FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_lookup.c
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_ktrace.h"
41 #include "opt_mac.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/vnode.h>
51 #include <sys/mount.h>
52 #include <sys/filedesc.h>
53 #include <sys/proc.h>
54 #include <sys/syscallsubr.h>
55 #ifdef KTRACE
56 #include <sys/ktrace.h>
57 #endif
58
59 #include <vm/uma.h>
60
61 /*
62 * Allocation zone for namei
63 */
64 uma_zone_t namei_zone;
65
66 static void
67 nameiinit(void *dummy __unused)
68 {
69 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
70 UMA_ALIGN_PTR, 0);
71
72 }
73 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL)
74
75 /*
76 * Convert a pathname into a pointer to a locked inode.
77 *
78 * The FOLLOW flag is set when symbolic links are to be followed
79 * when they occur at the end of the name translation process.
80 * Symbolic links are always followed for all other pathname
81 * components other than the last.
82 *
83 * The segflg defines whether the name is to be copied from user
84 * space or kernel space.
85 *
86 * Overall outline of namei:
87 *
88 * copy in name
89 * get starting directory
90 * while (!done && !error) {
91 * call lookup to search path.
92 * if symbolic link, massage name in buffer and continue
93 * }
94 */
95 int
96 namei(ndp)
97 register struct nameidata *ndp;
98 {
99 register struct filedesc *fdp; /* pointer to file descriptor state */
100 register char *cp; /* pointer into pathname argument */
101 register struct vnode *dp; /* the directory we are searching */
102 struct iovec aiov; /* uio for reading symbolic links */
103 struct uio auio;
104 int error, linklen;
105 struct componentname *cnp = &ndp->ni_cnd;
106 struct thread *td = cnp->cn_thread;
107 struct proc *p = td->td_proc;
108
109 GIANT_REQUIRED;
110
111 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
112 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
113 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
114 ("namei: nameiop contaminated with flags"));
115 KASSERT((cnp->cn_flags & OPMASK) == 0,
116 ("namei: flags contaminated with nameiops"));
117 fdp = p->p_fd;
118
119 /*
120 * Get a buffer for the name to be translated, and copy the
121 * name into the buffer.
122 */
123 if ((cnp->cn_flags & HASBUF) == 0)
124 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
125 if (ndp->ni_segflg == UIO_SYSSPACE)
126 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
127 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
128 else
129 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
130 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
131
132 /*
133 * Don't allow empty pathnames.
134 */
135 if (!error && *cnp->cn_pnbuf == '\0')
136 error = ENOENT;
137
138 if (error) {
139 uma_zfree(namei_zone, cnp->cn_pnbuf);
140 #ifdef DIAGNOSTIC
141 cnp->cn_pnbuf = NULL;
142 cnp->cn_nameptr = NULL;
143 #endif
144 ndp->ni_vp = NULL;
145 return (error);
146 }
147 ndp->ni_loopcnt = 0;
148 #ifdef KTRACE
149 if (KTRPOINT(td, KTR_NAMEI)) {
150 KASSERT(cnp->cn_thread == curthread,
151 ("namei not using curthread"));
152 ktrnamei(cnp->cn_pnbuf);
153 }
154 #endif
155
156 /*
157 * Get starting point for the translation.
158 */
159 FILEDESC_LOCK(fdp);
160 ndp->ni_rootdir = fdp->fd_rdir;
161 ndp->ni_topdir = fdp->fd_jdir;
162
163 dp = fdp->fd_cdir;
164 VREF(dp);
165 FILEDESC_UNLOCK(fdp);
166 for (;;) {
167 /*
168 * Check if root directory should replace current directory.
169 * Done at start of translation and after symbolic link.
170 */
171 cnp->cn_nameptr = cnp->cn_pnbuf;
172 if (*(cnp->cn_nameptr) == '/') {
173 vrele(dp);
174 while (*(cnp->cn_nameptr) == '/') {
175 cnp->cn_nameptr++;
176 ndp->ni_pathlen--;
177 }
178 dp = ndp->ni_rootdir;
179 VREF(dp);
180 }
181 ndp->ni_startdir = dp;
182 error = lookup(ndp);
183 if (error) {
184 uma_zfree(namei_zone, cnp->cn_pnbuf);
185 #ifdef DIAGNOSTIC
186 cnp->cn_pnbuf = NULL;
187 cnp->cn_nameptr = NULL;
188 #endif
189 return (error);
190 }
191 /*
192 * Check for symbolic link
193 */
194 if ((cnp->cn_flags & ISSYMLINK) == 0) {
195 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
196 uma_zfree(namei_zone, cnp->cn_pnbuf);
197 #ifdef DIAGNOSTIC
198 cnp->cn_pnbuf = NULL;
199 cnp->cn_nameptr = NULL;
200 #endif
201 } else
202 cnp->cn_flags |= HASBUF;
203
204 if (vn_canvmio(ndp->ni_vp) == TRUE &&
205 (cnp->cn_nameiop != DELETE) &&
206 ((cnp->cn_flags & (NOOBJ|LOCKLEAF)) ==
207 LOCKLEAF))
208 vfs_object_create(ndp->ni_vp, td,
209 ndp->ni_cnd.cn_cred);
210
211 return (0);
212 }
213 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
214 VOP_UNLOCK(ndp->ni_dvp, 0, td);
215 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
216 error = ELOOP;
217 break;
218 }
219 #ifdef MAC
220 if ((cnp->cn_flags & NOMACCHECK) == 0) {
221 error = mac_check_vnode_readlink(td->td_ucred,
222 ndp->ni_vp);
223 if (error)
224 break;
225 }
226 #endif
227 if (ndp->ni_pathlen > 1)
228 cp = uma_zalloc(namei_zone, M_WAITOK);
229 else
230 cp = cnp->cn_pnbuf;
231 aiov.iov_base = cp;
232 aiov.iov_len = MAXPATHLEN;
233 auio.uio_iov = &aiov;
234 auio.uio_iovcnt = 1;
235 auio.uio_offset = 0;
236 auio.uio_rw = UIO_READ;
237 auio.uio_segflg = UIO_SYSSPACE;
238 auio.uio_td = (struct thread *)0;
239 auio.uio_resid = MAXPATHLEN;
240 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
241 if (error) {
242 if (ndp->ni_pathlen > 1)
243 uma_zfree(namei_zone, cp);
244 break;
245 }
246 linklen = MAXPATHLEN - auio.uio_resid;
247 if (linklen == 0) {
248 if (ndp->ni_pathlen > 1)
249 uma_zfree(namei_zone, cp);
250 error = ENOENT;
251 break;
252 }
253 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
254 if (ndp->ni_pathlen > 1)
255 uma_zfree(namei_zone, cp);
256 error = ENAMETOOLONG;
257 break;
258 }
259 if (ndp->ni_pathlen > 1) {
260 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
261 uma_zfree(namei_zone, cnp->cn_pnbuf);
262 cnp->cn_pnbuf = cp;
263 } else
264 cnp->cn_pnbuf[linklen] = '\0';
265 ndp->ni_pathlen += linklen;
266 vput(ndp->ni_vp);
267 dp = ndp->ni_dvp;
268 }
269 uma_zfree(namei_zone, cnp->cn_pnbuf);
270 #ifdef DIAGNOSTIC
271 cnp->cn_pnbuf = NULL;
272 cnp->cn_nameptr = NULL;
273 #endif
274 vrele(ndp->ni_dvp);
275 vput(ndp->ni_vp);
276 ndp->ni_vp = NULL;
277 return (error);
278 }
279
280 /*
281 * Search a pathname.
282 * This is a very central and rather complicated routine.
283 *
284 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
285 * The starting directory is taken from ni_startdir. The pathname is
286 * descended until done, or a symbolic link is encountered. The variable
287 * ni_more is clear if the path is completed; it is set to one if a
288 * symbolic link needing interpretation is encountered.
289 *
290 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
291 * whether the name is to be looked up, created, renamed, or deleted.
292 * When CREATE, RENAME, or DELETE is specified, information usable in
293 * creating, renaming, or deleting a directory entry may be calculated.
294 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
295 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
296 * returned unlocked. Otherwise the parent directory is not returned. If
297 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
298 * the target is returned locked, otherwise it is returned unlocked.
299 * When creating or renaming and LOCKPARENT is specified, the target may not
300 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
301 *
302 * Overall outline of lookup:
303 *
304 * dirloop:
305 * identify next component of name at ndp->ni_ptr
306 * handle degenerate case where name is null string
307 * if .. and crossing mount points and on mounted filesys, find parent
308 * call VOP_LOOKUP routine for next component name
309 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
310 * component vnode returned in ni_vp (if it exists), locked.
311 * if result vnode is mounted on and crossing mount points,
312 * find mounted on vnode
313 * if more components of name, do next level at dirloop
314 * return the answer in ni_vp, locked if LOCKLEAF set
315 * if LOCKPARENT set, return locked parent in ni_dvp
316 * if WANTPARENT set, return unlocked parent in ni_dvp
317 */
318 int
319 lookup(ndp)
320 register struct nameidata *ndp;
321 {
322 register char *cp; /* pointer into pathname argument */
323 register struct vnode *dp = 0; /* the directory we are searching */
324 struct vnode *tdp; /* saved dp */
325 struct mount *mp; /* mount table entry */
326 int docache; /* == 0 do not cache last component */
327 int wantparent; /* 1 => wantparent or lockparent flag */
328 int rdonly; /* lookup read-only flag bit */
329 int trailing_slash;
330 int error = 0;
331 int dpunlocked = 0; /* dp has already been unlocked */
332 struct componentname *cnp = &ndp->ni_cnd;
333 struct thread *td = cnp->cn_thread;
334
335 /*
336 * Setup: break out flag bits into variables.
337 */
338 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
339 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
340 if (cnp->cn_nameiop == DELETE ||
341 (wantparent && cnp->cn_nameiop != CREATE &&
342 cnp->cn_nameiop != LOOKUP))
343 docache = 0;
344 rdonly = cnp->cn_flags & RDONLY;
345 ndp->ni_dvp = NULL;
346 cnp->cn_flags &= ~ISSYMLINK;
347 dp = ndp->ni_startdir;
348 ndp->ni_startdir = NULLVP;
349 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
350
351 dirloop:
352 /*
353 * Search a new directory.
354 *
355 * The last component of the filename is left accessible via
356 * cnp->cn_nameptr for callers that need the name. Callers needing
357 * the name set the SAVENAME flag. When done, they assume
358 * responsibility for freeing the pathname buffer.
359 */
360 cnp->cn_consume = 0;
361 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
362 continue;
363 cnp->cn_namelen = cp - cnp->cn_nameptr;
364 if (cnp->cn_namelen > NAME_MAX) {
365 error = ENAMETOOLONG;
366 goto bad;
367 }
368 #ifdef NAMEI_DIAGNOSTIC
369 { char c = *cp;
370 *cp = '\0';
371 printf("{%s}: ", cnp->cn_nameptr);
372 *cp = c; }
373 #endif
374 ndp->ni_pathlen -= cnp->cn_namelen;
375 ndp->ni_next = cp;
376
377 /*
378 * Replace multiple slashes by a single slash and trailing slashes
379 * by a null. This must be done before VOP_LOOKUP() because some
380 * fs's don't know about trailing slashes. Remember if there were
381 * trailing slashes to handle symlinks, existing non-directories
382 * and non-existing files that won't be directories specially later.
383 */
384 trailing_slash = 0;
385 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
386 cp++;
387 ndp->ni_pathlen--;
388 if (*cp == '\0') {
389 trailing_slash = 1;
390 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
391 }
392 }
393 ndp->ni_next = cp;
394
395 cnp->cn_flags |= MAKEENTRY;
396 if (*cp == '\0' && docache == 0)
397 cnp->cn_flags &= ~MAKEENTRY;
398 if (cnp->cn_namelen == 2 &&
399 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
400 cnp->cn_flags |= ISDOTDOT;
401 else
402 cnp->cn_flags &= ~ISDOTDOT;
403 if (*ndp->ni_next == 0)
404 cnp->cn_flags |= ISLASTCN;
405 else
406 cnp->cn_flags &= ~ISLASTCN;
407
408
409 /*
410 * Check for degenerate name (e.g. / or "")
411 * which is a way of talking about a directory,
412 * e.g. like "/." or ".".
413 */
414 if (cnp->cn_nameptr[0] == '\0') {
415 if (dp->v_type != VDIR) {
416 error = ENOTDIR;
417 goto bad;
418 }
419 if (cnp->cn_nameiop != LOOKUP) {
420 error = EISDIR;
421 goto bad;
422 }
423 if (wantparent) {
424 ndp->ni_dvp = dp;
425 VREF(dp);
426 }
427 ndp->ni_vp = dp;
428 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
429 VOP_UNLOCK(dp, 0, td);
430 /* XXX This should probably move to the top of function. */
431 if (cnp->cn_flags & SAVESTART)
432 panic("lookup: SAVESTART");
433 return (0);
434 }
435
436 /*
437 * Handle "..": four special cases.
438 * 1. Return an error if this is the last component of
439 * the name and the operation is DELETE or RENAME.
440 * 2. If at root directory (e.g. after chroot)
441 * or at absolute root directory
442 * then ignore it so can't get out.
443 * 3. If this vnode is the root of a mounted
444 * filesystem, then replace it with the
445 * vnode which was mounted on so we take the
446 * .. in the other filesystem.
447 * 4. If the vnode is the top directory of
448 * the jail or chroot, don't let them out.
449 */
450 if (cnp->cn_flags & ISDOTDOT) {
451 if ((cnp->cn_flags & ISLASTCN) != 0 &&
452 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
453 error = EINVAL;
454 goto bad;
455 }
456 for (;;) {
457 if (dp == ndp->ni_rootdir ||
458 dp == ndp->ni_topdir ||
459 dp == rootvnode) {
460 ndp->ni_dvp = dp;
461 ndp->ni_vp = dp;
462 VREF(dp);
463 goto nextname;
464 }
465 if ((dp->v_vflag & VV_ROOT) == 0 ||
466 (cnp->cn_flags & NOCROSSMOUNT))
467 break;
468 if (dp->v_mount == NULL) { /* forced unmount */
469 error = EBADF;
470 goto bad;
471 }
472 tdp = dp;
473 dp = dp->v_mount->mnt_vnodecovered;
474 vput(tdp);
475 VREF(dp);
476 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
477 }
478 }
479
480 /*
481 * We now have a segment name to search for, and a directory to search.
482 */
483 unionlookup:
484 #ifdef MAC
485 if ((cnp->cn_flags & NOMACCHECK) == 0) {
486 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
487 if (error)
488 goto bad;
489 }
490 #endif
491 ndp->ni_dvp = dp;
492 ndp->ni_vp = NULL;
493 cnp->cn_flags &= ~PDIRUNLOCK;
494 ASSERT_VOP_LOCKED(dp, "lookup");
495 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
496 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
497 #ifdef NAMEI_DIAGNOSTIC
498 printf("not found\n");
499 #endif
500 if ((error == ENOENT) &&
501 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
502 (dp->v_mount->mnt_flag & MNT_UNION)) {
503 tdp = dp;
504 dp = dp->v_mount->mnt_vnodecovered;
505 if (cnp->cn_flags & PDIRUNLOCK)
506 vrele(tdp);
507 else
508 vput(tdp);
509 VREF(dp);
510 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
511 goto unionlookup;
512 }
513
514 if (error != EJUSTRETURN)
515 goto bad;
516 /*
517 * If creating and at end of pathname, then can consider
518 * allowing file to be created.
519 */
520 if (rdonly) {
521 error = EROFS;
522 goto bad;
523 }
524 if (*cp == '\0' && trailing_slash &&
525 !(cnp->cn_flags & WILLBEDIR)) {
526 error = ENOENT;
527 goto bad;
528 }
529 /*
530 * We return with ni_vp NULL to indicate that the entry
531 * doesn't currently exist, leaving a pointer to the
532 * (possibly locked) directory inode in ndp->ni_dvp.
533 */
534 if (cnp->cn_flags & SAVESTART) {
535 ndp->ni_startdir = ndp->ni_dvp;
536 VREF(ndp->ni_startdir);
537 }
538 return (0);
539 }
540 #ifdef NAMEI_DIAGNOSTIC
541 printf("found\n");
542 #endif
543
544 ASSERT_VOP_LOCKED(ndp->ni_vp, "lookup");
545
546 /*
547 * Take into account any additional components consumed by
548 * the underlying filesystem.
549 */
550 if (cnp->cn_consume > 0) {
551 cnp->cn_nameptr += cnp->cn_consume;
552 ndp->ni_next += cnp->cn_consume;
553 ndp->ni_pathlen -= cnp->cn_consume;
554 cnp->cn_consume = 0;
555 }
556
557 dp = ndp->ni_vp;
558
559 /*
560 * Check to see if the vnode has been mounted on;
561 * if so find the root of the mounted filesystem.
562 */
563 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
564 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
565 if (vfs_busy(mp, 0, 0, td))
566 continue;
567 VOP_UNLOCK(dp, 0, td);
568 error = VFS_ROOT(mp, &tdp, td);
569 vfs_unbusy(mp, td);
570 if (error) {
571 dpunlocked = 1;
572 goto bad2;
573 }
574 vrele(dp);
575 ndp->ni_vp = dp = tdp;
576 }
577
578 /*
579 * Check for symbolic link
580 */
581 if ((dp->v_type == VLNK) &&
582 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
583 *ndp->ni_next == '/')) {
584 cnp->cn_flags |= ISSYMLINK;
585 if (dp->v_mount == NULL) {
586 /* We can't know whether the directory was mounted with
587 * NOSYMFOLLOW, so we can't follow safely. */
588 error = EBADF;
589 goto bad2;
590 }
591 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
592 error = EACCES;
593 goto bad2;
594 }
595 return (0);
596 }
597
598 /*
599 * Check for bogus trailing slashes.
600 */
601 if (trailing_slash && dp->v_type != VDIR) {
602 error = ENOTDIR;
603 goto bad2;
604 }
605
606 nextname:
607 /*
608 * Not a symbolic link. If more pathname,
609 * continue at next component, else return.
610 */
611 if (*ndp->ni_next == '/') {
612 cnp->cn_nameptr = ndp->ni_next;
613 while (*cnp->cn_nameptr == '/') {
614 cnp->cn_nameptr++;
615 ndp->ni_pathlen--;
616 }
617 if (ndp->ni_dvp != ndp->ni_vp)
618 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "lookup");
619 vrele(ndp->ni_dvp);
620 goto dirloop;
621 }
622 /*
623 * Disallow directory write attempts on read-only filesystems.
624 */
625 if (rdonly &&
626 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
627 error = EROFS;
628 goto bad2;
629 }
630 if (cnp->cn_flags & SAVESTART) {
631 ndp->ni_startdir = ndp->ni_dvp;
632 VREF(ndp->ni_startdir);
633 }
634 if (!wantparent)
635 vrele(ndp->ni_dvp);
636
637 if ((cnp->cn_flags & LOCKLEAF) == 0)
638 VOP_UNLOCK(dp, 0, td);
639 return (0);
640
641 bad2:
642 if ((cnp->cn_flags & (LOCKPARENT | PDIRUNLOCK)) == LOCKPARENT &&
643 *ndp->ni_next == '\0')
644 VOP_UNLOCK(ndp->ni_dvp, 0, td);
645 vrele(ndp->ni_dvp);
646 bad:
647 if (dpunlocked)
648 vrele(dp);
649 else
650 vput(dp);
651 ndp->ni_vp = NULL;
652 return (error);
653 }
654
655 /*
656 * relookup - lookup a path name component
657 * Used by lookup to re-aquire things.
658 */
659 int
660 relookup(dvp, vpp, cnp)
661 struct vnode *dvp, **vpp;
662 struct componentname *cnp;
663 {
664 struct thread *td = cnp->cn_thread;
665 struct vnode *dp = 0; /* the directory we are searching */
666 int docache; /* == 0 do not cache last component */
667 int wantparent; /* 1 => wantparent or lockparent flag */
668 int rdonly; /* lookup read-only flag bit */
669 int error = 0;
670 #ifdef NAMEI_DIAGNOSTIC
671 int newhash; /* DEBUG: check name hash */
672 char *cp; /* DEBUG: check name ptr/len */
673 #endif
674
675 /*
676 * Setup: break out flag bits into variables.
677 */
678 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
679 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
680 if (cnp->cn_nameiop == DELETE ||
681 (wantparent && cnp->cn_nameiop != CREATE))
682 docache = 0;
683 rdonly = cnp->cn_flags & RDONLY;
684 cnp->cn_flags &= ~ISSYMLINK;
685 dp = dvp;
686 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
687
688 /* dirloop: */
689 /*
690 * Search a new directory.
691 *
692 * The last component of the filename is left accessible via
693 * cnp->cn_nameptr for callers that need the name. Callers needing
694 * the name set the SAVENAME flag. When done, they assume
695 * responsibility for freeing the pathname buffer.
696 */
697 #ifdef NAMEI_DIAGNOSTIC
698 if (cnp->cn_namelen != cp - cnp->cn_nameptr)
699 panic ("relookup: bad len");
700 if (*cp != 0)
701 panic("relookup: not last component");
702 printf("{%s}: ", cnp->cn_nameptr);
703 #endif
704
705 /*
706 * Check for degenerate name (e.g. / or "")
707 * which is a way of talking about a directory,
708 * e.g. like "/." or ".".
709 */
710 if (cnp->cn_nameptr[0] == '\0') {
711 if (cnp->cn_nameiop != LOOKUP || wantparent) {
712 error = EISDIR;
713 goto bad;
714 }
715 if (dp->v_type != VDIR) {
716 error = ENOTDIR;
717 goto bad;
718 }
719 if (!(cnp->cn_flags & LOCKLEAF))
720 VOP_UNLOCK(dp, 0, td);
721 *vpp = dp;
722 /* XXX This should probably move to the top of function. */
723 if (cnp->cn_flags & SAVESTART)
724 panic("lookup: SAVESTART");
725 return (0);
726 }
727
728 if (cnp->cn_flags & ISDOTDOT)
729 panic ("relookup: lookup on dot-dot");
730
731 /*
732 * We now have a segment name to search for, and a directory to search.
733 */
734 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
735 KASSERT(*vpp == NULL, ("leaf should be empty"));
736 if (error != EJUSTRETURN)
737 goto bad;
738 /*
739 * If creating and at end of pathname, then can consider
740 * allowing file to be created.
741 */
742 if (rdonly) {
743 error = EROFS;
744 goto bad;
745 }
746 /* ASSERT(dvp == ndp->ni_startdir) */
747 if (cnp->cn_flags & SAVESTART)
748 VREF(dvp);
749 /*
750 * We return with ni_vp NULL to indicate that the entry
751 * doesn't currently exist, leaving a pointer to the
752 * (possibly locked) directory inode in ndp->ni_dvp.
753 */
754 return (0);
755 }
756 dp = *vpp;
757
758 /*
759 * Check for symbolic link
760 */
761 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
762 ("relookup: symlink found.\n"));
763
764 /*
765 * Disallow directory write attempts on read-only filesystems.
766 */
767 if (rdonly &&
768 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
769 error = EROFS;
770 goto bad2;
771 }
772 /* ASSERT(dvp == ndp->ni_startdir) */
773 if (cnp->cn_flags & SAVESTART)
774 VREF(dvp);
775
776 if (!wantparent)
777 vrele(dvp);
778
779 if (vn_canvmio(dp) == TRUE &&
780 ((cnp->cn_flags & (NOOBJ|LOCKLEAF)) == LOCKLEAF))
781 vfs_object_create(dp, td, cnp->cn_cred);
782
783 if ((cnp->cn_flags & LOCKLEAF) == 0)
784 VOP_UNLOCK(dp, 0, td);
785 return (0);
786
787 bad2:
788 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
789 VOP_UNLOCK(dvp, 0, td);
790 vrele(dvp);
791 bad:
792 vput(dp);
793 *vpp = NULL;
794 return (error);
795 }
796
797 /*
798 * Determine if there is a suitable alternate filename under the specified
799 * prefix for the specified path. If the create flag is set, then the
800 * alternate prefix will be used so long as the parent directory exists.
801 * This is used by the various compatiblity ABIs so that Linux binaries prefer
802 * files under /compat/linux for example. The chosen path (whether under
803 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
804 * to by pathbuf. The caller is responsible for free'ing the buffer from
805 * the M_TEMP bucket if one is returned.
806 */
807 int
808 kern_alternate_path(struct thread *td, const char *prefix, char *path,
809 enum uio_seg pathseg, char **pathbuf, int create)
810 {
811 struct nameidata nd, ndroot;
812 char *ptr, *buf, *cp;
813 size_t len, sz;
814 int error;
815
816 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
817 *pathbuf = buf;
818
819 /* Copy the prefix into the new pathname as a starting point. */
820 len = strlcpy(buf, prefix, MAXPATHLEN);
821 if (len >= MAXPATHLEN) {
822 *pathbuf = NULL;
823 free(buf, M_TEMP);
824 return (EINVAL);
825 }
826 sz = MAXPATHLEN - len;
827 ptr = buf + len;
828
829 /* Append the filename to the prefix. */
830 if (pathseg == UIO_SYSSPACE)
831 error = copystr(path, ptr, sz, &len);
832 else
833 error = copyinstr(path, ptr, sz, &len);
834
835 if (error) {
836 *pathbuf = NULL;
837 free(buf, M_TEMP);
838 return (error);
839 }
840
841 /* Only use a prefix with absolute pathnames. */
842 if (*ptr != '/') {
843 error = EINVAL;
844 goto keeporig;
845 }
846
847 /* XXX: VFS_LOCK_GIANT? */
848 mtx_lock(&Giant);
849
850 /*
851 * We know that there is a / somewhere in this pathname.
852 * Search backwards for it, to find the file's parent dir
853 * to see if it exists in the alternate tree. If it does,
854 * and we want to create a file (cflag is set). We don't
855 * need to worry about the root comparison in this case.
856 */
857
858 if (create) {
859 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
860 *cp = '\0';
861
862 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
863 error = namei(&nd);
864 *cp = '/';
865 if (error != 0)
866 goto nd_failed;
867 } else {
868 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
869
870 error = namei(&nd);
871 if (error != 0)
872 goto nd_failed;
873
874 /*
875 * We now compare the vnode of the prefix to the one
876 * vnode asked. If they resolve to be the same, then we
877 * ignore the match so that the real root gets used.
878 * This avoids the problem of traversing "../.." to find the
879 * root directory and never finding it, because "/" resolves
880 * to the emulation root directory. This is expensive :-(
881 */
882 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, td);
883
884 /* We shouldn't ever get an error from this namei(). */
885 error = namei(&ndroot);
886 if (error == 0) {
887 if (nd.ni_vp == ndroot.ni_vp)
888 error = ENOENT;
889
890 NDFREE(&ndroot, NDF_ONLY_PNBUF);
891 vrele(ndroot.ni_vp);
892 }
893 }
894
895 NDFREE(&nd, NDF_ONLY_PNBUF);
896 vrele(nd.ni_vp);
897
898 nd_failed:
899 /* XXX: VFS_UNLOCK_GIANT? */
900 mtx_unlock(&Giant);
901
902 keeporig:
903 /* If there was an error, use the original path name. */
904 if (error)
905 bcopy(ptr, buf, len);
906 return (error);
907 }
Cache object: 44af4e6f3186b5a792846ba840f67113
|