = NULL;
211 error = VOP_GETWRITEMOUNT(dp, &mp);
212 if (error != 0)
213 return (error);
214 nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head);
215 if (nt != NULL && nt->mp == mp) {
216 vfs_rel(mp);
217 return (0);
218 }
219 nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK);
220 nt->mp = mp;
221 error = lockmgr(&mp->mnt_renamelock, LK_SHARED | LK_NOWAIT, 0);
222 if (error != 0) {
223 MPASS(ndp->ni_nctrack_mnt == NULL);
224 ndp->ni_nctrack_mnt = mp;
225 free(nt, M_NAMEITRACKER);
226 error = ERESTART;
227 } else {
228 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
229 }
230 return (error);
231 }
232
233 static void
nameicap_cleanup(struct nameidata * ndp,int error)234 nameicap_cleanup(struct nameidata *ndp, int error)
235 {
236 struct nameicap_tracker *nt, *nt1;
237 struct mount *mp;
238
239 KASSERT((ndp->ni_nctrack_mnt == NULL &&
240 TAILQ_EMPTY(&ndp->ni_cap_tracker)) ||
241 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0,
242 ("tracker active and not strictrelative"));
243
244 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
245 mp = nt->mp;
246 lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0);
247 vfs_rel(mp);
248 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
249 free(nt, M_NAMEITRACKER);
250 }
251
252 mp = ndp->ni_nctrack_mnt;
253 if (mp != NULL) {
254 if (error == ERESTART) {
255 lockmgr(&mp->mnt_renamelock, LK_EXCLUSIVE, 0);
256 lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0);
257 }
258 vfs_rel(mp);
259 ndp->ni_nctrack_mnt = NULL;
260 }
261 }
262
263 /*
264 * For dotdot lookups in capability mode, disallow walking over the
265 * directory no_rbeneath_dpp that was used as the starting point of
266 * the lookup. Since we take the mnt_renamelocks of all mounts we
267 * ever walked over during lookup, parallel renames are disabled.
268 * This prevents the situation where we circumvent walk over
269 * ni_rbeneath_dpp following dotdots.
270 *
271 * Also allow to force failure of dotdot lookups for non-local
272 * filesystems, where external agents might assist local lookups to
273 * escape the compartment.
274 */
275 static int
nameicap_check_dotdot(struct nameidata * ndp,struct vnode * dp)276 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
277 {
278 struct mount *mp;
279
280 if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf &
281 NI_LCF_STRICTREL) == 0)
282 return (0);
283 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL_KTR |
284 NI_LCF_CAP_DOTDOT_KTR)) == NI_LCF_STRICTREL_KTR))
285 NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
286 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0)
287 goto violation;
288 if (dp == ndp->ni_rbeneath_dpp)
289 goto violation;
290 mp = dp->v_mount;
291 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
292 (mp->mnt_flag & MNT_LOCAL) == 0)
293 goto violation;
294 return (0);
295
296 violation:
297 if (__predict_false((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0))
298 NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
299 return (ENOTCAPABLE);
300 }
301
302 static void
namei_cleanup_cnp(struct componentname * cnp)303 namei_cleanup_cnp(struct componentname *cnp)
304 {
305
306 uma_zfree(namei_zone, cnp->cn_pnbuf);
307 cnp->cn_pnbuf = NULL;
308 cnp->cn_nameptr = NULL;
309 }
310
311 static int
namei_handle_root(struct nameidata * ndp,struct vnode ** dpp)312 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
313 {
314 struct componentname *cnp;
315
316 cnp = &ndp->ni_cnd;
317 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL |
318 NI_LCF_STRICTREL_KTR)) != 0)) {
319 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
320 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
321 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
322 return (ENOTCAPABLE);
323 }
324 while (*(cnp->cn_nameptr) == '/') {
325 cnp->cn_nameptr++;
326 ndp->ni_pathlen--;
327 }
328 *dpp = ndp->ni_rootdir;
329 vrefact(*dpp);
330 return (0);
331 }
332
333 static int
namei_setup(struct nameidata * ndp,struct vnode ** dpp,struct pwd ** pwdp)334 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
335 {
336 struct componentname *cnp;
337 struct thread *td;
338 struct pwd *pwd;
339 int error;
340 bool startdir_used;
341
342 cnp = &ndp->ni_cnd;
343 td = curthread;
344
345 startdir_used = false;
346 *pwdp = NULL;
347 *dpp = NULL;
348
349 #ifdef CAPABILITY_MODE
350 /*
351 * In capability mode, lookups must be restricted to happen in
352 * the subtree with the root specified by the file descriptor:
353 * - The root must be real file descriptor, not the pseudo-descriptor
354 * AT_FDCWD.
355 * - The passed path must be relative and not absolute.
356 * - If lookup_cap_dotdot is disabled, path must not contain the
357 * '..' components.
358 * - If lookup_cap_dotdot is enabled, we verify that all '..'
359 * components lookups result in the directories which were
360 * previously walked by us, which prevents an escape from
361 * the relative root.
362 */
363 if ((cnp->cn_flags & NOCAPCHECK) == 0) {
364 if (CAP_TRACING(td)) {
365 ndp->ni_lcf |= NI_LCF_STRICTREL_KTR;
366 if (ndp->ni_dirfd == AT_FDCWD)
367 NI_CAP_VIOLATION(ndp, "AT_FDCWD");
368 }
369 if (IN_CAPABILITY_MODE(td)) {
370 ndp->ni_lcf |= NI_LCF_STRICTREL;
371 ndp->ni_resflags |= NIRES_STRICTREL;
372 if (ndp->ni_dirfd == AT_FDCWD)
373 return (ECAPMODE);
374 }
375 }
376 #endif
377 error = 0;
378
379 /*
380 * Get starting point for the translation.
381 */
382 pwd = pwd_hold(td);
383 /*
384 * The reference on ni_rootdir is acquired in the block below to avoid
385 * back-to-back atomics for absolute lookups.
386 */
387 namei_setup_rootdir(ndp, cnp, pwd);
388 ndp->ni_topdir = pwd->pwd_jdir;
389
390 if (cnp->cn_pnbuf[0] == '/') {
391 ndp->ni_resflags |= NIRES_ABS;
392 error = namei_handle_root(ndp, dpp);
393 } else {
394 if (ndp->ni_startdir != NULL) {
395 *dpp = ndp->ni_startdir;
396 startdir_used = true;
397 } else if (ndp->ni_dirfd == AT_FDCWD) {
398 *dpp = pwd->pwd_cdir;
399 vrefact(*dpp);
400 } else {
401 if (cnp->cn_flags & AUDITVNODE1)
402 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
403 if (cnp->cn_flags & AUDITVNODE2)
404 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
405
406 error = fgetvp_lookup(ndp, dpp);
407 }
408 if (error == 0 && (*dpp)->v_type != VDIR &&
409 (cnp->cn_flags & OPENNAMED) == 0 &&
410 (cnp->cn_pnbuf[0] != '\0' ||
411 (cnp->cn_flags & EMPTYPATH) == 0))
412 error = ENOTDIR;
413 }
414 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
415 if (cnp->cn_pnbuf[0] == '/') {
416 error = ENOTCAPABLE;
417 } else if ((ndp->ni_lcf & NI_LCF_STRICTREL) == 0) {
418 ndp->ni_lcf |= NI_LCF_STRICTREL |
419 NI_LCF_CAP_DOTDOT;
420 }
421 }
422 if (error == 0 && (ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
423 ndp->ni_rbeneath_dpp = *dpp;
424
425 /*
426 * If we are auditing the kernel pathname, save the user pathname.
427 */
428 if (AUDITING_TD(td)) {
429 if (cnp->cn_flags & AUDITVNODE1)
430 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
431 if (cnp->cn_flags & AUDITVNODE2)
432 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
433 }
434 if (ndp->ni_startdir != NULL && !startdir_used)
435 vrele(ndp->ni_startdir);
436 if (error != 0) {
437 if (*dpp != NULL)
438 vrele(*dpp);
439 pwd_drop(pwd);
440 return (error);
441 }
442 if (lookup_cap_dotdot != 0) {
443 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
444 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT_KTR;
445 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
446 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
447 }
448 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
449 cnp->cn_flags, false);
450 *pwdp = pwd;
451 return (0);
452 }
453
454 static int
namei_getpath(struct nameidata * ndp)455 namei_getpath(struct nameidata *ndp)
456 {
457 struct componentname *cnp;
458 int error;
459
460 cnp = &ndp->ni_cnd;
461
462 /*
463 * Get a buffer for the name to be translated, and copy the
464 * name into the buffer.
465 */
466 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
467 if (ndp->ni_segflg == UIO_SYSSPACE) {
468 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
469 &ndp->ni_pathlen);
470 } else {
471 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
472 &ndp->ni_pathlen);
473 }
474
475 return (error);
476 }
477
478 static int
namei_emptypath(struct nameidata * ndp)479 namei_emptypath(struct nameidata *ndp)
480 {
481 struct componentname *cnp;
482 struct pwd *pwd;
483 struct vnode *dp;
484 int error;
485
486 cnp = &ndp->ni_cnd;
487 MPASS(*cnp->cn_pnbuf == '\0');
488 MPASS((cnp->cn_flags & EMPTYPATH) != 0);
489 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
490
491 ndp->ni_resflags |= NIRES_EMPTYPATH;
492 error = namei_setup(ndp, &dp, &pwd);
493 if (error != 0) {
494 goto errout;
495 }
496
497 /*
498 * Usecount on dp already provided by namei_setup.
499 */
500 ndp->ni_vp = dp;
501 pwd_drop(pwd);
502 NDVALIDATE(ndp);
503 if ((cnp->cn_flags & LOCKLEAF) != 0) {
504 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ?
505 LK_SHARED : LK_EXCLUSIVE);
506 if (VN_IS_DOOMED(dp)) {
507 vput(dp);
508 error = ENOENT;
509 goto errout;
510 }
511 }
512 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp);
513 return (0);
514
515 errout:
516 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
517 namei_cleanup_cnp(cnp);
518 return (error);
519 }
520
521 static int __noinline
namei_follow_link(struct nameidata * ndp)522 namei_follow_link(struct nameidata *ndp)
523 {
524 char *cp;
525 struct iovec aiov;
526 struct uio auio;
527 struct componentname *cnp;
528 struct thread *td;
529 int error, linklen;
530
531 error = 0;
532 cnp = &ndp->ni_cnd;
533 td = curthread;
534
535 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
536 error = ELOOP;
537 goto out;
538 }
539 #ifdef MAC
540 if ((cnp->cn_flags & NOMACCHECK) == 0) {
541 error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp);
542 if (error != 0)
543 goto out;
544 }
545 #endif
546 if (ndp->ni_pathlen > 1)
547 cp = uma_zalloc(namei_zone, M_WAITOK);
548 else
549 cp = cnp->cn_pnbuf;
550 aiov.iov_base = cp;
551 aiov.iov_len = MAXPATHLEN;
552 auio.uio_iov = &aiov;
553 auio.uio_iovcnt = 1;
554 auio.uio_offset = 0;
555 auio.uio_rw = UIO_READ;
556 auio.uio_segflg = UIO_SYSSPACE;
557 auio.uio_td = td;
558 auio.uio_resid = MAXPATHLEN;
559 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
560 if (error != 0) {
561 if (ndp->ni_pathlen > 1)
562 uma_zfree(namei_zone, cp);
563 goto out;
564 }
565 linklen = MAXPATHLEN - auio.uio_resid;
566 if (linklen == 0) {
567 if (ndp->ni_pathlen > 1)
568 uma_zfree(namei_zone, cp);
569 error = ENOENT;
570 goto out;
571 }
572 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
573 if (ndp->ni_pathlen > 1)
574 uma_zfree(namei_zone, cp);
575 error = ENAMETOOLONG;
576 goto out;
577 }
578 if (ndp->ni_pathlen > 1) {
579 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
580 uma_zfree(namei_zone, cnp->cn_pnbuf);
581 cnp->cn_pnbuf = cp;
582 } else
583 cnp->cn_pnbuf[linklen] = '\0';
584 ndp->ni_pathlen += linklen;
585 out:
586 return (error);
587 }
588
589 /*
590 * Convert a pathname into a pointer to a locked vnode.
591 *
592 * The FOLLOW flag is set when symbolic links are to be followed
593 * when they occur at the end of the name translation process.
594 * Symbolic links are always followed for all other pathname
595 * components other than the last.
596 *
597 * The segflg defines whether the name is to be copied from user
598 * space or kernel space.
599 *
600 * Overall outline of namei:
601 *
602 * copy in name
603 * get starting directory
604 * while (!done && !error) {
605 * call lookup to search path.
606 * if symbolic link, massage name in buffer and continue
607 * }
608 */
609 int
namei(struct nameidata * ndp)610 namei(struct nameidata *ndp)
611 {
612 struct vnode *dp; /* the directory we are searching */
613 struct componentname *cnp;
614 struct thread *td;
615 struct pwd *pwd;
616 int error;
617 enum cache_fpl_status status;
618
619 cnp = &ndp->ni_cnd;
620 td = curthread;
621 #ifdef INVARIANTS
622 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0,
623 ("%s: repeated call to namei without NDREINIT", __func__));
624 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED,
625 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags));
626 ndp->ni_debugflags |= NAMEI_DBG_CALLED;
627 if (ndp->ni_startdir != NULL)
628 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR;
629 if (cnp->cn_flags & FAILIFEXISTS) {
630 KASSERT(cnp->cn_nameiop == CREATE,
631 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop));
632 /*
633 * The limitation below is to restrict hairy corner cases.
634 */
635 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT,
636 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF",
637 __func__));
638 }
639 #endif
640 ndp->ni_cnd.cn_cred = td->td_ucred;
641 KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x",
642 __func__, ndp->ni_resflags));
643 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
644 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
645 ("namei: unexpected flags: %#jx",
646 (uintmax_t)(cnp->cn_flags & NAMEI_INTERNAL_FLAGS)));
647 if (cnp->cn_flags & NOCACHE)
648 KASSERT(cnp->cn_nameiop != LOOKUP,
649 ("%s: NOCACHE passed with LOOKUP", __func__));
650 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
651 ndp->ni_startdir->v_type == VBAD);
652
653 restart:
654 ndp->ni_lcf = 0;
655 ndp->ni_loopcnt = 0;
656 ndp->ni_vp = NULL;
657
658 error = namei_getpath(ndp);
659 if (__predict_false(error != 0)) {
660 namei_cleanup_cnp(cnp);
661 nameicap_cleanup(ndp, error);
662 SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
663 false, ndp);
664 return (error);
665 }
666
667 cnp->cn_nameptr = cnp->cn_pnbuf;
668
669 #ifdef KTRACE
670 if (KTRPOINT(td, KTR_NAMEI)) {
671 ktrnamei(cnp->cn_pnbuf);
672 }
673 #endif
674 TSNAMEI(curthread->td_proc->p_pid, cnp->cn_pnbuf);
675
676 /*
677 * First try looking up the target without locking any vnodes.
678 *
679 * We may need to start from scratch or pick up where it left off.
680 */
681 error = cache_fplookup(ndp, &status, &pwd);
682 switch (status) {
683 case CACHE_FPL_STATUS_UNSET:
684 __assert_unreachable();
685 break;
686 case CACHE_FPL_STATUS_HANDLED:
687 if (error == 0)
688 NDVALIDATE(ndp);
689 else if (__predict_false(pwd->pwd_adir != pwd->pwd_rdir &&
690 (cnp->cn_flags & ISRESTARTED) == 0)) {
691 namei_cleanup_cnp(cnp);
692 nameicap_cleanup(ndp, ERESTART);
693 NDRESTART(ndp);
694 goto restart;
695 }
696 return (error);
697 case CACHE_FPL_STATUS_PARTIAL:
698 dp = ndp->ni_startdir;
699 break;
700 case CACHE_FPL_STATUS_DESTROYED:
701 ndp->ni_loopcnt = 0;
702 error = namei_getpath(ndp);
703 if (__predict_false(error != 0)) {
704 namei_cleanup_cnp(cnp);
705 nameicap_cleanup(ndp, error);
706 return (error);
707 }
708 cnp->cn_nameptr = cnp->cn_pnbuf;
709 /* FALLTHROUGH */
710 case CACHE_FPL_STATUS_ABORTED:
711 MPASS(ndp->ni_lcf == 0);
712 if (*cnp->cn_pnbuf == '\0') {
713 if ((cnp->cn_flags & EMPTYPATH) != 0) {
714 error = namei_emptypath(ndp);
715 nameicap_cleanup(ndp, error);
716 return (error);
717 }
718 namei_cleanup_cnp(cnp);
719 nameicap_cleanup(ndp, ENOENT);
720 SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
721 false, ndp);
722 return (ENOENT);
723 }
724 error = namei_setup(ndp, &dp, &pwd);
725 if (error != 0) {
726 namei_cleanup_cnp(cnp);
727 nameicap_cleanup(ndp, error);
728 return (error);
729 }
730 break;
731 }
732
733 /*
734 * Locked lookup.
735 */
736 for (;;) {
737 ndp->ni_startdir = dp;
738 error = vfs_lookup(ndp);
739 if (error != 0) {
740 uint64_t was_restarted;
741 bool abi_restart;
742
743 was_restarted = ndp->ni_cnd.cn_flags &
744 ISRESTARTED;
745 abi_restart = pwd->pwd_adir != pwd->pwd_rdir &&
746 error == ENOENT && was_restarted == 0;
747 if (error != ERESTART && !abi_restart)
748 goto out;
749 nameicap_cleanup(ndp, error);
750 pwd_drop(pwd);
751 namei_cleanup_cnp(cnp);
752 NDRESET(ndp);
753 if (abi_restart)
754 was_restarted = ISRESTARTED;
755 ndp->ni_cnd.cn_flags |= was_restarted;
756 goto restart;
757 }
758
759 /*
760 * If not a symbolic link, we're done.
761 */
762 if ((cnp->cn_flags & ISSYMLINK) == 0) {
763 SDT_PROBE4(vfs, namei, lookup, return, error,
764 ndp->ni_vp, false, ndp);
765 nameicap_cleanup(ndp, 0);
766 pwd_drop(pwd);
767 NDVALIDATE(ndp);
768 return (0);
769 }
770 error = namei_follow_link(ndp);
771 if (error != 0)
772 break;
773 vput(ndp->ni_vp);
774 dp = ndp->ni_dvp;
775 /*
776 * Check if root directory should replace current directory.
777 */
778 cnp->cn_nameptr = cnp->cn_pnbuf;
779 if (*(cnp->cn_nameptr) == '/') {
780 /*
781 * Reset the lookup to start from the real root without
782 * origin path name reloading.
783 */
784 if (__predict_false(ndp->ni_rootdir != pwd->pwd_rdir)) {
785 cnp->cn_flags |= ISRESTARTED;
786 ndp->ni_rootdir = pwd->pwd_rdir;
787 }
788 vrele(dp);
789 error = namei_handle_root(ndp, &dp);
790 if (error != 0)
791 goto out;
792 }
793 }
794 vput(ndp->ni_vp);
795 ndp->ni_vp = NULL;
796 vrele(ndp->ni_dvp);
797 out:
798 MPASS(error != 0 && error != ERESTART);
799 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
800 namei_cleanup_cnp(cnp);
801 nameicap_cleanup(ndp, error);
802 pwd_drop(pwd);
803 return (error);
804 }
805
806 static int
enforce_lkflags(struct mount * mp,int lkflags)807 enforce_lkflags(struct mount *mp, int lkflags)
808 {
809
810 if (mp == NULL || ((lkflags & LK_SHARED) &&
811 !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
812 lkflags &= ~LK_SHARED;
813 lkflags |= LK_EXCLUSIVE;
814 }
815 lkflags |= LK_NODDLKTREAT;
816 return (lkflags);
817 }
818
819 static __inline int
needs_exclusive_leaf(struct mount * mp,int flags)820 needs_exclusive_leaf(struct mount *mp, int flags)
821 {
822
823 /*
824 * Intermediate nodes can use shared locks, we only need to
825 * force an exclusive lock for leaf nodes.
826 */
827 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
828 return (0);
829
830 /* Always use exclusive locks if LOCKSHARED isn't set. */
831 if (!(flags & LOCKSHARED))
832 return (1);
833
834 /*
835 * For lookups during open(), if the mount point supports
836 * extended shared operations, then use a shared lock for the
837 * leaf node, otherwise use an exclusive lock.
838 */
839 if ((flags & ISOPEN) != 0)
840 return (!MNT_EXTENDED_SHARED(mp));
841
842 /*
843 * Lookup requests outside of open() that specify LOCKSHARED
844 * only need a shared lock on the leaf vnode.
845 */
846 return (0);
847 }
848
849 /*
850 * Various filesystems expect to be able to copy a name component with length
851 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make
852 * sure that these are the same size.
853 */
854 _Static_assert(MAXNAMLEN == NAME_MAX,
855 "MAXNAMLEN and NAME_MAX have different values");
856
857 static int __noinline
vfs_lookup_degenerate(struct nameidata * ndp,struct vnode * dp,int wantparent)858 vfs_lookup_degenerate(struct nameidata *ndp, struct vnode *dp, int wantparent)
859 {
860 struct componentname *cnp;
861 struct mount *mp;
862 int error;
863
864 cnp = &ndp->ni_cnd;
865
866 cnp->cn_flags |= ISLASTCN;
867
868 mp = atomic_load_ptr(&dp->v_mount);
869 if (needs_exclusive_leaf(mp, cnp->cn_flags)) {
870 cnp->cn_lkflags &= ~LK_SHARED;
871 cnp->cn_lkflags |= LK_EXCLUSIVE;
872 }
873
874 vn_lock(dp, enforce_lkflags(mp, cnp->cn_lkflags | LK_RETRY));
875
876 if (dp->v_type != VDIR) {
877 error = ENOTDIR;
878 goto bad;
879 }
880 if (cnp->cn_nameiop != LOOKUP) {
881 error = EISDIR;
882 goto bad;
883 }
884 if (wantparent) {
885 ndp->ni_dvp = dp;
886 VREF(dp);
887 }
888 ndp->ni_vp = dp;
889 cnp->cn_namelen = 0;
890
891 if (cnp->cn_flags & AUDITVNODE1)
892 AUDIT_ARG_VNODE1(dp);
893 else if (cnp->cn_flags & AUDITVNODE2)
894 AUDIT_ARG_VNODE2(dp);
895
896 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
897 VOP_UNLOCK(dp);
898 return (0);
899 bad:
900 VOP_UNLOCK(dp);
901 return (error);
902 }
903
904 struct nameidata *
vfs_lookup_nameidata(struct componentname * cnp)905 vfs_lookup_nameidata(struct componentname *cnp)
906 {
907 if ((cnp->cn_flags & NAMEILOOKUP) == 0)
908 return (NULL);
909 return (__containerof(cnp, struct nameidata, ni_cnd));
910 }
911
912 /*
913 * Would a dotdot lookup relative to dvp cause this lookup to cross a jail or
914 * chroot boundary?
915 */
916 bool
vfs_lookup_isroot(struct nameidata * ndp,struct vnode * dvp)917 vfs_lookup_isroot(struct nameidata *ndp, struct vnode *dvp)
918 {
919 for (struct prison *pr = ndp->ni_cnd.cn_cred->cr_prison; pr != NULL;
920 pr = pr->pr_parent) {
921 if (dvp == pr->pr_root)
922 return (true);
923 }
924 return (dvp == ndp->ni_rootdir || dvp == ndp->ni_topdir ||
925 dvp == rootvnode);
926 }
927
928 /*
929 * FAILIFEXISTS handling.
930 *
931 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange
932 * behaviour of leaving the vnode unlocked if the target is the same
933 * vnode as the parent.
934 */
935 static int __noinline
vfs_lookup_failifexists(struct nameidata * ndp)936 vfs_lookup_failifexists(struct nameidata *ndp)
937 {
938 struct componentname *cnp __diagused;
939
940 cnp = &ndp->ni_cnd;
941
942 MPASS((cnp->cn_flags & ISSYMLINK) == 0);
943 if (ndp->ni_vp == ndp->ni_dvp)
944 vrele(ndp->ni_dvp);
945 else
946 vput(ndp->ni_dvp);
947 vrele(ndp->ni_vp);
948 ndp->ni_dvp = NULL;
949 ndp->ni_vp = NULL;
950 NDFREE_PNBUF(ndp);
951 return (EEXIST);
952 }
953
954 static int __noinline
vfs_lookup_cross_mount(struct nameidata * ndp)955 vfs_lookup_cross_mount(struct nameidata *ndp)
956 {
957 struct componentname *cnp;
958 struct mount *mp;
959 struct vnode *dp, *tdp;
960 int error, crosslkflags;
961 bool crosslock;
962
963 cnp = &ndp->ni_cnd;
964 dp = ndp->ni_vp;
965
966 /*
967 * The vnode has been mounted on, find the root of the mounted
968 * filesystem.
969 */
970 do {
971 mp = dp->v_mountedhere;
972 ASSERT_VOP_LOCKED(dp, __func__);
973 VNPASS((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 && mp != NULL, dp);
974
975 crosslock = (dp->v_vflag & VV_CROSSLOCK) != 0;
976 crosslkflags = enforce_lkflags(mp, cnp->cn_lkflags);
977 if (__predict_false(crosslock)) {
978 /*
979 * We are going to be holding the vnode lock, which
980 * in this case is shared by the root vnode of the
981 * filesystem mounted at mp, across the call to
982 * VFS_ROOT(). Make the situation clear to the
983 * filesystem by passing LK_CANRECURSE if the
984 * lock is held exclusive, or by clearinng
985 * LK_NODDLKTREAT to allow recursion on the shared
986 * lock in the presence of an exclusive waiter.
987 */
988 if (VOP_ISLOCKED(dp) == LK_EXCLUSIVE) {
989 crosslkflags &= ~LK_SHARED;
990 crosslkflags |= LK_EXCLUSIVE | LK_CANRECURSE;
991 } else if ((crosslkflags & LK_EXCLUSIVE) != 0) {
992 error = vn_lock(dp, LK_UPGRADE);
993 if (error != 0) {
994 MPASS(error == ENOENT);
995 vrele(dp);
996 if (dp != ndp->ni_dvp)
997 vput(ndp->ni_dvp);
998 else
999 vrele(ndp->ni_dvp);
1000 break;
1001 }
1002 if (dp->v_mountedhere != mp) {
1003 /*
1004 * Note that we rely on the
1005 * VIRF_MOUNTPOINT loop condition to
1006 * ensure we stop iterating if dp is
1007 * no longer a mountpoint at all.
1008 */
1009 continue;
1010 }
1011 } else
1012 crosslkflags &= ~LK_NODDLKTREAT;
1013 }
1014 if (vfs_busy(mp, 0) != 0)
1015 continue;
1016 if (__predict_true(!crosslock))
1017 vput(dp);
1018 if (dp != ndp->ni_dvp)
1019 vput(ndp->ni_dvp);
1020 else
1021 vrele(ndp->ni_dvp);
1022 vrefact(vp_crossmp);
1023 ndp->ni_dvp = vp_crossmp;
1024 error = VFS_ROOT(mp, crosslkflags, &tdp);
1025 vfs_unbusy(mp);
1026 if (__predict_false(crosslock))
1027 vput(dp);
1028 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1029 panic("vp_crossmp exclusively locked or reclaimed");
1030 if (error != 0)
1031 break;
1032 ndp->ni_vp = dp = tdp;
1033 } while ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0);
1034
1035 return (error);
1036 }
1037
1038 /*
1039 * Search a pathname.
1040 * This is a very central and rather complicated routine.
1041 *
1042 * The pathname is pointed to by cn_nameptr and is of length ni_pathlen.
1043 * The starting directory is taken from ni_startdir. The pathname is
1044 * descended until done, or a symbolic link is encountered. The cn_flags
1045 * has ISLASTCN or'ed if the path is completed or ISSYMLINK or'ed if a
1046 * symbolic link needing interpretation is encountered.
1047 *
1048 * The cn_nameiop is LOOKUP, CREATE, RENAME, or DELETE depending on
1049 * whether the name is to be looked up, created, renamed, or deleted.
1050 * When CREATE, RENAME, or DELETE is specified, information usable in
1051 * creating, renaming, or deleting a directory entry may be calculated.
1052 * If cn_flags has LOCKPARENT or'ed into it, the parent directory is returned
1053 * locked. If it has WANTPARENT or'ed into it, the parent directory is
1054 * returned unlocked. Otherwise the parent directory is not returned. If
1055 * the target of the pathname exists and LOCKLEAF is or'ed into the cn_flags
1056 * the target is returned locked, otherwise it is returned unlocked.
1057 *
1058 * Overall outline of lookup:
1059 *
1060 * handle degenerate case where name is null string
1061 *
1062 * dirloop:
1063 * identify next component of name at ndp->ni_cnd.cn_nameptr
1064 * handle .. special cases related to capabilities, chroot, jail
1065 * if .. and crossing mount points and on mounted filesys, find parent
1066 * call VOP_LOOKUP routine for next component name
1067 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
1068 * component vnode returned in ni_vp (if it exists), locked.
1069 * if result vnode is mounted on and crossing mount points,
1070 * find mounted on vnode
1071 * if more components of name, do next level at dirloop
1072 * if VOP_LOOKUP returns ERELOOKUP, repeat the same level at dirloop
1073 * return the answer in ni_vp, locked if LOCKLEAF set
1074 * if LOCKPARENT set, return locked parent in ni_dvp
1075 * if WANTPARENT set, return unlocked parent in ni_dvp
1076 */
1077 int
vfs_lookup(struct nameidata * ndp)1078 vfs_lookup(struct nameidata *ndp)
1079 {
1080 char *cp; /* pointer into pathname argument */
1081 char *prev_ni_next; /* saved ndp->ni_next */
1082 char *nulchar; /* location of '\0' in cn_pnbuf */
1083 char *lastchar; /* location of the last character */
1084 struct vnode *dp = NULL; /* the directory we are searching */
1085 struct vnode *tdp; /* saved dp */
1086 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
1087 int docache; /* == 0 do not cache last component */
1088 int wantparent; /* 1 => wantparent or lockparent flag */
1089 int rdonly; /* lookup read-only flag bit */
1090 int error = 0;
1091 int relookup = 0; /* do not consume the path component */
1092 struct componentname *cnp = &ndp->ni_cnd;
1093 int lkflags_save;
1094 int ni_dvp_unlocked;
1095
1096 /*
1097 * Setup: break out flag bits into variables.
1098 */
1099 ni_dvp_unlocked = 0;
1100 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
1101 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
1102 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
1103 /*
1104 * When set to zero, docache causes the last component of the
1105 * pathname to be deleted from the cache and the full lookup
1106 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
1107 * filesystems need some pre-computed values that are made
1108 * during the full lookup, for instance UFS sets dp->i_offset.
1109 *
1110 * The docache variable is set to zero when requested by the
1111 * NOCACHE flag and for all modifying operations except CREATE.
1112 */
1113 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
1114 if (cnp->cn_nameiop == DELETE ||
1115 (wantparent && cnp->cn_nameiop != CREATE &&
1116 cnp->cn_nameiop != LOOKUP))
1117 docache = 0;
1118 rdonly = cnp->cn_flags & RDONLY;
1119 cnp->cn_flags &= ~ISSYMLINK;
1120 ndp->ni_dvp = NULL;
1121
1122 cnp->cn_lkflags = LK_SHARED;
1123 dp = ndp->ni_startdir;
1124 ndp->ni_startdir = NULLVP;
1125
1126 /*
1127 * Leading slashes, if any, are supposed to be skipped by the caller.
1128 */
1129 MPASS(cnp->cn_nameptr[0] != '/');
1130
1131 /*
1132 * Check for degenerate name (e.g. / or "") which is a way of talking
1133 * about a directory, e.g. like "/." or ".".
1134 */
1135 if (__predict_false(cnp->cn_nameptr[0] == '\0')) {
1136 error = vfs_lookup_degenerate(ndp, dp, wantparent);
1137 if (error == 0)
1138 goto success_right_lock;
1139 goto bad_unlocked;
1140 }
1141
1142 /*
1143 * Nul-out trailing slashes (e.g., "foo///" -> "foo").
1144 *
1145 * This must be done before VOP_LOOKUP() because some fs's don't know
1146 * about trailing slashes. Remember if there were trailing slashes to
1147 * handle symlinks, existing non-directories and non-existing files
1148 * that won't be directories specially later.
1149 */
1150 MPASS(ndp->ni_pathlen >= 2);
1151 lastchar = &cnp->cn_nameptr[ndp->ni_pathlen - 2];
1152 if (*lastchar == '/') {
1153 while (lastchar >= cnp->cn_pnbuf) {
1154 *lastchar = '\0';
1155 lastchar--;
1156 ndp->ni_pathlen--;
1157 if (*lastchar != '/') {
1158 break;
1159 }
1160 }
1161 cnp->cn_flags |= TRAILINGSLASH;
1162 }
1163
1164 /*
1165 * We use shared locks until we hit the parent of the last cn then
1166 * we adjust based on the requesting flags.
1167 */
1168 vn_lock(dp,
1169 enforce_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
1170
1171 dirloop:
1172 /*
1173 * Search a new directory.
1174 *
1175 * The last component of the filename is left accessible via
1176 * cnp->cn_nameptr. It has to be freed with a call to NDFREE*.
1177 *
1178 * Store / as a temporary sentinel so that we only have one character
1179 * to test for. Pathnames tend to be short so this should not be
1180 * resulting in cache misses.
1181 */
1182 nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1];
1183 KASSERT(*nulchar == '\0',
1184 ("%s: expected nul at %p; string [%s]\n", __func__, nulchar,
1185 cnp->cn_pnbuf));
1186 *nulchar = '/';
1187 for (cp = cnp->cn_nameptr; *cp != '/'; cp++) {
1188 KASSERT(*cp != '\0',
1189 ("%s: encountered unexpected nul; string [%s]\n", __func__,
1190 cnp->cn_nameptr));
1191 continue;
1192 }
1193 *nulchar = '\0';
1194 cnp->cn_namelen = cp - cnp->cn_nameptr;
1195 if (__predict_false(cnp->cn_namelen > NAME_MAX)) {
1196 error = ENAMETOOLONG;
1197 goto bad;
1198 }
1199 prev_ni_pathlen = ndp->ni_pathlen;
1200 ndp->ni_pathlen -= cnp->cn_namelen;
1201 KASSERT(ndp->ni_pathlen <= PATH_MAX,
1202 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
1203 prev_ni_next = ndp->ni_next;
1204 ndp->ni_next = cp;
1205
1206 /*
1207 * Something else should be clearing this.
1208 */
1209 cnp->cn_flags &= ~(ISDOTDOT|ISLASTCN);
1210
1211 cnp->cn_flags |= MAKEENTRY;
1212 if (*cp == '\0' && docache == 0)
1213 cnp->cn_flags &= ~MAKEENTRY;
1214 if (cnp->cn_namelen == 2 &&
1215 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1216 cnp->cn_flags |= ISDOTDOT;
1217 if (*ndp->ni_next == 0) {
1218 cnp->cn_flags |= ISLASTCN;
1219
1220 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
1221 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))) {
1222 error = EINVAL;
1223 goto bad;
1224 }
1225 }
1226
1227 error = nameicap_tracker_add(ndp, dp);
1228 if (error != 0)
1229 goto bad;
1230
1231 /*
1232 * Make sure degenerate names don't get here, their handling was
1233 * previously found in this spot.
1234 */
1235 MPASS(cnp->cn_nameptr[0] != '\0');
1236
1237 /*
1238 * Handle "..": five special cases.
1239 * 0. If doing a capability lookup and lookup_cap_dotdot is
1240 * disabled, return ENOTCAPABLE.
1241 * 1. Return an error if this is the last component of
1242 * the name and the operation is DELETE or RENAME.
1243 * 2. If at root directory (e.g. after chroot)
1244 * or at absolute root directory
1245 * then ignore it so can't get out.
1246 * 3. If this vnode is the root of a mounted
1247 * filesystem, then replace it with the
1248 * vnode which was mounted on so we take the
1249 * .. in the other filesystem.
1250 * 4. If the vnode is the top directory of
1251 * the jail or chroot, don't let them out.
1252 * 5. If doing a capability lookup and lookup_cap_dotdot is
1253 * enabled, return ENOTCAPABLE if the lookup would escape
1254 * from the initial file descriptor directory.
1255 */
1256 if (cnp->cn_flags & ISDOTDOT) {
1257 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL_KTR |
1258 NI_LCF_CAP_DOTDOT_KTR)) == NI_LCF_STRICTREL_KTR))
1259 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
1260 if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL |
1261 NI_LCF_CAP_DOTDOT)) == NI_LCF_STRICTREL)) {
1262 error = ENOTCAPABLE;
1263 goto bad;
1264 }
1265 if ((cnp->cn_flags & ISLASTCN) != 0 &&
1266 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1267 error = EINVAL;
1268 goto bad;
1269 }
1270 for (;;) {
1271 bool isroot;
1272
1273 isroot = vfs_lookup_isroot(ndp, dp);
1274 if (__predict_false(isroot && (ndp->ni_lcf &
1275 (NI_LCF_STRICTREL | NI_LCF_STRICTREL_KTR)) != 0)) {
1276 if ((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0)
1277 NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
1278 if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0) {
1279 error = ENOTCAPABLE;
1280 goto bad;
1281 }
1282 }
1283 if (isroot || ((dp->v_vflag & VV_ROOT) != 0 &&
1284 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
1285 ndp->ni_dvp = dp;
1286 ndp->ni_vp = dp;
1287 VREF(dp);
1288 goto nextname;
1289 }
1290 if ((dp->v_vflag & VV_ROOT) == 0)
1291 break;
1292 if (VN_IS_DOOMED(dp)) { /* forced unmount */
1293 error = ENOENT;
1294 goto bad;
1295 }
1296 tdp = dp;
1297 dp = dp->v_mount->mnt_vnodecovered;
1298 VREF(dp);
1299 vput(tdp);
1300 vn_lock(dp,
1301 enforce_lkflags(dp->v_mount, cnp->cn_lkflags |
1302 LK_RETRY));
1303 }
1304 }
1305
1306 /*
1307 * We now have a segment name to search for, and a directory to search.
1308 */
1309 unionlookup:
1310 #ifdef MAC
1311 error = mac_vnode_check_lookup(cnp->cn_cred, dp, cnp);
1312 if (__predict_false(error))
1313 goto bad;
1314 #endif
1315 ndp->ni_dvp = dp;
1316 ndp->ni_vp = NULL;
1317 ASSERT_VOP_LOCKED(dp, "lookup");
1318 /*
1319 * If we have a shared lock we may need to upgrade the lock for the
1320 * last operation.
1321 */
1322 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
1323 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
1324 vn_lock(dp, LK_UPGRADE|LK_RETRY);
1325 if (VN_IS_DOOMED(dp)) {
1326 error = ENOENT;
1327 goto bad;
1328 }
1329 /*
1330 * If we're looking up the last component and we need an exclusive
1331 * lock, adjust our lkflags.
1332 */
1333 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1334 cnp->cn_lkflags = LK_EXCLUSIVE;
1335 lkflags_save = cnp->cn_lkflags;
1336 cnp->cn_lkflags = enforce_lkflags(dp->v_mount, cnp->cn_lkflags);
1337 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1338 cnp->cn_lkflags = lkflags_save;
1339 if (error != 0) {
1340 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1341 if ((error == ENOENT) &&
1342 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1343 (dp->v_mount->mnt_flag & MNT_UNION)) {
1344 tdp = dp;
1345 dp = dp->v_mount->mnt_vnodecovered;
1346 VREF(dp);
1347 vput(tdp);
1348 vn_lock(dp,
1349 enforce_lkflags(dp->v_mount, cnp->cn_lkflags |
1350 LK_RETRY));
1351 error = nameicap_tracker_add(ndp, dp);
1352 if (error != 0)
1353 goto bad;
1354 goto unionlookup;
1355 }
1356
1357 if (error == ERELOOKUP) {
1358 vref(dp);
1359 ndp->ni_vp = dp;
1360 error = 0;
1361 relookup = 1;
1362 goto good;
1363 }
1364
1365 if (error != EJUSTRETURN)
1366 goto bad;
1367 /*
1368 * At this point, we know we're at the end of the
1369 * pathname. If creating / renaming, we can consider
1370 * allowing the file or directory to be created / renamed,
1371 * provided we're not on a read-only filesystem.
1372 */
1373 if (rdonly) {
1374 error = EROFS;
1375 goto bad;
1376 }
1377 /* trailing slash only allowed for directories */
1378 if ((cnp->cn_flags & TRAILINGSLASH) &&
1379 !(cnp->cn_flags & WILLBEDIR)) {
1380 error = ENOENT;
1381 goto bad;
1382 }
1383 if ((cnp->cn_flags & LOCKPARENT) == 0)
1384 VOP_UNLOCK(dp);
1385 /*
1386 * We return with ni_vp NULL to indicate that the entry
1387 * doesn't currently exist, leaving a pointer to the
1388 * (possibly locked) directory vnode in ndp->ni_dvp.
1389 */
1390 goto success;
1391 }
1392
1393 good:
1394 dp = ndp->ni_vp;
1395
1396 /*
1397 * Check for symbolic link
1398 */
1399 if ((dp->v_type == VLNK) &&
1400 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1401 *ndp->ni_next == '/')) {
1402 cnp->cn_flags |= ISSYMLINK;
1403 if (VN_IS_DOOMED(dp)) {
1404 /*
1405 * We can't know whether the directory was mounted with
1406 * NOSYMFOLLOW, so we can't follow safely.
1407 */
1408 error = ENOENT;
1409 goto bad2;
1410 }
1411 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1412 error = EACCES;
1413 goto bad2;
1414 }
1415 /*
1416 * Symlink code always expects an unlocked dvp.
1417 */
1418 if (ndp->ni_dvp != ndp->ni_vp) {
1419 VOP_UNLOCK(ndp->ni_dvp);
1420 ni_dvp_unlocked = 1;
1421 }
1422 goto success;
1423 }
1424
1425 if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 &&
1426 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1427 error = vfs_lookup_cross_mount(ndp);
1428 if (error != 0)
1429 goto bad_unlocked;
1430 /*
1431 * FALLTHROUGH to nextname
1432 */
1433 dp = ndp->ni_vp;
1434 }
1435
1436 nextname:
1437 /*
1438 * Not a symbolic link that we will follow. Continue with the
1439 * next component if there is any; otherwise, we're done.
1440 */
1441 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1442 ("lookup: invalid path state."));
1443 if (relookup) {
1444 relookup = 0;
1445 ndp->ni_pathlen = prev_ni_pathlen;
1446 ndp->ni_next = prev_ni_next;
1447 if (ndp->ni_dvp != dp)
1448 vput(ndp->ni_dvp);
1449 else
1450 vrele(ndp->ni_dvp);
1451 goto dirloop;
1452 }
1453 if (cnp->cn_flags & ISDOTDOT) {
1454 error = nameicap_check_dotdot(ndp, ndp->ni_dvp);
1455 if (error != 0)
1456 goto bad2;
1457 }
1458 if (*ndp->ni_next == '/') {
1459 cnp->cn_nameptr = ndp->ni_next;
1460 while (*cnp->cn_nameptr == '/') {
1461 cnp->cn_nameptr++;
1462 ndp->ni_pathlen--;
1463 }
1464 if (ndp->ni_dvp != dp)
1465 vput(ndp->ni_dvp);
1466 else
1467 vrele(ndp->ni_dvp);
1468 goto dirloop;
1469 }
1470 /*
1471 * If we're processing a path with a trailing slash,
1472 * check that the end result is a directory.
1473 */
1474 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1475 error = ENOTDIR;
1476 goto bad2;
1477 }
1478 /*
1479 * Disallow directory write attempts on read-only filesystems.
1480 */
1481 if (rdonly &&
1482 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1483 error = EROFS;
1484 goto bad2;
1485 }
1486 if (!wantparent) {
1487 ni_dvp_unlocked = 2;
1488 if (ndp->ni_dvp != dp)
1489 vput(ndp->ni_dvp);
1490 else
1491 vrele(ndp->ni_dvp);
1492 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1493 VOP_UNLOCK(ndp->ni_dvp);
1494 ni_dvp_unlocked = 1;
1495 }
1496
1497 if (cnp->cn_flags & AUDITVNODE1)
1498 AUDIT_ARG_VNODE1(dp);
1499 else if (cnp->cn_flags & AUDITVNODE2)
1500 AUDIT_ARG_VNODE2(dp);
1501
1502 if ((cnp->cn_flags & LOCKLEAF) == 0)
1503 VOP_UNLOCK(dp);
1504 success:
1505 /*
1506 * FIXME: for lookups which only cross a mount point to fetch the
1507 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
1508 * if either WANTPARENT or LOCKPARENT is set.
1509 */
1510 /*
1511 * Because of shared lookup we may have the vnode shared locked, but
1512 * the caller may want it to be exclusively locked.
1513 */
1514 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1515 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1516 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1517 if (VN_IS_DOOMED(dp)) {
1518 error = ENOENT;
1519 goto bad2;
1520 }
1521 }
1522 success_right_lock:
1523 if (ndp->ni_vp != NULL) {
1524 if ((cnp->cn_flags & ISDOTDOT) == 0) {
1525 error = nameicap_tracker_add(ndp, ndp->ni_vp);
1526 if (error != 0)
1527 goto bad2;
1528 }
1529 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS)
1530 return (vfs_lookup_failifexists(ndp));
1531 }
1532 return (0);
1533
1534 bad2:
1535 if (ni_dvp_unlocked != 2) {
1536 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1537 vput(ndp->ni_dvp);
1538 else
1539 vrele(ndp->ni_dvp);
1540 }
1541 bad:
1542 vput(dp);
1543 bad_unlocked:
1544 ndp->ni_vp = NULL;
1545 return (error);
1546 }
1547
1548 /*
1549 * relookup - lookup a path name component
1550 * Used by lookup to re-acquire things.
1551 */
1552 int
vfs_relookup(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,bool refstart)1553 vfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1554 bool refstart)
1555 {
1556 struct vnode *dp = NULL; /* the directory we are searching */
1557 int rdonly; /* lookup read-only flag bit */
1558 int error = 0;
1559
1560 KASSERT(cnp->cn_flags & ISLASTCN,
1561 ("relookup: Not given last component."));
1562 /*
1563 * Setup: break out flag bits into variables.
1564 */
1565 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0,
1566 ("relookup: parent not wanted"));
1567 rdonly = cnp->cn_flags & RDONLY;
1568 cnp->cn_flags &= ~ISSYMLINK;
1569 dp = dvp;
1570 cnp->cn_lkflags = LK_EXCLUSIVE;
1571 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1572
1573 /*
1574 * Search a new directory.
1575 *
1576 * See a comment in vfs_lookup for cnp->cn_nameptr.
1577 *
1578 * Check for "" which represents the root directory after slash
1579 * removal.
1580 */
1581 if (cnp->cn_nameptr[0] == '\0') {
1582 /*
1583 * Support only LOOKUP for "/" because lookup()
1584 * can't succeed for CREATE, DELETE and RENAME.
1585 */
1586 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1587 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1588
1589 if (!(cnp->cn_flags & LOCKLEAF))
1590 VOP_UNLOCK(dp);
1591 *vpp = dp;
1592 /* XXX This should probably move to the top of function. */
1593 if (refstart)
1594 panic("lookup: SAVESTART");
1595 return (0);
1596 }
1597
1598 if (cnp->cn_flags & ISDOTDOT)
1599 panic ("relookup: lookup on dot-dot");
1600
1601 /*
1602 * We now have a segment name to search for, and a directory to search.
1603 */
1604 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1605 KASSERT(*vpp == NULL, ("leaf should be empty"));
1606 if (error != EJUSTRETURN)
1607 goto bad;
1608 /*
1609 * If creating and at end of pathname, then can consider
1610 * allowing file to be created.
1611 */
1612 if (rdonly) {
1613 error = EROFS;
1614 goto bad;
1615 }
1616 /* ASSERT(dvp == ndp->ni_startdir) */
1617 if (refstart)
1618 VREF(dvp);
1619 if ((cnp->cn_flags & LOCKPARENT) == 0)
1620 VOP_UNLOCK(dp);
1621 /*
1622 * We return with ni_vp NULL to indicate that the entry
1623 * doesn't currently exist, leaving a pointer to the
1624 * (possibly locked) directory vnode in ndp->ni_dvp.
1625 */
1626 return (0);
1627 }
1628
1629 dp = *vpp;
1630
1631 /*
1632 * Disallow directory write attempts on read-only filesystems.
1633 */
1634 if (rdonly &&
1635 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1636 if (dvp == dp)
1637 vrele(dvp);
1638 else
1639 vput(dvp);
1640 error = EROFS;
1641 goto bad;
1642 }
1643 /*
1644 * Set the parent lock/ref state to the requested state.
1645 */
1646 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp)
1647 VOP_UNLOCK(dvp);
1648 /*
1649 * Check for symbolic link
1650 */
1651 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1652 ("relookup: symlink found.\n"));
1653
1654 /* ASSERT(dvp == ndp->ni_startdir) */
1655 if (refstart)
1656 VREF(dvp);
1657
1658 if ((cnp->cn_flags & LOCKLEAF) == 0)
1659 VOP_UNLOCK(dp);
1660 return (0);
1661 bad:
1662 vput(dp);
1663 *vpp = NULL;
1664 return (error);
1665 }
1666
1667 #ifdef INVARIANTS
1668 /*
1669 * Validate the final state of ndp after the lookup.
1670 */
1671 static void
NDVALIDATE_impl(struct nameidata * ndp,int line)1672 NDVALIDATE_impl(struct nameidata *ndp, int line)
1673 {
1674 struct componentname *cnp;
1675
1676 cnp = &ndp->ni_cnd;
1677 if (cnp->cn_pnbuf == NULL)
1678 panic("%s: got no buf! called from %d", __func__, line);
1679 }
1680
1681 #endif
1682