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: releng/6.1/sys/kern/vfs_lookup.c 158159 2006-04-30 03:58:12Z kris $");
39
40 #include "opt_ktrace.h"
41 #include "opt_mac.h"
42 #include "opt_vfs.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/mac.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/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60
61 #include <vm/uma.h>
62
63 #define NAMEI_DIAGNOSTIC 1
64 #undef NAMEI_DIAGNOSTIC
65
66 /*
67 * Allocation zone for namei
68 */
69 uma_zone_t namei_zone;
70
71 static void
72 nameiinit(void *dummy __unused)
73 {
74 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
75 UMA_ALIGN_PTR, 0);
76
77 }
78 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL)
79
80 #ifdef LOOKUP_SHARED
81 static int lookup_shared = 1;
82 #else
83 static int lookup_shared = 0;
84 #endif
85 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
86 "Enables/Disables shared locks for path name translation");
87
88 /*
89 * Convert a pathname into a pointer to a locked inode.
90 *
91 * The FOLLOW flag is set when symbolic links are to be followed
92 * when they occur at the end of the name translation process.
93 * Symbolic links are always followed for all other pathname
94 * components other than the last.
95 *
96 * The segflg defines whether the name is to be copied from user
97 * space or kernel space.
98 *
99 * Overall outline of namei:
100 *
101 * copy in name
102 * get starting directory
103 * while (!done && !error) {
104 * call lookup to search path.
105 * if symbolic link, massage name in buffer and continue
106 * }
107 */
108 int
109 namei(ndp)
110 register struct nameidata *ndp;
111 {
112 register struct filedesc *fdp; /* pointer to file descriptor state */
113 register char *cp; /* pointer into pathname argument */
114 register struct vnode *dp; /* the directory we are searching */
115 struct iovec aiov; /* uio for reading symbolic links */
116 struct uio auio;
117 int error, linklen;
118 struct componentname *cnp = &ndp->ni_cnd;
119 struct thread *td = cnp->cn_thread;
120 struct proc *p = td->td_proc;
121 int vfslocked;
122
123 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
124 ("NOT MPSAFE and Giant not held"));
125 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
126 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
127 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
128 ("namei: nameiop contaminated with flags"));
129 KASSERT((cnp->cn_flags & OPMASK) == 0,
130 ("namei: flags contaminated with nameiops"));
131 if (!lookup_shared)
132 cnp->cn_flags &= ~LOCKSHARED;
133 fdp = p->p_fd;
134
135 /*
136 * Get a buffer for the name to be translated, and copy the
137 * name into the buffer.
138 */
139 if ((cnp->cn_flags & HASBUF) == 0)
140 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
141 if (ndp->ni_segflg == UIO_SYSSPACE)
142 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
143 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
144 else
145 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
146 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
147
148 /*
149 * Don't allow empty pathnames.
150 */
151 if (!error && *cnp->cn_pnbuf == '\0')
152 error = ENOENT;
153
154 if (error) {
155 uma_zfree(namei_zone, cnp->cn_pnbuf);
156 #ifdef DIAGNOSTIC
157 cnp->cn_pnbuf = NULL;
158 cnp->cn_nameptr = NULL;
159 #endif
160 ndp->ni_vp = NULL;
161 return (error);
162 }
163 ndp->ni_loopcnt = 0;
164 #ifdef KTRACE
165 if (KTRPOINT(td, KTR_NAMEI)) {
166 KASSERT(cnp->cn_thread == curthread,
167 ("namei not using curthread"));
168 ktrnamei(cnp->cn_pnbuf);
169 }
170 #endif
171
172 /*
173 * Get starting point for the translation.
174 */
175 FILEDESC_LOCK(fdp);
176 ndp->ni_rootdir = fdp->fd_rdir;
177 ndp->ni_topdir = fdp->fd_jdir;
178
179 dp = fdp->fd_cdir;
180 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
181 VREF(dp);
182 FILEDESC_UNLOCK(fdp);
183 for (;;) {
184 /*
185 * Check if root directory should replace current directory.
186 * Done at start of translation and after symbolic link.
187 */
188 cnp->cn_nameptr = cnp->cn_pnbuf;
189 if (*(cnp->cn_nameptr) == '/') {
190 vrele(dp);
191 VFS_UNLOCK_GIANT(vfslocked);
192 while (*(cnp->cn_nameptr) == '/') {
193 cnp->cn_nameptr++;
194 ndp->ni_pathlen--;
195 }
196 dp = ndp->ni_rootdir;
197 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
198 VREF(dp);
199 }
200 if (vfslocked)
201 ndp->ni_cnd.cn_flags |= GIANTHELD;
202 ndp->ni_startdir = dp;
203 error = lookup(ndp);
204 if (error) {
205 uma_zfree(namei_zone, cnp->cn_pnbuf);
206 #ifdef DIAGNOSTIC
207 cnp->cn_pnbuf = NULL;
208 cnp->cn_nameptr = NULL;
209 #endif
210 return (error);
211 }
212 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
213 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
214 /*
215 * Check for symbolic link
216 */
217 if ((cnp->cn_flags & ISSYMLINK) == 0) {
218 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
219 uma_zfree(namei_zone, cnp->cn_pnbuf);
220 #ifdef DIAGNOSTIC
221 cnp->cn_pnbuf = NULL;
222 cnp->cn_nameptr = NULL;
223 #endif
224 } else
225 cnp->cn_flags |= HASBUF;
226
227 if ((cnp->cn_flags & MPSAFE) == 0) {
228 VFS_UNLOCK_GIANT(vfslocked);
229 } else if (vfslocked)
230 ndp->ni_cnd.cn_flags |= GIANTHELD;
231 return (0);
232 }
233 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
234 error = ELOOP;
235 break;
236 }
237 #ifdef MAC
238 if ((cnp->cn_flags & NOMACCHECK) == 0) {
239 error = mac_check_vnode_readlink(td->td_ucred,
240 ndp->ni_vp);
241 if (error)
242 break;
243 }
244 #endif
245 if (ndp->ni_pathlen > 1)
246 cp = uma_zalloc(namei_zone, M_WAITOK);
247 else
248 cp = cnp->cn_pnbuf;
249 aiov.iov_base = cp;
250 aiov.iov_len = MAXPATHLEN;
251 auio.uio_iov = &aiov;
252 auio.uio_iovcnt = 1;
253 auio.uio_offset = 0;
254 auio.uio_rw = UIO_READ;
255 auio.uio_segflg = UIO_SYSSPACE;
256 auio.uio_td = (struct thread *)0;
257 auio.uio_resid = MAXPATHLEN;
258 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
259 if (error) {
260 if (ndp->ni_pathlen > 1)
261 uma_zfree(namei_zone, cp);
262 break;
263 }
264 linklen = MAXPATHLEN - auio.uio_resid;
265 if (linklen == 0) {
266 if (ndp->ni_pathlen > 1)
267 uma_zfree(namei_zone, cp);
268 error = ENOENT;
269 break;
270 }
271 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
272 if (ndp->ni_pathlen > 1)
273 uma_zfree(namei_zone, cp);
274 error = ENAMETOOLONG;
275 break;
276 }
277 if (ndp->ni_pathlen > 1) {
278 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
279 uma_zfree(namei_zone, cnp->cn_pnbuf);
280 cnp->cn_pnbuf = cp;
281 } else
282 cnp->cn_pnbuf[linklen] = '\0';
283 ndp->ni_pathlen += linklen;
284 vput(ndp->ni_vp);
285 dp = ndp->ni_dvp;
286 }
287 uma_zfree(namei_zone, cnp->cn_pnbuf);
288 #ifdef DIAGNOSTIC
289 cnp->cn_pnbuf = NULL;
290 cnp->cn_nameptr = NULL;
291 #endif
292 vput(ndp->ni_vp);
293 ndp->ni_vp = NULL;
294 vrele(ndp->ni_dvp);
295 VFS_UNLOCK_GIANT(vfslocked);
296 return (error);
297 }
298
299 /*
300 * Search a pathname.
301 * This is a very central and rather complicated routine.
302 *
303 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
304 * The starting directory is taken from ni_startdir. The pathname is
305 * descended until done, or a symbolic link is encountered. The variable
306 * ni_more is clear if the path is completed; it is set to one if a
307 * symbolic link needing interpretation is encountered.
308 *
309 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
310 * whether the name is to be looked up, created, renamed, or deleted.
311 * When CREATE, RENAME, or DELETE is specified, information usable in
312 * creating, renaming, or deleting a directory entry may be calculated.
313 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
314 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
315 * returned unlocked. Otherwise the parent directory is not returned. If
316 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
317 * the target is returned locked, otherwise it is returned unlocked.
318 * When creating or renaming and LOCKPARENT is specified, the target may not
319 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
320 *
321 * Overall outline of lookup:
322 *
323 * dirloop:
324 * identify next component of name at ndp->ni_ptr
325 * handle degenerate case where name is null string
326 * if .. and crossing mount points and on mounted filesys, find parent
327 * call VOP_LOOKUP routine for next component name
328 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
329 * component vnode returned in ni_vp (if it exists), locked.
330 * if result vnode is mounted on and crossing mount points,
331 * find mounted on vnode
332 * if more components of name, do next level at dirloop
333 * return the answer in ni_vp, locked if LOCKLEAF set
334 * if LOCKPARENT set, return locked parent in ni_dvp
335 * if WANTPARENT set, return unlocked parent in ni_dvp
336 */
337 int
338 lookup(ndp)
339 register struct nameidata *ndp;
340 {
341 register char *cp; /* pointer into pathname argument */
342 register struct vnode *dp = 0; /* the directory we are searching */
343 struct vnode *tdp; /* saved dp */
344 struct mount *mp; /* mount table entry */
345 int docache; /* == 0 do not cache last component */
346 int wantparent; /* 1 => wantparent or lockparent flag */
347 int rdonly; /* lookup read-only flag bit */
348 int trailing_slash;
349 int error = 0;
350 int dpunlocked = 0; /* dp has already been unlocked */
351 struct componentname *cnp = &ndp->ni_cnd;
352 struct thread *td = cnp->cn_thread;
353 int vfslocked; /* VFS Giant state for child */
354 int dvfslocked; /* VFS Giant state for parent */
355 int tvfslocked;
356
357 /*
358 * Setup: break out flag bits into variables.
359 */
360 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
361 vfslocked = 0;
362 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
363 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
364 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
365 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
366 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
367 if (cnp->cn_nameiop == DELETE ||
368 (wantparent && cnp->cn_nameiop != CREATE &&
369 cnp->cn_nameiop != LOOKUP))
370 docache = 0;
371 rdonly = cnp->cn_flags & RDONLY;
372 cnp->cn_flags &= ~ISSYMLINK;
373 ndp->ni_dvp = NULL;
374 /*
375 * We use shared locks until we hit the parent of the last cn then
376 * we adjust based on the requesting flags.
377 */
378 if (lookup_shared)
379 cnp->cn_lkflags = LK_SHARED;
380 else
381 cnp->cn_lkflags = LK_EXCLUSIVE;
382 dp = ndp->ni_startdir;
383 ndp->ni_startdir = NULLVP;
384 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
385
386 dirloop:
387 /*
388 * Search a new directory.
389 *
390 * The last component of the filename is left accessible via
391 * cnp->cn_nameptr for callers that need the name. Callers needing
392 * the name set the SAVENAME flag. When done, they assume
393 * responsibility for freeing the pathname buffer.
394 */
395 cnp->cn_consume = 0;
396 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
397 continue;
398 cnp->cn_namelen = cp - cnp->cn_nameptr;
399 if (cnp->cn_namelen > NAME_MAX) {
400 error = ENAMETOOLONG;
401 goto bad;
402 }
403 #ifdef NAMEI_DIAGNOSTIC
404 { char c = *cp;
405 *cp = '\0';
406 printf("{%s}: ", cnp->cn_nameptr);
407 *cp = c; }
408 #endif
409 ndp->ni_pathlen -= cnp->cn_namelen;
410 ndp->ni_next = cp;
411
412 /*
413 * Replace multiple slashes by a single slash and trailing slashes
414 * by a null. This must be done before VOP_LOOKUP() because some
415 * fs's don't know about trailing slashes. Remember if there were
416 * trailing slashes to handle symlinks, existing non-directories
417 * and non-existing files that won't be directories specially later.
418 */
419 trailing_slash = 0;
420 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
421 cp++;
422 ndp->ni_pathlen--;
423 if (*cp == '\0') {
424 trailing_slash = 1;
425 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
426 }
427 }
428 ndp->ni_next = cp;
429
430 cnp->cn_flags |= MAKEENTRY;
431 if (*cp == '\0' && docache == 0)
432 cnp->cn_flags &= ~MAKEENTRY;
433 if (cnp->cn_namelen == 2 &&
434 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
435 cnp->cn_flags |= ISDOTDOT;
436 else
437 cnp->cn_flags &= ~ISDOTDOT;
438 if (*ndp->ni_next == 0)
439 cnp->cn_flags |= ISLASTCN;
440 else
441 cnp->cn_flags &= ~ISLASTCN;
442
443
444 /*
445 * Check for degenerate name (e.g. / or "")
446 * which is a way of talking about a directory,
447 * e.g. like "/." or ".".
448 */
449 if (cnp->cn_nameptr[0] == '\0') {
450 if (dp->v_type != VDIR) {
451 error = ENOTDIR;
452 goto bad;
453 }
454 if (cnp->cn_nameiop != LOOKUP) {
455 error = EISDIR;
456 goto bad;
457 }
458 if (wantparent) {
459 ndp->ni_dvp = dp;
460 VREF(dp);
461 }
462 ndp->ni_vp = dp;
463 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
464 VOP_UNLOCK(dp, 0, td);
465 /* XXX This should probably move to the top of function. */
466 if (cnp->cn_flags & SAVESTART)
467 panic("lookup: SAVESTART");
468 goto success;
469 }
470
471 /*
472 * Handle "..": four special cases.
473 * 1. Return an error if this is the last component of
474 * the name and the operation is DELETE or RENAME.
475 * 2. If at root directory (e.g. after chroot)
476 * or at absolute root directory
477 * then ignore it so can't get out.
478 * 3. If this vnode is the root of a mounted
479 * filesystem, then replace it with the
480 * vnode which was mounted on so we take the
481 * .. in the other filesystem.
482 * 4. If the vnode is the top directory of
483 * the jail or chroot, don't let them out.
484 */
485 if (cnp->cn_flags & ISDOTDOT) {
486 if ((cnp->cn_flags & ISLASTCN) != 0 &&
487 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
488 error = EINVAL;
489 goto bad;
490 }
491 for (;;) {
492 if (dp == ndp->ni_rootdir ||
493 dp == ndp->ni_topdir ||
494 dp == rootvnode) {
495 ndp->ni_dvp = dp;
496 ndp->ni_vp = dp;
497 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
498 VREF(dp);
499 goto nextname;
500 }
501 if ((dp->v_vflag & VV_ROOT) == 0 ||
502 (cnp->cn_flags & NOCROSSMOUNT))
503 break;
504 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
505 error = EBADF;
506 goto bad;
507 }
508 tdp = dp;
509 dp = dp->v_mount->mnt_vnodecovered;
510 tvfslocked = dvfslocked;
511 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
512 VREF(dp);
513 vput(tdp);
514 VFS_UNLOCK_GIANT(tvfslocked);
515 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
516 }
517 }
518
519 /*
520 * We now have a segment name to search for, and a directory to search.
521 */
522 unionlookup:
523 #ifdef MAC
524 if ((cnp->cn_flags & NOMACCHECK) == 0) {
525 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
526 if (error)
527 goto bad;
528 }
529 #endif
530 ndp->ni_dvp = dp;
531 ndp->ni_vp = NULL;
532 ASSERT_VOP_LOCKED(dp, "lookup");
533 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
534 /*
535 * If we have a shared lock we may need to upgrade the lock for the
536 * last operation.
537 */
538 if (VOP_ISLOCKED(dp, td) == LK_SHARED &&
539 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
540 vn_lock(dp, LK_UPGRADE|LK_RETRY, td);
541 /*
542 * If we're looking up the last component and we need an exclusive
543 * lock, adjust our lkflags.
544 */
545 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
546 (ISLASTCN|LOCKLEAF))
547 cnp->cn_lkflags = LK_EXCLUSIVE;
548 #ifdef NAMEI_DIAGNOSTIC
549 vprint("lookup in", dp);
550 #endif
551 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
552 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
553 #ifdef NAMEI_DIAGNOSTIC
554 printf("not found\n");
555 #endif
556 if ((error == ENOENT) &&
557 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
558 (dp->v_mount->mnt_flag & MNT_UNION)) {
559 tdp = dp;
560 dp = dp->v_mount->mnt_vnodecovered;
561 tvfslocked = dvfslocked;
562 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
563 VREF(dp);
564 vput(tdp);
565 VFS_UNLOCK_GIANT(tvfslocked);
566 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
567 goto unionlookup;
568 }
569
570 if (error != EJUSTRETURN)
571 goto bad;
572 /*
573 * If creating and at end of pathname, then can consider
574 * allowing file to be created.
575 */
576 if (rdonly) {
577 error = EROFS;
578 goto bad;
579 }
580 if (*cp == '\0' && trailing_slash &&
581 !(cnp->cn_flags & WILLBEDIR)) {
582 error = ENOENT;
583 goto bad;
584 }
585 if ((cnp->cn_flags & LOCKPARENT) == 0)
586 VOP_UNLOCK(dp, 0, td);
587 /*
588 * This is a temporary assert to make sure I know what the
589 * behavior here was.
590 */
591 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
592 ("lookup: Unhandled case."));
593 /*
594 * We return with ni_vp NULL to indicate that the entry
595 * doesn't currently exist, leaving a pointer to the
596 * (possibly locked) directory inode in ndp->ni_dvp.
597 */
598 if (cnp->cn_flags & SAVESTART) {
599 ndp->ni_startdir = ndp->ni_dvp;
600 VREF(ndp->ni_startdir);
601 }
602 goto success;
603 }
604 #ifdef NAMEI_DIAGNOSTIC
605 printf("found\n");
606 #endif
607 /*
608 * Take into account any additional components consumed by
609 * the underlying filesystem.
610 */
611 if (cnp->cn_consume > 0) {
612 cnp->cn_nameptr += cnp->cn_consume;
613 ndp->ni_next += cnp->cn_consume;
614 ndp->ni_pathlen -= cnp->cn_consume;
615 cnp->cn_consume = 0;
616 }
617
618 dp = ndp->ni_vp;
619 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
620
621 /*
622 * Check to see if the vnode has been mounted on;
623 * if so find the root of the mounted filesystem.
624 */
625 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
626 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
627 if (vfs_busy(mp, 0, 0, td))
628 continue;
629 vput(dp);
630 VFS_UNLOCK_GIANT(vfslocked);
631 vfslocked = VFS_LOCK_GIANT(mp);
632 if (dp != ndp->ni_dvp)
633 VOP_UNLOCK(ndp->ni_dvp, 0, td);
634 error = VFS_ROOT(mp, cnp->cn_lkflags, &tdp, td);
635 vfs_unbusy(mp, td);
636 vn_lock(ndp->ni_dvp, cnp->cn_lkflags | LK_RETRY, td);
637 if (error) {
638 dpunlocked = 1;
639 goto bad2;
640 }
641 ndp->ni_vp = dp = tdp;
642 }
643
644 /*
645 * Check for symbolic link
646 */
647 if ((dp->v_type == VLNK) &&
648 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
649 *ndp->ni_next == '/')) {
650 cnp->cn_flags |= ISSYMLINK;
651 if (dp->v_iflag & VI_DOOMED) {
652 /* We can't know whether the directory was mounted with
653 * NOSYMFOLLOW, so we can't follow safely. */
654 error = EBADF;
655 goto bad2;
656 }
657 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
658 error = EACCES;
659 goto bad2;
660 }
661 /*
662 * Symlink code always expects an unlocked dvp.
663 */
664 if (ndp->ni_dvp != ndp->ni_vp)
665 VOP_UNLOCK(ndp->ni_dvp, 0, td);
666 goto success;
667 }
668
669 /*
670 * Check for bogus trailing slashes.
671 */
672 if (trailing_slash && dp->v_type != VDIR) {
673 error = ENOTDIR;
674 goto bad2;
675 }
676
677 nextname:
678 /*
679 * Not a symbolic link. If more pathname,
680 * continue at next component, else return.
681 */
682 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
683 ("lookup: invalid path state."));
684 if (*ndp->ni_next == '/') {
685 cnp->cn_nameptr = ndp->ni_next;
686 while (*cnp->cn_nameptr == '/') {
687 cnp->cn_nameptr++;
688 ndp->ni_pathlen--;
689 }
690 if (ndp->ni_dvp != dp)
691 vput(ndp->ni_dvp);
692 else
693 vrele(ndp->ni_dvp);
694 VFS_UNLOCK_GIANT(dvfslocked);
695 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
696 vfslocked = 0;
697 goto dirloop;
698 }
699 /*
700 * Disallow directory write attempts on read-only filesystems.
701 */
702 if (rdonly &&
703 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
704 error = EROFS;
705 goto bad2;
706 }
707 if (cnp->cn_flags & SAVESTART) {
708 ndp->ni_startdir = ndp->ni_dvp;
709 VREF(ndp->ni_startdir);
710 }
711 if (!wantparent) {
712 if (ndp->ni_dvp != dp)
713 vput(ndp->ni_dvp);
714 else
715 vrele(ndp->ni_dvp);
716 VFS_UNLOCK_GIANT(dvfslocked);
717 dvfslocked = 0;
718 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
719 VOP_UNLOCK(ndp->ni_dvp, 0, td);
720
721 if ((cnp->cn_flags & LOCKLEAF) == 0)
722 VOP_UNLOCK(dp, 0, td);
723 success:
724 if (vfslocked && dvfslocked)
725 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
726 if (vfslocked || dvfslocked)
727 ndp->ni_cnd.cn_flags |= GIANTHELD;
728 return (0);
729
730 bad2:
731 if (dp != ndp->ni_dvp)
732 vput(ndp->ni_dvp);
733 else
734 vrele(ndp->ni_dvp);
735 bad:
736 if (!dpunlocked)
737 vput(dp);
738 VFS_UNLOCK_GIANT(vfslocked);
739 VFS_UNLOCK_GIANT(dvfslocked);
740 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
741 ndp->ni_vp = NULL;
742 return (error);
743 }
744
745 /*
746 * relookup - lookup a path name component
747 * Used by lookup to re-aquire things.
748 */
749 int
750 relookup(dvp, vpp, cnp)
751 struct vnode *dvp, **vpp;
752 struct componentname *cnp;
753 {
754 struct thread *td = cnp->cn_thread;
755 struct vnode *dp = 0; /* the directory we are searching */
756 int wantparent; /* 1 => wantparent or lockparent flag */
757 int rdonly; /* lookup read-only flag bit */
758 int error = 0;
759
760 KASSERT(cnp->cn_flags & ISLASTCN,
761 ("relookup: Not given last component."));
762 /*
763 * Setup: break out flag bits into variables.
764 */
765 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
766 KASSERT(wantparent, ("relookup: parent not wanted."));
767 rdonly = cnp->cn_flags & RDONLY;
768 cnp->cn_flags &= ~ISSYMLINK;
769 dp = dvp;
770 cnp->cn_lkflags = LK_EXCLUSIVE;
771 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
772
773 /*
774 * Search a new directory.
775 *
776 * The last component of the filename is left accessible via
777 * cnp->cn_nameptr for callers that need the name. Callers needing
778 * the name set the SAVENAME flag. When done, they assume
779 * responsibility for freeing the pathname buffer.
780 */
781 #ifdef NAMEI_DIAGNOSTIC
782 printf("{%s}: ", cnp->cn_nameptr);
783 #endif
784
785 /*
786 * Check for degenerate name (e.g. / or "")
787 * which is a way of talking about a directory,
788 * e.g. like "/." or ".".
789 */
790 if (cnp->cn_nameptr[0] == '\0') {
791 if (cnp->cn_nameiop != LOOKUP || wantparent) {
792 error = EISDIR;
793 goto bad;
794 }
795 if (dp->v_type != VDIR) {
796 error = ENOTDIR;
797 goto bad;
798 }
799 if (!(cnp->cn_flags & LOCKLEAF))
800 VOP_UNLOCK(dp, 0, td);
801 *vpp = dp;
802 /* XXX This should probably move to the top of function. */
803 if (cnp->cn_flags & SAVESTART)
804 panic("lookup: SAVESTART");
805 return (0);
806 }
807
808 if (cnp->cn_flags & ISDOTDOT)
809 panic ("relookup: lookup on dot-dot");
810
811 /*
812 * We now have a segment name to search for, and a directory to search.
813 */
814 #ifdef NAMEI_DIAGNOSTIC
815 vprint("search in:", dp);
816 #endif
817 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
818 KASSERT(*vpp == NULL, ("leaf should be empty"));
819 if (error != EJUSTRETURN)
820 goto bad;
821 /*
822 * If creating and at end of pathname, then can consider
823 * allowing file to be created.
824 */
825 if (rdonly) {
826 error = EROFS;
827 goto bad;
828 }
829 /* ASSERT(dvp == ndp->ni_startdir) */
830 if (cnp->cn_flags & SAVESTART)
831 VREF(dvp);
832 if ((cnp->cn_flags & LOCKPARENT) == 0)
833 VOP_UNLOCK(dp, 0, td);
834 /*
835 * This is a temporary assert to make sure I know what the
836 * behavior here was.
837 */
838 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
839 ("relookup: Unhandled case."));
840 /*
841 * We return with ni_vp NULL to indicate that the entry
842 * doesn't currently exist, leaving a pointer to the
843 * (possibly locked) directory inode in ndp->ni_dvp.
844 */
845 return (0);
846 }
847 dp = *vpp;
848
849 /*
850 * Disallow directory write attempts on read-only filesystems.
851 */
852 if (rdonly &&
853 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
854 if (dvp == dp)
855 vrele(dvp);
856 else
857 vput(dvp);
858 error = EROFS;
859 goto bad;
860 }
861 /*
862 * Set the parent lock/ref state to the requested state.
863 */
864 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
865 if (wantparent)
866 VOP_UNLOCK(dvp, 0, td);
867 else
868 vput(dvp);
869 } else if (!wantparent)
870 vrele(dvp);
871 /*
872 * Check for symbolic link
873 */
874 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
875 ("relookup: symlink found.\n"));
876
877 /* ASSERT(dvp == ndp->ni_startdir) */
878 if (cnp->cn_flags & SAVESTART)
879 VREF(dvp);
880
881 if ((cnp->cn_flags & LOCKLEAF) == 0)
882 VOP_UNLOCK(dp, 0, td);
883 return (0);
884 bad:
885 vput(dp);
886 *vpp = NULL;
887 return (error);
888 }
889
890 /*
891 * Free data allocated by namei(); see namei(9) for details.
892 */
893 void
894 NDFREE(ndp, flags)
895 struct nameidata *ndp;
896 const u_int flags;
897 {
898 int unlock_dvp;
899 int unlock_vp;
900
901 unlock_dvp = 0;
902 unlock_vp = 0;
903
904 if (!(flags & NDF_NO_FREE_PNBUF) &&
905 (ndp->ni_cnd.cn_flags & HASBUF)) {
906 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
907 ndp->ni_cnd.cn_flags &= ~HASBUF;
908 }
909 if (!(flags & NDF_NO_VP_UNLOCK) &&
910 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
911 unlock_vp = 1;
912 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
913 if (unlock_vp) {
914 vput(ndp->ni_vp);
915 unlock_vp = 0;
916 } else
917 vrele(ndp->ni_vp);
918 ndp->ni_vp = NULL;
919 }
920 if (unlock_vp)
921 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
922 if (!(flags & NDF_NO_DVP_UNLOCK) &&
923 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
924 ndp->ni_dvp != ndp->ni_vp)
925 unlock_dvp = 1;
926 if (!(flags & NDF_NO_DVP_RELE) &&
927 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
928 if (unlock_dvp) {
929 vput(ndp->ni_dvp);
930 unlock_dvp = 0;
931 } else
932 vrele(ndp->ni_dvp);
933 ndp->ni_dvp = NULL;
934 }
935 if (unlock_dvp)
936 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
937 if (!(flags & NDF_NO_STARTDIR_RELE) &&
938 (ndp->ni_cnd.cn_flags & SAVESTART)) {
939 vrele(ndp->ni_startdir);
940 ndp->ni_startdir = NULL;
941 }
942 }
943
944 /*
945 * Determine if there is a suitable alternate filename under the specified
946 * prefix for the specified path. If the create flag is set, then the
947 * alternate prefix will be used so long as the parent directory exists.
948 * This is used by the various compatiblity ABIs so that Linux binaries prefer
949 * files under /compat/linux for example. The chosen path (whether under
950 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
951 * to by pathbuf. The caller is responsible for free'ing the buffer from
952 * the M_TEMP bucket if one is returned.
953 */
954 int
955 kern_alternate_path(struct thread *td, const char *prefix, char *path,
956 enum uio_seg pathseg, char **pathbuf, int create)
957 {
958 struct nameidata nd, ndroot;
959 char *ptr, *buf, *cp;
960 size_t len, sz;
961 int error;
962
963 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
964 *pathbuf = buf;
965
966 /* Copy the prefix into the new pathname as a starting point. */
967 len = strlcpy(buf, prefix, MAXPATHLEN);
968 if (len >= MAXPATHLEN) {
969 *pathbuf = NULL;
970 free(buf, M_TEMP);
971 return (EINVAL);
972 }
973 sz = MAXPATHLEN - len;
974 ptr = buf + len;
975
976 /* Append the filename to the prefix. */
977 if (pathseg == UIO_SYSSPACE)
978 error = copystr(path, ptr, sz, &len);
979 else
980 error = copyinstr(path, ptr, sz, &len);
981
982 if (error) {
983 *pathbuf = NULL;
984 free(buf, M_TEMP);
985 return (error);
986 }
987
988 /* Only use a prefix with absolute pathnames. */
989 if (*ptr != '/') {
990 error = EINVAL;
991 goto keeporig;
992 }
993
994 /*
995 * We know that there is a / somewhere in this pathname.
996 * Search backwards for it, to find the file's parent dir
997 * to see if it exists in the alternate tree. If it does,
998 * and we want to create a file (cflag is set). We don't
999 * need to worry about the root comparison in this case.
1000 */
1001
1002 if (create) {
1003 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1004 *cp = '\0';
1005
1006 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1007 error = namei(&nd);
1008 *cp = '/';
1009 if (error != 0)
1010 goto keeporig;
1011 } else {
1012 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1013
1014 error = namei(&nd);
1015 if (error != 0)
1016 goto keeporig;
1017
1018 /*
1019 * We now compare the vnode of the prefix to the one
1020 * vnode asked. If they resolve to be the same, then we
1021 * ignore the match so that the real root gets used.
1022 * This avoids the problem of traversing "../.." to find the
1023 * root directory and never finding it, because "/" resolves
1024 * to the emulation root directory. This is expensive :-(
1025 */
1026 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1027 td);
1028
1029 /* We shouldn't ever get an error from this namei(). */
1030 error = namei(&ndroot);
1031 if (error == 0) {
1032 if (nd.ni_vp == ndroot.ni_vp)
1033 error = ENOENT;
1034
1035 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1036 vrele(ndroot.ni_vp);
1037 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1038 }
1039 }
1040
1041 NDFREE(&nd, NDF_ONLY_PNBUF);
1042 vrele(nd.ni_vp);
1043 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1044
1045 keeporig:
1046 /* If there was an error, use the original path name. */
1047 if (error)
1048 bcopy(ptr, buf, len);
1049 return (error);
1050 }
Cache object: 9dc6193620aeda3dcaba538632a3e0b0
|