1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_capsicum.h"
41 #include "opt_ktrace.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/capsicum.h>
47 #include <sys/fcntl.h>
48 #include <sys/jail.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/vnode.h>
53 #include <sys/mount.h>
54 #include <sys/filedesc.h>
55 #include <sys/proc.h>
56 #include <sys/sdt.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #ifdef KTRACE
60 #include <sys/ktrace.h>
61 #endif
62
63 #include <security/audit/audit.h>
64 #include <security/mac/mac_framework.h>
65
66 #include <vm/uma.h>
67
68 #define NAMEI_DIAGNOSTIC 1
69 #undef NAMEI_DIAGNOSTIC
70
71 SDT_PROVIDER_DECLARE(vfs);
72 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
73 "unsigned long");
74 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
75
76 /* Allocation zone for namei. */
77 uma_zone_t namei_zone;
78
79 /* Placeholder vnode for mp traversal. */
80 static struct vnode *vp_crossmp;
81
82 static int
crossmp_vop_islocked(struct vop_islocked_args * ap)83 crossmp_vop_islocked(struct vop_islocked_args *ap)
84 {
85
86 return (LK_SHARED);
87 }
88
89 static int
crossmp_vop_lock1(struct vop_lock1_args * ap)90 crossmp_vop_lock1(struct vop_lock1_args *ap)
91 {
92 struct vnode *vp;
93 struct lock *lk;
94 const char *file;
95 int flags, line;
96
97 vp = ap->a_vp;
98 lk = vp->v_vnlock;
99 flags = ap->a_flags;
100 file = ap->a_file;
101 line = ap->a_line;
102
103 if ((flags & LK_SHARED) == 0)
104 panic("invalid lock request for crossmp");
105
106 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
107 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
108 WITNESS_LOCK(&lk->lock_object, 0, file, line);
109 if ((flags & LK_INTERLOCK) != 0)
110 VI_UNLOCK(vp);
111 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
112 return (0);
113 }
114
115 static int
crossmp_vop_unlock(struct vop_unlock_args * ap)116 crossmp_vop_unlock(struct vop_unlock_args *ap)
117 {
118 struct vnode *vp;
119 struct lock *lk;
120 int flags;
121
122 vp = ap->a_vp;
123 lk = vp->v_vnlock;
124 flags = ap->a_flags;
125
126 if ((flags & LK_INTERLOCK) != 0)
127 VI_UNLOCK(vp);
128 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
129 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
130 LOCK_LINE);
131 return (0);
132 }
133
134 static struct vop_vector crossmp_vnodeops = {
135 .vop_default = &default_vnodeops,
136 .vop_islocked = crossmp_vop_islocked,
137 .vop_lock1 = crossmp_vop_lock1,
138 .vop_unlock = crossmp_vop_unlock,
139 };
140
141 struct nameicap_tracker {
142 struct vnode *dp;
143 TAILQ_ENTRY(nameicap_tracker) nm_link;
144 };
145
146 /* Zone for cap mode tracker elements used for dotdot capability checks. */
147 static uma_zone_t nt_zone;
148
149 static void
nameiinit(void * dummy __unused)150 nameiinit(void *dummy __unused)
151 {
152
153 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
154 UMA_ALIGN_PTR, 0);
155 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
156 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
157 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
158 }
159 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
160
161 static int lookup_shared = 1;
162 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0,
163 "enables shared locks for path name translation");
164
165 static int lookup_cap_dotdot = 1;
166 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
167 &lookup_cap_dotdot, 0,
168 "enables \"..\" components in path lookup in capability mode");
169 static int lookup_cap_dotdot_nonlocal = 1;
170 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
171 &lookup_cap_dotdot_nonlocal, 0,
172 "enables \"..\" components in path lookup in capability mode "
173 "on non-local mount");
174
175 static void
nameicap_tracker_add(struct nameidata * ndp,struct vnode * dp)176 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
177 {
178 struct nameicap_tracker *nt;
179
180 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
181 return;
182 nt = uma_zalloc(nt_zone, M_WAITOK);
183 vhold(dp);
184 nt->dp = dp;
185 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
186 }
187
188 static void
nameicap_cleanup(struct nameidata * ndp)189 nameicap_cleanup(struct nameidata *ndp)
190 {
191 struct nameicap_tracker *nt, *nt1;
192
193 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
194 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
195 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
196 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
197 vdrop(nt->dp);
198 uma_zfree(nt_zone, nt);
199 }
200 }
201
202 /*
203 * For dotdot lookups in capability mode, only allow the component
204 * lookup to succeed if the resulting directory was already traversed
205 * during the operation. Also fail dotdot lookups for non-local
206 * filesystems, where external agents might assist local lookups to
207 * escape the compartment.
208 */
209 static int
nameicap_check_dotdot(struct nameidata * ndp,struct vnode * dp)210 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
211 {
212 struct nameicap_tracker *nt;
213 struct mount *mp;
214
215 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
216 dp->v_type != VDIR)
217 return (0);
218 mp = dp->v_mount;
219 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
220 (mp->mnt_flag & MNT_LOCAL) == 0)
221 return (ENOTCAPABLE);
222 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
223 nm_link) {
224 if (dp == nt->dp)
225 return (0);
226 }
227 return (ENOTCAPABLE);
228 }
229
230 static void
namei_cleanup_cnp(struct componentname * cnp)231 namei_cleanup_cnp(struct componentname *cnp)
232 {
233
234 uma_zfree(namei_zone, cnp->cn_pnbuf);
235 #ifdef DIAGNOSTIC
236 cnp->cn_pnbuf = NULL;
237 cnp->cn_nameptr = NULL;
238 #endif
239 }
240
241 static int
namei_handle_root(struct nameidata * ndp,struct vnode ** dpp)242 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
243 {
244 struct componentname *cnp;
245
246 cnp = &ndp->ni_cnd;
247 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
248 #ifdef KTRACE
249 if (KTRPOINT(curthread, KTR_CAPFAIL))
250 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
251 #endif
252 return (ENOTCAPABLE);
253 }
254 while (*(cnp->cn_nameptr) == '/') {
255 cnp->cn_nameptr++;
256 ndp->ni_pathlen--;
257 }
258 *dpp = ndp->ni_rootdir;
259 vrefact(*dpp);
260 return (0);
261 }
262
263 /*
264 * Convert a pathname into a pointer to a locked vnode.
265 *
266 * The FOLLOW flag is set when symbolic links are to be followed
267 * when they occur at the end of the name translation process.
268 * Symbolic links are always followed for all other pathname
269 * components other than the last.
270 *
271 * The segflg defines whether the name is to be copied from user
272 * space or kernel space.
273 *
274 * Overall outline of namei:
275 *
276 * copy in name
277 * get starting directory
278 * while (!done && !error) {
279 * call lookup to search path.
280 * if symbolic link, massage name in buffer and continue
281 * }
282 */
283 int
namei(struct nameidata * ndp)284 namei(struct nameidata *ndp)
285 {
286 struct filedesc *fdp; /* pointer to file descriptor state */
287 char *cp; /* pointer into pathname argument */
288 struct vnode *dp; /* the directory we are searching */
289 struct iovec aiov; /* uio for reading symbolic links */
290 struct componentname *cnp;
291 struct file *dfp;
292 struct thread *td;
293 struct proc *p;
294 cap_rights_t rights;
295 struct uio auio;
296 int error, linklen, startdir_used;
297
298 cnp = &ndp->ni_cnd;
299 td = cnp->cn_thread;
300 p = td->td_proc;
301 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
302 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
303 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
304 ("namei: nameiop contaminated with flags"));
305 KASSERT((cnp->cn_flags & OPMASK) == 0,
306 ("namei: flags contaminated with nameiops"));
307 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
308 ndp->ni_startdir->v_type == VBAD);
309 if (!lookup_shared)
310 cnp->cn_flags &= ~LOCKSHARED;
311 fdp = p->p_fd;
312 TAILQ_INIT(&ndp->ni_cap_tracker);
313 ndp->ni_lcf = 0;
314
315 /* We will set this ourselves if we need it. */
316 cnp->cn_flags &= ~TRAILINGSLASH;
317
318 /*
319 * Get a buffer for the name to be translated, and copy the
320 * name into the buffer.
321 */
322 if ((cnp->cn_flags & HASBUF) == 0)
323 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
324 if (ndp->ni_segflg == UIO_SYSSPACE)
325 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
326 &ndp->ni_pathlen);
327 else
328 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
329 &ndp->ni_pathlen);
330
331 /*
332 * Don't allow empty pathnames.
333 */
334 if (error == 0 && *cnp->cn_pnbuf == '\0')
335 error = ENOENT;
336
337 #ifdef CAPABILITY_MODE
338 /*
339 * In capability mode, lookups must be restricted to happen in
340 * the subtree with the root specified by the file descriptor:
341 * - The root must be real file descriptor, not the pseudo-descriptor
342 * AT_FDCWD.
343 * - The passed path must be relative and not absolute.
344 * - If lookup_cap_dotdot is disabled, path must not contain the
345 * '..' components.
346 * - If lookup_cap_dotdot is enabled, we verify that all '..'
347 * components lookups result in the directories which were
348 * previously walked by us, which prevents an escape from
349 * the relative root.
350 */
351 if (error == 0 && IN_CAPABILITY_MODE(td) &&
352 (cnp->cn_flags & NOCAPCHECK) == 0) {
353 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
354 if (ndp->ni_dirfd == AT_FDCWD) {
355 #ifdef KTRACE
356 if (KTRPOINT(td, KTR_CAPFAIL))
357 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
358 #endif
359 error = ECAPMODE;
360 }
361 }
362 #endif
363 if (error != 0) {
364 namei_cleanup_cnp(cnp);
365 ndp->ni_vp = NULL;
366 return (error);
367 }
368 ndp->ni_loopcnt = 0;
369 #ifdef KTRACE
370 if (KTRPOINT(td, KTR_NAMEI)) {
371 KASSERT(cnp->cn_thread == curthread,
372 ("namei not using curthread"));
373 ktrnamei(cnp->cn_pnbuf);
374 }
375 #endif
376 /*
377 * Get starting point for the translation.
378 */
379 FILEDESC_SLOCK(fdp);
380 ndp->ni_rootdir = fdp->fd_rdir;
381 vrefact(ndp->ni_rootdir);
382 ndp->ni_topdir = fdp->fd_jdir;
383
384 /*
385 * If we are auditing the kernel pathname, save the user pathname.
386 */
387 if (cnp->cn_flags & AUDITVNODE1)
388 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
389 if (cnp->cn_flags & AUDITVNODE2)
390 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
391
392 startdir_used = 0;
393 dp = NULL;
394 cnp->cn_nameptr = cnp->cn_pnbuf;
395 if (cnp->cn_pnbuf[0] == '/') {
396 error = namei_handle_root(ndp, &dp);
397 } else {
398 if (ndp->ni_startdir != NULL) {
399 dp = ndp->ni_startdir;
400 startdir_used = 1;
401 } else if (ndp->ni_dirfd == AT_FDCWD) {
402 dp = fdp->fd_cdir;
403 vrefact(dp);
404 } else {
405 rights = ndp->ni_rightsneeded;
406 cap_rights_set(&rights, CAP_LOOKUP);
407
408 if (cnp->cn_flags & AUDITVNODE1)
409 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
410 if (cnp->cn_flags & AUDITVNODE2)
411 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
412 /*
413 * Effectively inlined fgetvp_rights, because we need to
414 * inspect the file as well as grabbing the vnode.
415 */
416 error = fget_cap_locked(fdp, ndp->ni_dirfd, &rights,
417 &dfp, &ndp->ni_filecaps);
418 if (error != 0) {
419 /*
420 * Preserve the error; it should either be EBADF
421 * or capability-related, both of which can be
422 * safely returned to the caller.
423 */
424 } else if (dfp->f_ops == &badfileops) {
425 error = EBADF;
426 } else if (dfp->f_vnode == NULL) {
427 error = ENOTDIR;
428 } else {
429 dp = dfp->f_vnode;
430 vrefact(dp);
431
432 if ((dfp->f_flag & FSEARCH) != 0)
433 cnp->cn_flags |= NOEXECCHECK;
434 }
435 #ifdef CAPABILITIES
436 /*
437 * If file descriptor doesn't have all rights,
438 * all lookups relative to it must also be
439 * strictly relative.
440 */
441 CAP_ALL(&rights);
442 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
443 &rights) ||
444 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
445 ndp->ni_filecaps.fc_nioctls != -1) {
446 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
447 }
448 #endif
449 }
450 if (error == 0 && dp->v_type != VDIR)
451 error = ENOTDIR;
452 }
453 FILEDESC_SUNLOCK(fdp);
454 if (ndp->ni_startdir != NULL && !startdir_used)
455 vrele(ndp->ni_startdir);
456 if (error != 0) {
457 if (dp != NULL)
458 vrele(dp);
459 goto out;
460 }
461 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
462 lookup_cap_dotdot != 0)
463 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
464 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
465 cnp->cn_flags);
466 for (;;) {
467 ndp->ni_startdir = dp;
468 error = lookup(ndp);
469 if (error != 0)
470 goto out;
471 /*
472 * If not a symbolic link, we're done.
473 */
474 if ((cnp->cn_flags & ISSYMLINK) == 0) {
475 vrele(ndp->ni_rootdir);
476 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
477 namei_cleanup_cnp(cnp);
478 } else
479 cnp->cn_flags |= HASBUF;
480 nameicap_cleanup(ndp);
481 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
482 return (0);
483 }
484 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
485 error = ELOOP;
486 break;
487 }
488 #ifdef MAC
489 if ((cnp->cn_flags & NOMACCHECK) == 0) {
490 error = mac_vnode_check_readlink(td->td_ucred,
491 ndp->ni_vp);
492 if (error != 0)
493 break;
494 }
495 #endif
496 if (ndp->ni_pathlen > 1)
497 cp = uma_zalloc(namei_zone, M_WAITOK);
498 else
499 cp = cnp->cn_pnbuf;
500 aiov.iov_base = cp;
501 aiov.iov_len = MAXPATHLEN;
502 auio.uio_iov = &aiov;
503 auio.uio_iovcnt = 1;
504 auio.uio_offset = 0;
505 auio.uio_rw = UIO_READ;
506 auio.uio_segflg = UIO_SYSSPACE;
507 auio.uio_td = td;
508 auio.uio_resid = MAXPATHLEN;
509 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
510 if (error != 0) {
511 if (ndp->ni_pathlen > 1)
512 uma_zfree(namei_zone, cp);
513 break;
514 }
515 linklen = MAXPATHLEN - auio.uio_resid;
516 if (linklen == 0) {
517 if (ndp->ni_pathlen > 1)
518 uma_zfree(namei_zone, cp);
519 error = ENOENT;
520 break;
521 }
522 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
523 if (ndp->ni_pathlen > 1)
524 uma_zfree(namei_zone, cp);
525 error = ENAMETOOLONG;
526 break;
527 }
528 if (ndp->ni_pathlen > 1) {
529 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
530 uma_zfree(namei_zone, cnp->cn_pnbuf);
531 cnp->cn_pnbuf = cp;
532 } else
533 cnp->cn_pnbuf[linklen] = '\0';
534 ndp->ni_pathlen += linklen;
535 vput(ndp->ni_vp);
536 dp = ndp->ni_dvp;
537 /*
538 * Check if root directory should replace current directory.
539 */
540 cnp->cn_nameptr = cnp->cn_pnbuf;
541 if (*(cnp->cn_nameptr) == '/') {
542 vrele(dp);
543 error = namei_handle_root(ndp, &dp);
544 if (error != 0)
545 goto out;
546 }
547 }
548 vput(ndp->ni_vp);
549 ndp->ni_vp = NULL;
550 vrele(ndp->ni_dvp);
551 out:
552 vrele(ndp->ni_rootdir);
553 namei_cleanup_cnp(cnp);
554 nameicap_cleanup(ndp);
555 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
556 return (error);
557 }
558
559 static int
compute_cn_lkflags(struct mount * mp,int lkflags,int cnflags)560 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
561 {
562
563 if (mp == NULL || ((lkflags & LK_SHARED) &&
564 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
565 ((cnflags & ISDOTDOT) &&
566 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
567 lkflags &= ~LK_SHARED;
568 lkflags |= LK_EXCLUSIVE;
569 }
570 lkflags |= LK_NODDLKTREAT;
571 return (lkflags);
572 }
573
574 static __inline int
needs_exclusive_leaf(struct mount * mp,int flags)575 needs_exclusive_leaf(struct mount *mp, int flags)
576 {
577
578 /*
579 * Intermediate nodes can use shared locks, we only need to
580 * force an exclusive lock for leaf nodes.
581 */
582 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
583 return (0);
584
585 /* Always use exclusive locks if LOCKSHARED isn't set. */
586 if (!(flags & LOCKSHARED))
587 return (1);
588
589 /*
590 * For lookups during open(), if the mount point supports
591 * extended shared operations, then use a shared lock for the
592 * leaf node, otherwise use an exclusive lock.
593 */
594 if ((flags & ISOPEN) != 0)
595 return (!MNT_EXTENDED_SHARED(mp));
596
597 /*
598 * Lookup requests outside of open() that specify LOCKSHARED
599 * only need a shared lock on the leaf vnode.
600 */
601 return (0);
602 }
603
604 /*
605 * Search a pathname.
606 * This is a very central and rather complicated routine.
607 *
608 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
609 * The starting directory is taken from ni_startdir. The pathname is
610 * descended until done, or a symbolic link is encountered. The variable
611 * ni_more is clear if the path is completed; it is set to one if a
612 * symbolic link needing interpretation is encountered.
613 *
614 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
615 * whether the name is to be looked up, created, renamed, or deleted.
616 * When CREATE, RENAME, or DELETE is specified, information usable in
617 * creating, renaming, or deleting a directory entry may be calculated.
618 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
619 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
620 * returned unlocked. Otherwise the parent directory is not returned. If
621 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
622 * the target is returned locked, otherwise it is returned unlocked.
623 * When creating or renaming and LOCKPARENT is specified, the target may not
624 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
625 *
626 * Overall outline of lookup:
627 *
628 * dirloop:
629 * identify next component of name at ndp->ni_ptr
630 * handle degenerate case where name is null string
631 * if .. and crossing mount points and on mounted filesys, find parent
632 * call VOP_LOOKUP routine for next component name
633 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
634 * component vnode returned in ni_vp (if it exists), locked.
635 * if result vnode is mounted on and crossing mount points,
636 * find mounted on vnode
637 * if more components of name, do next level at dirloop
638 * return the answer in ni_vp, locked if LOCKLEAF set
639 * if LOCKPARENT set, return locked parent in ni_dvp
640 * if WANTPARENT set, return unlocked parent in ni_dvp
641 */
642 int
lookup(struct nameidata * ndp)643 lookup(struct nameidata *ndp)
644 {
645 char *cp; /* pointer into pathname argument */
646 char *prev_ni_next; /* saved ndp->ni_next */
647 struct vnode *dp = NULL; /* the directory we are searching */
648 struct vnode *tdp; /* saved dp */
649 struct mount *mp; /* mount table entry */
650 struct prison *pr;
651 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
652 int docache; /* == 0 do not cache last component */
653 int wantparent; /* 1 => wantparent or lockparent flag */
654 int rdonly; /* lookup read-only flag bit */
655 int error = 0;
656 int dpunlocked = 0; /* dp has already been unlocked */
657 int relookup = 0; /* do not consume the path component */
658 struct componentname *cnp = &ndp->ni_cnd;
659 int lkflags_save;
660 int ni_dvp_unlocked;
661
662 /*
663 * Setup: break out flag bits into variables.
664 */
665 ni_dvp_unlocked = 0;
666 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
667 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
668 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
669 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
670 if (cnp->cn_nameiop == DELETE ||
671 (wantparent && cnp->cn_nameiop != CREATE &&
672 cnp->cn_nameiop != LOOKUP))
673 docache = 0;
674 rdonly = cnp->cn_flags & RDONLY;
675 cnp->cn_flags &= ~ISSYMLINK;
676 ndp->ni_dvp = NULL;
677 /*
678 * We use shared locks until we hit the parent of the last cn then
679 * we adjust based on the requesting flags.
680 */
681 if (lookup_shared)
682 cnp->cn_lkflags = LK_SHARED;
683 else
684 cnp->cn_lkflags = LK_EXCLUSIVE;
685 dp = ndp->ni_startdir;
686 ndp->ni_startdir = NULLVP;
687 vn_lock(dp,
688 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
689 cnp->cn_flags));
690
691 dirloop:
692 /*
693 * Search a new directory.
694 *
695 * The last component of the filename is left accessible via
696 * cnp->cn_nameptr for callers that need the name. Callers needing
697 * the name set the SAVENAME flag. When done, they assume
698 * responsibility for freeing the pathname buffer.
699 */
700 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
701 continue;
702 cnp->cn_namelen = cp - cnp->cn_nameptr;
703 if (cnp->cn_namelen > NAME_MAX) {
704 error = ENAMETOOLONG;
705 goto bad;
706 }
707 #ifdef NAMEI_DIAGNOSTIC
708 { char c = *cp;
709 *cp = '\0';
710 printf("{%s}: ", cnp->cn_nameptr);
711 *cp = c; }
712 #endif
713 prev_ni_pathlen = ndp->ni_pathlen;
714 ndp->ni_pathlen -= cnp->cn_namelen;
715 KASSERT(ndp->ni_pathlen <= PATH_MAX,
716 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
717 prev_ni_next = ndp->ni_next;
718 ndp->ni_next = cp;
719
720 /*
721 * Replace multiple slashes by a single slash and trailing slashes
722 * by a null. This must be done before VOP_LOOKUP() because some
723 * fs's don't know about trailing slashes. Remember if there were
724 * trailing slashes to handle symlinks, existing non-directories
725 * and non-existing files that won't be directories specially later.
726 */
727 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
728 cp++;
729 ndp->ni_pathlen--;
730 if (*cp == '\0') {
731 *ndp->ni_next = '\0';
732 cnp->cn_flags |= TRAILINGSLASH;
733 }
734 }
735 ndp->ni_next = cp;
736
737 cnp->cn_flags |= MAKEENTRY;
738 if (*cp == '\0' && docache == 0)
739 cnp->cn_flags &= ~MAKEENTRY;
740 if (cnp->cn_namelen == 2 &&
741 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
742 cnp->cn_flags |= ISDOTDOT;
743 else
744 cnp->cn_flags &= ~ISDOTDOT;
745 if (*ndp->ni_next == 0)
746 cnp->cn_flags |= ISLASTCN;
747 else
748 cnp->cn_flags &= ~ISLASTCN;
749
750 if ((cnp->cn_flags & ISLASTCN) != 0 &&
751 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
752 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
753 error = EINVAL;
754 goto bad;
755 }
756
757 nameicap_tracker_add(ndp, dp);
758
759 /*
760 * Check for degenerate name (e.g. / or "")
761 * which is a way of talking about a directory,
762 * e.g. like "/." or ".".
763 */
764 if (cnp->cn_nameptr[0] == '\0') {
765 if (dp->v_type != VDIR) {
766 error = ENOTDIR;
767 goto bad;
768 }
769 if (cnp->cn_nameiop != LOOKUP) {
770 error = EISDIR;
771 goto bad;
772 }
773 if (wantparent) {
774 ndp->ni_dvp = dp;
775 VREF(dp);
776 }
777 ndp->ni_vp = dp;
778
779 if (cnp->cn_flags & AUDITVNODE1)
780 AUDIT_ARG_VNODE1(dp);
781 else if (cnp->cn_flags & AUDITVNODE2)
782 AUDIT_ARG_VNODE2(dp);
783
784 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
785 VOP_UNLOCK(dp, 0);
786 /* XXX This should probably move to the top of function. */
787 if (cnp->cn_flags & SAVESTART)
788 panic("lookup: SAVESTART");
789 goto success;
790 }
791
792 /*
793 * Handle "..": five special cases.
794 * 0. If doing a capability lookup and lookup_cap_dotdot is
795 * disabled, return ENOTCAPABLE.
796 * 1. Return an error if this is the last component of
797 * the name and the operation is DELETE or RENAME.
798 * 2. If at root directory (e.g. after chroot)
799 * or at absolute root directory
800 * then ignore it so can't get out.
801 * 3. If this vnode is the root of a mounted
802 * filesystem, then replace it with the
803 * vnode which was mounted on so we take the
804 * .. in the other filesystem.
805 * 4. If the vnode is the top directory of
806 * the jail or chroot, don't let them out.
807 * 5. If doing a capability lookup and lookup_cap_dotdot is
808 * enabled, return ENOTCAPABLE if the lookup would escape
809 * from the initial file descriptor directory. Checks are
810 * done by ensuring that namei() already traversed the
811 * result of dotdot lookup.
812 */
813 if (cnp->cn_flags & ISDOTDOT) {
814 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
815 == NI_LCF_STRICTRELATIVE) {
816 #ifdef KTRACE
817 if (KTRPOINT(curthread, KTR_CAPFAIL))
818 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
819 #endif
820 error = ENOTCAPABLE;
821 goto bad;
822 }
823 if ((cnp->cn_flags & ISLASTCN) != 0 &&
824 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
825 error = EINVAL;
826 goto bad;
827 }
828 for (;;) {
829 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
830 pr = pr->pr_parent)
831 if (dp == pr->pr_root)
832 break;
833 if (dp == ndp->ni_rootdir ||
834 dp == ndp->ni_topdir ||
835 dp == rootvnode ||
836 pr != NULL ||
837 ((dp->v_vflag & VV_ROOT) != 0 &&
838 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
839 ndp->ni_dvp = dp;
840 ndp->ni_vp = dp;
841 VREF(dp);
842 goto nextname;
843 }
844 if ((dp->v_vflag & VV_ROOT) == 0)
845 break;
846 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
847 error = ENOENT;
848 goto bad;
849 }
850 tdp = dp;
851 dp = dp->v_mount->mnt_vnodecovered;
852 VREF(dp);
853 vput(tdp);
854 vn_lock(dp,
855 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
856 LK_RETRY, ISDOTDOT));
857 error = nameicap_check_dotdot(ndp, dp);
858 if (error != 0) {
859 #ifdef KTRACE
860 if (KTRPOINT(curthread, KTR_CAPFAIL))
861 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
862 #endif
863 goto bad;
864 }
865 }
866 }
867
868 /*
869 * We now have a segment name to search for, and a directory to search.
870 */
871 unionlookup:
872 #ifdef MAC
873 if ((cnp->cn_flags & NOMACCHECK) == 0) {
874 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
875 cnp);
876 if (error)
877 goto bad;
878 }
879 #endif
880 ndp->ni_dvp = dp;
881 ndp->ni_vp = NULL;
882 ASSERT_VOP_LOCKED(dp, "lookup");
883 /*
884 * If we have a shared lock we may need to upgrade the lock for the
885 * last operation.
886 */
887 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
888 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
889 vn_lock(dp, LK_UPGRADE|LK_RETRY);
890 if ((dp->v_iflag & VI_DOOMED) != 0) {
891 error = ENOENT;
892 goto bad;
893 }
894 /*
895 * If we're looking up the last component and we need an exclusive
896 * lock, adjust our lkflags.
897 */
898 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
899 cnp->cn_lkflags = LK_EXCLUSIVE;
900 #ifdef NAMEI_DIAGNOSTIC
901 vn_printf(dp, "lookup in ");
902 #endif
903 lkflags_save = cnp->cn_lkflags;
904 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
905 cnp->cn_flags);
906 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
907 cnp->cn_lkflags = lkflags_save;
908 if (error != 0) {
909 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
910 #ifdef NAMEI_DIAGNOSTIC
911 printf("not found\n");
912 #endif
913 if ((error == ENOENT) &&
914 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
915 (dp->v_mount->mnt_flag & MNT_UNION)) {
916 tdp = dp;
917 dp = dp->v_mount->mnt_vnodecovered;
918 VREF(dp);
919 vput(tdp);
920 vn_lock(dp,
921 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
922 LK_RETRY, cnp->cn_flags));
923 nameicap_tracker_add(ndp, dp);
924 goto unionlookup;
925 }
926
927 if (error == ERELOOKUP) {
928 vref(dp);
929 ndp->ni_vp = dp;
930 error = 0;
931 relookup = 1;
932 goto good;
933 }
934
935 if (error != EJUSTRETURN)
936 goto bad;
937 /*
938 * At this point, we know we're at the end of the
939 * pathname. If creating / renaming, we can consider
940 * allowing the file or directory to be created / renamed,
941 * provided we're not on a read-only filesystem.
942 */
943 if (rdonly) {
944 error = EROFS;
945 goto bad;
946 }
947 /* trailing slash only allowed for directories */
948 if ((cnp->cn_flags & TRAILINGSLASH) &&
949 !(cnp->cn_flags & WILLBEDIR)) {
950 error = ENOENT;
951 goto bad;
952 }
953 if ((cnp->cn_flags & LOCKPARENT) == 0)
954 VOP_UNLOCK(dp, 0);
955 /*
956 * We return with ni_vp NULL to indicate that the entry
957 * doesn't currently exist, leaving a pointer to the
958 * (possibly locked) directory vnode in ndp->ni_dvp.
959 */
960 if (cnp->cn_flags & SAVESTART) {
961 ndp->ni_startdir = ndp->ni_dvp;
962 VREF(ndp->ni_startdir);
963 }
964 goto success;
965 }
966
967 good:
968 #ifdef NAMEI_DIAGNOSTIC
969 printf("found\n");
970 #endif
971 dp = ndp->ni_vp;
972
973 /*
974 * Check to see if the vnode has been mounted on;
975 * if so find the root of the mounted filesystem.
976 */
977 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
978 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
979 if (vfs_busy(mp, 0))
980 continue;
981 vput(dp);
982 if (dp != ndp->ni_dvp)
983 vput(ndp->ni_dvp);
984 else
985 vrele(ndp->ni_dvp);
986 vrefact(vp_crossmp);
987 ndp->ni_dvp = vp_crossmp;
988 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
989 cnp->cn_flags), &tdp);
990 vfs_unbusy(mp);
991 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
992 panic("vp_crossmp exclusively locked or reclaimed");
993 if (error) {
994 dpunlocked = 1;
995 goto bad2;
996 }
997 ndp->ni_vp = dp = tdp;
998 }
999
1000 /*
1001 * Check for symbolic link
1002 */
1003 if ((dp->v_type == VLNK) &&
1004 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1005 *ndp->ni_next == '/')) {
1006 cnp->cn_flags |= ISSYMLINK;
1007 if (dp->v_iflag & VI_DOOMED) {
1008 /*
1009 * We can't know whether the directory was mounted with
1010 * NOSYMFOLLOW, so we can't follow safely.
1011 */
1012 error = ENOENT;
1013 goto bad2;
1014 }
1015 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1016 error = EACCES;
1017 goto bad2;
1018 }
1019 /*
1020 * Symlink code always expects an unlocked dvp.
1021 */
1022 if (ndp->ni_dvp != ndp->ni_vp) {
1023 VOP_UNLOCK(ndp->ni_dvp, 0);
1024 ni_dvp_unlocked = 1;
1025 }
1026 goto success;
1027 }
1028
1029 nextname:
1030 /*
1031 * Not a symbolic link that we will follow. Continue with the
1032 * next component if there is any; otherwise, we're done.
1033 */
1034 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1035 ("lookup: invalid path state."));
1036 if (relookup) {
1037 relookup = 0;
1038 ndp->ni_pathlen = prev_ni_pathlen;
1039 ndp->ni_next = prev_ni_next;
1040 if (ndp->ni_dvp != dp)
1041 vput(ndp->ni_dvp);
1042 else
1043 vrele(ndp->ni_dvp);
1044 goto dirloop;
1045 }
1046 if (cnp->cn_flags & ISDOTDOT) {
1047 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1048 if (error != 0) {
1049 #ifdef KTRACE
1050 if (KTRPOINT(curthread, KTR_CAPFAIL))
1051 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1052 #endif
1053 goto bad2;
1054 }
1055 }
1056 if (*ndp->ni_next == '/') {
1057 cnp->cn_nameptr = ndp->ni_next;
1058 while (*cnp->cn_nameptr == '/') {
1059 cnp->cn_nameptr++;
1060 ndp->ni_pathlen--;
1061 }
1062 if (ndp->ni_dvp != dp)
1063 vput(ndp->ni_dvp);
1064 else
1065 vrele(ndp->ni_dvp);
1066 goto dirloop;
1067 }
1068 /*
1069 * If we're processing a path with a trailing slash,
1070 * check that the end result is a directory.
1071 */
1072 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1073 error = ENOTDIR;
1074 goto bad2;
1075 }
1076 /*
1077 * Disallow directory write attempts on read-only filesystems.
1078 */
1079 if (rdonly &&
1080 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1081 error = EROFS;
1082 goto bad2;
1083 }
1084 if (cnp->cn_flags & SAVESTART) {
1085 ndp->ni_startdir = ndp->ni_dvp;
1086 VREF(ndp->ni_startdir);
1087 }
1088 if (!wantparent) {
1089 ni_dvp_unlocked = 2;
1090 if (ndp->ni_dvp != dp)
1091 vput(ndp->ni_dvp);
1092 else
1093 vrele(ndp->ni_dvp);
1094 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1095 VOP_UNLOCK(ndp->ni_dvp, 0);
1096 ni_dvp_unlocked = 1;
1097 }
1098
1099 if (cnp->cn_flags & AUDITVNODE1)
1100 AUDIT_ARG_VNODE1(dp);
1101 else if (cnp->cn_flags & AUDITVNODE2)
1102 AUDIT_ARG_VNODE2(dp);
1103
1104 if ((cnp->cn_flags & LOCKLEAF) == 0)
1105 VOP_UNLOCK(dp, 0);
1106 success:
1107 /*
1108 * Because of lookup_shared we may have the vnode shared locked, but
1109 * the caller may want it to be exclusively locked.
1110 */
1111 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1112 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1113 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1114 if (dp->v_iflag & VI_DOOMED) {
1115 error = ENOENT;
1116 goto bad2;
1117 }
1118 }
1119 return (0);
1120
1121 bad2:
1122 if (ni_dvp_unlocked != 2) {
1123 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1124 vput(ndp->ni_dvp);
1125 else
1126 vrele(ndp->ni_dvp);
1127 }
1128 bad:
1129 if (!dpunlocked)
1130 vput(dp);
1131 ndp->ni_vp = NULL;
1132 return (error);
1133 }
1134
1135 /*
1136 * relookup - lookup a path name component
1137 * Used by lookup to re-acquire things.
1138 */
1139 int
relookup(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp)1140 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1141 {
1142 struct vnode *dp = NULL; /* the directory we are searching */
1143 int wantparent; /* 1 => wantparent or lockparent flag */
1144 int rdonly; /* lookup read-only flag bit */
1145 int error = 0;
1146
1147 KASSERT(cnp->cn_flags & ISLASTCN,
1148 ("relookup: Not given last component."));
1149 /*
1150 * Setup: break out flag bits into variables.
1151 */
1152 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1153 KASSERT(wantparent, ("relookup: parent not wanted."));
1154 rdonly = cnp->cn_flags & RDONLY;
1155 cnp->cn_flags &= ~ISSYMLINK;
1156 dp = dvp;
1157 cnp->cn_lkflags = LK_EXCLUSIVE;
1158 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1159
1160 /*
1161 * Search a new directory.
1162 *
1163 * The last component of the filename is left accessible via
1164 * cnp->cn_nameptr for callers that need the name. Callers needing
1165 * the name set the SAVENAME flag. When done, they assume
1166 * responsibility for freeing the pathname buffer.
1167 */
1168 #ifdef NAMEI_DIAGNOSTIC
1169 printf("{%s}: ", cnp->cn_nameptr);
1170 #endif
1171
1172 /*
1173 * Check for "" which represents the root directory after slash
1174 * removal.
1175 */
1176 if (cnp->cn_nameptr[0] == '\0') {
1177 /*
1178 * Support only LOOKUP for "/" because lookup()
1179 * can't succeed for CREATE, DELETE and RENAME.
1180 */
1181 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1182 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1183
1184 if (!(cnp->cn_flags & LOCKLEAF))
1185 VOP_UNLOCK(dp, 0);
1186 *vpp = dp;
1187 /* XXX This should probably move to the top of function. */
1188 if (cnp->cn_flags & SAVESTART)
1189 panic("lookup: SAVESTART");
1190 return (0);
1191 }
1192
1193 if (cnp->cn_flags & ISDOTDOT)
1194 panic ("relookup: lookup on dot-dot");
1195
1196 /*
1197 * We now have a segment name to search for, and a directory to search.
1198 */
1199 #ifdef NAMEI_DIAGNOSTIC
1200 vn_printf(dp, "search in ");
1201 #endif
1202 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1203 KASSERT(*vpp == NULL, ("leaf should be empty"));
1204 if (error != EJUSTRETURN)
1205 goto bad;
1206 /*
1207 * If creating and at end of pathname, then can consider
1208 * allowing file to be created.
1209 */
1210 if (rdonly) {
1211 error = EROFS;
1212 goto bad;
1213 }
1214 /* ASSERT(dvp == ndp->ni_startdir) */
1215 if (cnp->cn_flags & SAVESTART)
1216 VREF(dvp);
1217 if ((cnp->cn_flags & LOCKPARENT) == 0)
1218 VOP_UNLOCK(dp, 0);
1219 /*
1220 * We return with ni_vp NULL to indicate that the entry
1221 * doesn't currently exist, leaving a pointer to the
1222 * (possibly locked) directory vnode in ndp->ni_dvp.
1223 */
1224 return (0);
1225 }
1226
1227 dp = *vpp;
1228
1229 /*
1230 * Disallow directory write attempts on read-only filesystems.
1231 */
1232 if (rdonly &&
1233 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1234 if (dvp == dp)
1235 vrele(dvp);
1236 else
1237 vput(dvp);
1238 error = EROFS;
1239 goto bad;
1240 }
1241 /*
1242 * Set the parent lock/ref state to the requested state.
1243 */
1244 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1245 if (wantparent)
1246 VOP_UNLOCK(dvp, 0);
1247 else
1248 vput(dvp);
1249 } else if (!wantparent)
1250 vrele(dvp);
1251 /*
1252 * Check for symbolic link
1253 */
1254 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1255 ("relookup: symlink found.\n"));
1256
1257 /* ASSERT(dvp == ndp->ni_startdir) */
1258 if (cnp->cn_flags & SAVESTART)
1259 VREF(dvp);
1260
1261 if ((cnp->cn_flags & LOCKLEAF) == 0)
1262 VOP_UNLOCK(dp, 0);
1263 return (0);
1264 bad:
1265 vput(dp);
1266 *vpp = NULL;
1267 return (error);
1268 }
1269
1270 void
NDINIT_ALL(struct nameidata * ndp,u_long op,u_long flags,enum uio_seg segflg,const char * namep,int dirfd,struct vnode * startdir,cap_rights_t * rightsp,struct thread * td)1271 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1272 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1273 struct thread *td)
1274 {
1275
1276 ndp->ni_cnd.cn_nameiop = op;
1277 ndp->ni_cnd.cn_flags = flags;
1278 ndp->ni_segflg = segflg;
1279 ndp->ni_dirp = namep;
1280 ndp->ni_dirfd = dirfd;
1281 ndp->ni_startdir = startdir;
1282 if (rightsp != NULL)
1283 ndp->ni_rightsneeded = *rightsp;
1284 else
1285 cap_rights_init(&ndp->ni_rightsneeded);
1286 filecaps_init(&ndp->ni_filecaps);
1287 ndp->ni_cnd.cn_thread = td;
1288 }
1289
1290 /*
1291 * Free data allocated by namei(); see namei(9) for details.
1292 */
1293 void
NDFREE(struct nameidata * ndp,const u_int flags)1294 NDFREE(struct nameidata *ndp, const u_int flags)
1295 {
1296 int unlock_dvp;
1297 int unlock_vp;
1298
1299 unlock_dvp = 0;
1300 unlock_vp = 0;
1301
1302 if (!(flags & NDF_NO_FREE_PNBUF) &&
1303 (ndp->ni_cnd.cn_flags & HASBUF)) {
1304 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1305 ndp->ni_cnd.cn_flags &= ~HASBUF;
1306 }
1307 if (!(flags & NDF_NO_VP_UNLOCK) &&
1308 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1309 unlock_vp = 1;
1310 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1311 if (unlock_vp) {
1312 vput(ndp->ni_vp);
1313 unlock_vp = 0;
1314 } else
1315 vrele(ndp->ni_vp);
1316 ndp->ni_vp = NULL;
1317 }
1318 if (unlock_vp)
1319 VOP_UNLOCK(ndp->ni_vp, 0);
1320 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1321 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1322 ndp->ni_dvp != ndp->ni_vp)
1323 unlock_dvp = 1;
1324 if (!(flags & NDF_NO_DVP_RELE) &&
1325 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1326 if (unlock_dvp) {
1327 vput(ndp->ni_dvp);
1328 unlock_dvp = 0;
1329 } else
1330 vrele(ndp->ni_dvp);
1331 ndp->ni_dvp = NULL;
1332 }
1333 if (unlock_dvp)
1334 VOP_UNLOCK(ndp->ni_dvp, 0);
1335 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1336 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1337 vrele(ndp->ni_startdir);
1338 ndp->ni_startdir = NULL;
1339 }
1340 }
1341
1342 /*
1343 * Determine if there is a suitable alternate filename under the specified
1344 * prefix for the specified path. If the create flag is set, then the
1345 * alternate prefix will be used so long as the parent directory exists.
1346 * This is used by the various compatibility ABIs so that Linux binaries prefer
1347 * files under /compat/linux for example. The chosen path (whether under
1348 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1349 * to by pathbuf. The caller is responsible for free'ing the buffer from
1350 * the M_TEMP bucket if one is returned.
1351 */
1352 int
kern_alternate_path(struct thread * td,const char * prefix,const char * path,enum uio_seg pathseg,char ** pathbuf,int create,int dirfd)1353 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1354 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1355 {
1356 struct nameidata nd, ndroot;
1357 char *ptr, *buf, *cp;
1358 size_t len, sz;
1359 int error;
1360
1361 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1362 *pathbuf = buf;
1363
1364 /* Copy the prefix into the new pathname as a starting point. */
1365 len = strlcpy(buf, prefix, MAXPATHLEN);
1366 if (len >= MAXPATHLEN) {
1367 *pathbuf = NULL;
1368 free(buf, M_TEMP);
1369 return (EINVAL);
1370 }
1371 sz = MAXPATHLEN - len;
1372 ptr = buf + len;
1373
1374 /* Append the filename to the prefix. */
1375 if (pathseg == UIO_SYSSPACE)
1376 error = copystr(path, ptr, sz, &len);
1377 else
1378 error = copyinstr(path, ptr, sz, &len);
1379
1380 if (error) {
1381 *pathbuf = NULL;
1382 free(buf, M_TEMP);
1383 return (error);
1384 }
1385
1386 /* Only use a prefix with absolute pathnames. */
1387 if (*ptr != '/') {
1388 error = EINVAL;
1389 goto keeporig;
1390 }
1391
1392 if (dirfd != AT_FDCWD) {
1393 /*
1394 * We want the original because the "prefix" is
1395 * included in the already opened dirfd.
1396 */
1397 bcopy(ptr, buf, len);
1398 return (0);
1399 }
1400
1401 /*
1402 * We know that there is a / somewhere in this pathname.
1403 * Search backwards for it, to find the file's parent dir
1404 * to see if it exists in the alternate tree. If it does,
1405 * and we want to create a file (cflag is set). We don't
1406 * need to worry about the root comparison in this case.
1407 */
1408
1409 if (create) {
1410 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1411 *cp = '\0';
1412
1413 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1414 error = namei(&nd);
1415 *cp = '/';
1416 if (error != 0)
1417 goto keeporig;
1418 } else {
1419 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1420
1421 error = namei(&nd);
1422 if (error != 0)
1423 goto keeporig;
1424
1425 /*
1426 * We now compare the vnode of the prefix to the one
1427 * vnode asked. If they resolve to be the same, then we
1428 * ignore the match so that the real root gets used.
1429 * This avoids the problem of traversing "../.." to find the
1430 * root directory and never finding it, because "/" resolves
1431 * to the emulation root directory. This is expensive :-(
1432 */
1433 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1434 td);
1435
1436 /* We shouldn't ever get an error from this namei(). */
1437 error = namei(&ndroot);
1438 if (error == 0) {
1439 if (nd.ni_vp == ndroot.ni_vp)
1440 error = ENOENT;
1441
1442 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1443 vrele(ndroot.ni_vp);
1444 }
1445 }
1446
1447 NDFREE(&nd, NDF_ONLY_PNBUF);
1448 vrele(nd.ni_vp);
1449
1450 keeporig:
1451 /* If there was an error, use the original path name. */
1452 if (error)
1453 bcopy(ptr, buf, len);
1454 return (error);
1455 }
1456