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/6.0/sys/kern/vfs_lookup.c 150725 2005-09-29 18:53:10Z jhb $");
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;
354 int tvfslocked;
355
356 /*
357 * Setup: break out flag bits into variables.
358 */
359 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
360 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
361 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
362 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
363 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
364 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
365 if (cnp->cn_nameiop == DELETE ||
366 (wantparent && cnp->cn_nameiop != CREATE &&
367 cnp->cn_nameiop != LOOKUP))
368 docache = 0;
369 rdonly = cnp->cn_flags & RDONLY;
370 cnp->cn_flags &= ~ISSYMLINK;
371 ndp->ni_dvp = NULL;
372 /*
373 * We use shared locks until we hit the parent of the last cn then
374 * we adjust based on the requesting flags.
375 */
376 if (lookup_shared)
377 cnp->cn_lkflags = LK_SHARED;
378 else
379 cnp->cn_lkflags = LK_EXCLUSIVE;
380 dp = ndp->ni_startdir;
381 ndp->ni_startdir = NULLVP;
382 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
383
384 dirloop:
385 /*
386 * Search a new directory.
387 *
388 * The last component of the filename is left accessible via
389 * cnp->cn_nameptr for callers that need the name. Callers needing
390 * the name set the SAVENAME flag. When done, they assume
391 * responsibility for freeing the pathname buffer.
392 */
393 cnp->cn_consume = 0;
394 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
395 continue;
396 cnp->cn_namelen = cp - cnp->cn_nameptr;
397 if (cnp->cn_namelen > NAME_MAX) {
398 error = ENAMETOOLONG;
399 goto bad;
400 }
401 #ifdef NAMEI_DIAGNOSTIC
402 { char c = *cp;
403 *cp = '\0';
404 printf("{%s}: ", cnp->cn_nameptr);
405 *cp = c; }
406 #endif
407 ndp->ni_pathlen -= cnp->cn_namelen;
408 ndp->ni_next = cp;
409
410 /*
411 * Replace multiple slashes by a single slash and trailing slashes
412 * by a null. This must be done before VOP_LOOKUP() because some
413 * fs's don't know about trailing slashes. Remember if there were
414 * trailing slashes to handle symlinks, existing non-directories
415 * and non-existing files that won't be directories specially later.
416 */
417 trailing_slash = 0;
418 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
419 cp++;
420 ndp->ni_pathlen--;
421 if (*cp == '\0') {
422 trailing_slash = 1;
423 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
424 }
425 }
426 ndp->ni_next = cp;
427
428 cnp->cn_flags |= MAKEENTRY;
429 if (*cp == '\0' && docache == 0)
430 cnp->cn_flags &= ~MAKEENTRY;
431 if (cnp->cn_namelen == 2 &&
432 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
433 cnp->cn_flags |= ISDOTDOT;
434 else
435 cnp->cn_flags &= ~ISDOTDOT;
436 if (*ndp->ni_next == 0)
437 cnp->cn_flags |= ISLASTCN;
438 else
439 cnp->cn_flags &= ~ISLASTCN;
440
441
442 /*
443 * Check for degenerate name (e.g. / or "")
444 * which is a way of talking about a directory,
445 * e.g. like "/." or ".".
446 */
447 if (cnp->cn_nameptr[0] == '\0') {
448 if (dp->v_type != VDIR) {
449 error = ENOTDIR;
450 goto bad;
451 }
452 if (cnp->cn_nameiop != LOOKUP) {
453 error = EISDIR;
454 goto bad;
455 }
456 if (wantparent) {
457 ndp->ni_dvp = dp;
458 VREF(dp);
459 }
460 ndp->ni_vp = dp;
461 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
462 VOP_UNLOCK(dp, 0, td);
463 /* XXX This should probably move to the top of function. */
464 if (cnp->cn_flags & SAVESTART)
465 panic("lookup: SAVESTART");
466 goto success;
467 }
468
469 /*
470 * Handle "..": two special cases.
471 * 1. If at root directory (e.g. after chroot)
472 * or at absolute root directory
473 * then ignore it so can't get out.
474 * 2. If this vnode is the root of a mounted
475 * filesystem, then replace it with the
476 * vnode which was mounted on so we take the
477 * .. in the other filesystem.
478 * 3. If the vnode is the top directory of
479 * the jail or chroot, don't let them out.
480 */
481 if (cnp->cn_flags & ISDOTDOT) {
482 for (;;) {
483 if (dp == ndp->ni_rootdir ||
484 dp == ndp->ni_topdir ||
485 dp == rootvnode) {
486 ndp->ni_dvp = dp;
487 ndp->ni_vp = dp;
488 VREF(dp);
489 goto nextname;
490 }
491 if ((dp->v_vflag & VV_ROOT) == 0 ||
492 (cnp->cn_flags & NOCROSSMOUNT))
493 break;
494 if (dp->v_mount == NULL) { /* forced unmount */
495 error = EBADF;
496 goto bad;
497 }
498 tdp = dp;
499 dp = dp->v_mount->mnt_vnodecovered;
500 tvfslocked = vfslocked;
501 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
502 VREF(dp);
503 vput(tdp);
504 VFS_UNLOCK_GIANT(tvfslocked);
505 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
506 }
507 }
508
509 /*
510 * We now have a segment name to search for, and a directory to search.
511 */
512 unionlookup:
513 #ifdef MAC
514 if ((cnp->cn_flags & NOMACCHECK) == 0) {
515 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
516 if (error)
517 goto bad;
518 }
519 #endif
520 ndp->ni_dvp = dp;
521 ndp->ni_vp = NULL;
522 ASSERT_VOP_LOCKED(dp, "lookup");
523 /*
524 * If we have a shared lock we may need to upgrade the lock for the
525 * last operation.
526 */
527 if (VOP_ISLOCKED(dp, td) == LK_SHARED &&
528 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
529 vn_lock(dp, LK_UPGRADE|LK_RETRY, td);
530 /*
531 * If we're looking up the last component and we need an exclusive
532 * lock, adjust our lkflags.
533 */
534 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
535 (ISLASTCN|LOCKLEAF))
536 cnp->cn_lkflags = LK_EXCLUSIVE;
537 #ifdef NAMEI_DIAGNOSTIC
538 vprint("lookup in", dp);
539 #endif
540 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
541 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
542 #ifdef NAMEI_DIAGNOSTIC
543 printf("not found\n");
544 #endif
545 if ((error == ENOENT) &&
546 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
547 (dp->v_mount->mnt_flag & MNT_UNION)) {
548 tdp = dp;
549 dp = dp->v_mount->mnt_vnodecovered;
550 tvfslocked = vfslocked;
551 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
552 VREF(dp);
553 vput(tdp);
554 VFS_UNLOCK_GIANT(tvfslocked);
555 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
556 goto unionlookup;
557 }
558
559 if (error != EJUSTRETURN)
560 goto bad;
561 /*
562 * If creating and at end of pathname, then can consider
563 * allowing file to be created.
564 */
565 if (rdonly) {
566 error = EROFS;
567 goto bad;
568 }
569 if (*cp == '\0' && trailing_slash &&
570 !(cnp->cn_flags & WILLBEDIR)) {
571 error = ENOENT;
572 goto bad;
573 }
574 if ((cnp->cn_flags & LOCKPARENT) == 0)
575 VOP_UNLOCK(dp, 0, td);
576 /*
577 * This is a temporary assert to make sure I know what the
578 * behavior here was.
579 */
580 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
581 ("lookup: Unhandled case."));
582 /*
583 * We return with ni_vp NULL to indicate that the entry
584 * doesn't currently exist, leaving a pointer to the
585 * (possibly locked) directory inode in ndp->ni_dvp.
586 */
587 if (cnp->cn_flags & SAVESTART) {
588 ndp->ni_startdir = ndp->ni_dvp;
589 VREF(ndp->ni_startdir);
590 }
591 goto success;
592 }
593 #ifdef NAMEI_DIAGNOSTIC
594 printf("found\n");
595 #endif
596 /*
597 * Take into account any additional components consumed by
598 * the underlying filesystem.
599 */
600 if (cnp->cn_consume > 0) {
601 cnp->cn_nameptr += cnp->cn_consume;
602 ndp->ni_next += cnp->cn_consume;
603 ndp->ni_pathlen -= cnp->cn_consume;
604 cnp->cn_consume = 0;
605 }
606
607 dp = ndp->ni_vp;
608
609 /*
610 * Check to see if the vnode has been mounted on;
611 * if so find the root of the mounted filesystem.
612 */
613 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
614 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
615 KASSERT(dp != ndp->ni_dvp, ("XXX"));
616 if (vfs_busy(mp, 0, 0, td))
617 continue;
618 vput(dp);
619 tvfslocked = VFS_LOCK_GIANT(mp);
620 VFS_UNLOCK_GIANT(vfslocked);
621 vfslocked = tvfslocked;
622 VOP_UNLOCK(ndp->ni_dvp, 0, td);
623 error = VFS_ROOT(mp, cnp->cn_lkflags, &tdp, td);
624 VOP_LOCK(ndp->ni_dvp, cnp->cn_lkflags | LK_RETRY, td);
625 vfs_unbusy(mp, td);
626 if (error) {
627 dpunlocked = 1;
628 goto bad2;
629 }
630 ndp->ni_vp = dp = tdp;
631 }
632
633 /*
634 * Check for symbolic link
635 */
636 if ((dp->v_type == VLNK) &&
637 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
638 *ndp->ni_next == '/')) {
639 cnp->cn_flags |= ISSYMLINK;
640 if (dp->v_mount == NULL) {
641 /* We can't know whether the directory was mounted with
642 * NOSYMFOLLOW, so we can't follow safely. */
643 error = EBADF;
644 goto bad2;
645 }
646 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
647 error = EACCES;
648 goto bad2;
649 }
650 /*
651 * Symlink code always expects an unlocked dvp.
652 */
653 if (ndp->ni_dvp != ndp->ni_vp)
654 VOP_UNLOCK(ndp->ni_dvp, 0, td);
655 goto success;
656 }
657
658 /*
659 * Check for bogus trailing slashes.
660 */
661 if (trailing_slash && dp->v_type != VDIR) {
662 error = ENOTDIR;
663 goto bad2;
664 }
665
666 nextname:
667 /*
668 * Not a symbolic link. If more pathname,
669 * continue at next component, else return.
670 */
671 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
672 ("lookup: invalid path state."));
673 if (*ndp->ni_next == '/') {
674 cnp->cn_nameptr = ndp->ni_next;
675 while (*cnp->cn_nameptr == '/') {
676 cnp->cn_nameptr++;
677 ndp->ni_pathlen--;
678 }
679 if (ndp->ni_dvp != dp)
680 vput(ndp->ni_dvp);
681 else
682 vrele(ndp->ni_dvp);
683 goto dirloop;
684 }
685 /*
686 * Disallow directory write attempts on read-only filesystems.
687 */
688 if (rdonly &&
689 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
690 error = EROFS;
691 goto bad2;
692 }
693 if (cnp->cn_flags & SAVESTART) {
694 ndp->ni_startdir = ndp->ni_dvp;
695 VREF(ndp->ni_startdir);
696 }
697 if (!wantparent) {
698 if (ndp->ni_dvp != dp)
699 vput(ndp->ni_dvp);
700 else
701 vrele(ndp->ni_dvp);
702 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
703 VOP_UNLOCK(ndp->ni_dvp, 0, td);
704
705 if ((cnp->cn_flags & LOCKLEAF) == 0)
706 VOP_UNLOCK(dp, 0, td);
707 success:
708 if (vfslocked)
709 ndp->ni_cnd.cn_flags |= GIANTHELD;
710 return (0);
711
712 bad2:
713 if (dp != ndp->ni_dvp)
714 vput(ndp->ni_dvp);
715 else
716 vrele(ndp->ni_dvp);
717 bad:
718 if (!dpunlocked)
719 vput(dp);
720 VFS_UNLOCK_GIANT(vfslocked);
721 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
722 ndp->ni_vp = NULL;
723 return (error);
724 }
725
726 /*
727 * relookup - lookup a path name component
728 * Used by lookup to re-aquire things.
729 */
730 int
731 relookup(dvp, vpp, cnp)
732 struct vnode *dvp, **vpp;
733 struct componentname *cnp;
734 {
735 struct thread *td = cnp->cn_thread;
736 struct vnode *dp = 0; /* the directory we are searching */
737 int wantparent; /* 1 => wantparent or lockparent flag */
738 int rdonly; /* lookup read-only flag bit */
739 int error = 0;
740
741 KASSERT(cnp->cn_flags & ISLASTCN,
742 ("relookup: Not given last component."));
743 /*
744 * Setup: break out flag bits into variables.
745 */
746 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
747 KASSERT(wantparent, ("relookup: parent not wanted."));
748 rdonly = cnp->cn_flags & RDONLY;
749 cnp->cn_flags &= ~ISSYMLINK;
750 dp = dvp;
751 cnp->cn_lkflags = LK_EXCLUSIVE;
752 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
753
754 /*
755 * Search a new directory.
756 *
757 * The last component of the filename is left accessible via
758 * cnp->cn_nameptr for callers that need the name. Callers needing
759 * the name set the SAVENAME flag. When done, they assume
760 * responsibility for freeing the pathname buffer.
761 */
762 #ifdef NAMEI_DIAGNOSTIC
763 printf("{%s}: ", cnp->cn_nameptr);
764 #endif
765
766 /*
767 * Check for degenerate name (e.g. / or "")
768 * which is a way of talking about a directory,
769 * e.g. like "/." or ".".
770 */
771 if (cnp->cn_nameptr[0] == '\0') {
772 if (cnp->cn_nameiop != LOOKUP || wantparent) {
773 error = EISDIR;
774 goto bad;
775 }
776 if (dp->v_type != VDIR) {
777 error = ENOTDIR;
778 goto bad;
779 }
780 if (!(cnp->cn_flags & LOCKLEAF))
781 VOP_UNLOCK(dp, 0, td);
782 *vpp = dp;
783 /* XXX This should probably move to the top of function. */
784 if (cnp->cn_flags & SAVESTART)
785 panic("lookup: SAVESTART");
786 return (0);
787 }
788
789 if (cnp->cn_flags & ISDOTDOT)
790 panic ("relookup: lookup on dot-dot");
791
792 /*
793 * We now have a segment name to search for, and a directory to search.
794 */
795 #ifdef NAMEI_DIAGNOSTIC
796 vprint("search in:", dp);
797 #endif
798 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
799 KASSERT(*vpp == NULL, ("leaf should be empty"));
800 if (error != EJUSTRETURN)
801 goto bad;
802 /*
803 * If creating and at end of pathname, then can consider
804 * allowing file to be created.
805 */
806 if (rdonly) {
807 error = EROFS;
808 goto bad;
809 }
810 /* ASSERT(dvp == ndp->ni_startdir) */
811 if (cnp->cn_flags & SAVESTART)
812 VREF(dvp);
813 if ((cnp->cn_flags & LOCKPARENT) == 0)
814 VOP_UNLOCK(dp, 0, td);
815 /*
816 * This is a temporary assert to make sure I know what the
817 * behavior here was.
818 */
819 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
820 ("relookup: Unhandled case."));
821 /*
822 * We return with ni_vp NULL to indicate that the entry
823 * doesn't currently exist, leaving a pointer to the
824 * (possibly locked) directory inode in ndp->ni_dvp.
825 */
826 return (0);
827 }
828 dp = *vpp;
829
830 /*
831 * Disallow directory write attempts on read-only filesystems.
832 */
833 if (rdonly &&
834 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
835 if (dvp == dp)
836 vrele(dvp);
837 else
838 vput(dvp);
839 error = EROFS;
840 goto bad;
841 }
842 /*
843 * Set the parent lock/ref state to the requested state.
844 */
845 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
846 if (wantparent)
847 VOP_UNLOCK(dvp, 0, td);
848 else
849 vput(dvp);
850 } else if (!wantparent)
851 vrele(dvp);
852 /*
853 * Check for symbolic link
854 */
855 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
856 ("relookup: symlink found.\n"));
857
858 /* ASSERT(dvp == ndp->ni_startdir) */
859 if (cnp->cn_flags & SAVESTART)
860 VREF(dvp);
861
862 if ((cnp->cn_flags & LOCKLEAF) == 0)
863 VOP_UNLOCK(dp, 0, td);
864 return (0);
865 bad:
866 vput(dp);
867 *vpp = NULL;
868 return (error);
869 }
870
871 /*
872 * Free data allocated by namei(); see namei(9) for details.
873 */
874 void
875 NDFREE(ndp, flags)
876 struct nameidata *ndp;
877 const u_int flags;
878 {
879 int unlock_dvp;
880 int unlock_vp;
881
882 unlock_dvp = 0;
883 unlock_vp = 0;
884
885 if (!(flags & NDF_NO_FREE_PNBUF) &&
886 (ndp->ni_cnd.cn_flags & HASBUF)) {
887 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
888 ndp->ni_cnd.cn_flags &= ~HASBUF;
889 }
890 if (!(flags & NDF_NO_VP_UNLOCK) &&
891 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
892 unlock_vp = 1;
893 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
894 if (unlock_vp) {
895 vput(ndp->ni_vp);
896 unlock_vp = 0;
897 } else
898 vrele(ndp->ni_vp);
899 ndp->ni_vp = NULL;
900 }
901 if (unlock_vp)
902 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
903 if (!(flags & NDF_NO_DVP_UNLOCK) &&
904 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
905 ndp->ni_dvp != ndp->ni_vp)
906 unlock_dvp = 1;
907 if (!(flags & NDF_NO_DVP_RELE) &&
908 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
909 if (unlock_dvp) {
910 vput(ndp->ni_dvp);
911 unlock_dvp = 0;
912 } else
913 vrele(ndp->ni_dvp);
914 ndp->ni_dvp = NULL;
915 }
916 if (unlock_dvp)
917 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
918 if (!(flags & NDF_NO_STARTDIR_RELE) &&
919 (ndp->ni_cnd.cn_flags & SAVESTART)) {
920 vrele(ndp->ni_startdir);
921 ndp->ni_startdir = NULL;
922 }
923 }
924
925 /*
926 * Determine if there is a suitable alternate filename under the specified
927 * prefix for the specified path. If the create flag is set, then the
928 * alternate prefix will be used so long as the parent directory exists.
929 * This is used by the various compatiblity ABIs so that Linux binaries prefer
930 * files under /compat/linux for example. The chosen path (whether under
931 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
932 * to by pathbuf. The caller is responsible for free'ing the buffer from
933 * the M_TEMP bucket if one is returned.
934 */
935 int
936 kern_alternate_path(struct thread *td, const char *prefix, char *path,
937 enum uio_seg pathseg, char **pathbuf, int create)
938 {
939 struct nameidata nd, ndroot;
940 char *ptr, *buf, *cp;
941 size_t len, sz;
942 int error;
943
944 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
945 *pathbuf = buf;
946
947 /* Copy the prefix into the new pathname as a starting point. */
948 len = strlcpy(buf, prefix, MAXPATHLEN);
949 if (len >= MAXPATHLEN) {
950 *pathbuf = NULL;
951 free(buf, M_TEMP);
952 return (EINVAL);
953 }
954 sz = MAXPATHLEN - len;
955 ptr = buf + len;
956
957 /* Append the filename to the prefix. */
958 if (pathseg == UIO_SYSSPACE)
959 error = copystr(path, ptr, sz, &len);
960 else
961 error = copyinstr(path, ptr, sz, &len);
962
963 if (error) {
964 *pathbuf = NULL;
965 free(buf, M_TEMP);
966 return (error);
967 }
968
969 /* Only use a prefix with absolute pathnames. */
970 if (*ptr != '/') {
971 error = EINVAL;
972 goto keeporig;
973 }
974
975 /*
976 * We know that there is a / somewhere in this pathname.
977 * Search backwards for it, to find the file's parent dir
978 * to see if it exists in the alternate tree. If it does,
979 * and we want to create a file (cflag is set). We don't
980 * need to worry about the root comparison in this case.
981 */
982
983 if (create) {
984 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
985 *cp = '\0';
986
987 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
988 error = namei(&nd);
989 *cp = '/';
990 if (error != 0)
991 goto keeporig;
992 } else {
993 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
994
995 error = namei(&nd);
996 if (error != 0)
997 goto keeporig;
998
999 /*
1000 * We now compare the vnode of the prefix to the one
1001 * vnode asked. If they resolve to be the same, then we
1002 * ignore the match so that the real root gets used.
1003 * This avoids the problem of traversing "../.." to find the
1004 * root directory and never finding it, because "/" resolves
1005 * to the emulation root directory. This is expensive :-(
1006 */
1007 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1008 td);
1009
1010 /* We shouldn't ever get an error from this namei(). */
1011 error = namei(&ndroot);
1012 if (error == 0) {
1013 if (nd.ni_vp == ndroot.ni_vp)
1014 error = ENOENT;
1015
1016 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1017 vrele(ndroot.ni_vp);
1018 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1019 }
1020 }
1021
1022 NDFREE(&nd, NDF_ONLY_PNBUF);
1023 vrele(nd.ni_vp);
1024 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1025
1026 keeporig:
1027 /* If there was an error, use the original path name. */
1028 if (error)
1029 bcopy(ptr, buf, len);
1030 return (error);
1031 }
Cache object: 465d32340c9747f219bac98e634521dc
|