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/6.2/sys/kern/vfs_lookup.c 164286 2006-11-14 20:42:41Z cvs2svn $");
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 <security/audit/audit.h>
62
63 #include <vm/uma.h>
64
65 #define NAMEI_DIAGNOSTIC 1
66 #undef NAMEI_DIAGNOSTIC
67
68 /*
69 * Allocation zone for namei
70 */
71 uma_zone_t namei_zone;
72
73 static void
74 nameiinit(void *dummy __unused)
75 {
76 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
77 UMA_ALIGN_PTR, 0);
78
79 }
80 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL)
81
82 #ifdef LOOKUP_SHARED
83 static int lookup_shared = 1;
84 #else
85 static int lookup_shared = 0;
86 #endif
87 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
88 "Enables/Disables shared locks for path name translation");
89
90 /*
91 * Convert a pathname into a pointer to a locked vnode.
92 *
93 * The FOLLOW flag is set when symbolic links are to be followed
94 * when they occur at the end of the name translation process.
95 * Symbolic links are always followed for all other pathname
96 * components other than the last.
97 *
98 * The segflg defines whether the name is to be copied from user
99 * space or kernel space.
100 *
101 * Overall outline of namei:
102 *
103 * copy in name
104 * get starting directory
105 * while (!done && !error) {
106 * call lookup to search path.
107 * if symbolic link, massage name in buffer and continue
108 * }
109 */
110 int
111 namei(ndp)
112 register struct nameidata *ndp;
113 {
114 register struct filedesc *fdp; /* pointer to file descriptor state */
115 register char *cp; /* pointer into pathname argument */
116 register struct vnode *dp; /* the directory we are searching */
117 struct iovec aiov; /* uio for reading symbolic links */
118 struct uio auio;
119 int error, linklen;
120 struct componentname *cnp = &ndp->ni_cnd;
121 struct thread *td = cnp->cn_thread;
122 struct proc *p = td->td_proc;
123 int vfslocked;
124
125 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
126 ("NOT MPSAFE and Giant not held"));
127 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
128 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
129 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
130 ("namei: nameiop contaminated with flags"));
131 KASSERT((cnp->cn_flags & OPMASK) == 0,
132 ("namei: flags contaminated with nameiops"));
133 if (!lookup_shared)
134 cnp->cn_flags &= ~LOCKSHARED;
135 fdp = p->p_fd;
136
137 /*
138 * Get a buffer for the name to be translated, and copy the
139 * name into the buffer.
140 */
141 if ((cnp->cn_flags & HASBUF) == 0)
142 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
143 if (ndp->ni_segflg == UIO_SYSSPACE)
144 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
145 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
146 else
147 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
148 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
149
150 /* If we are auditing the kernel pathname, save the user pathname. */
151 if (cnp->cn_flags & AUDITVNODE1)
152 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH1);
153 if (cnp->cn_flags & AUDITVNODE2)
154 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH2);
155
156 /*
157 * Don't allow empty pathnames.
158 */
159 if (!error && *cnp->cn_pnbuf == '\0')
160 error = ENOENT;
161
162 if (error) {
163 uma_zfree(namei_zone, cnp->cn_pnbuf);
164 #ifdef DIAGNOSTIC
165 cnp->cn_pnbuf = NULL;
166 cnp->cn_nameptr = NULL;
167 #endif
168 ndp->ni_vp = NULL;
169 return (error);
170 }
171 ndp->ni_loopcnt = 0;
172 #ifdef KTRACE
173 if (KTRPOINT(td, KTR_NAMEI)) {
174 KASSERT(cnp->cn_thread == curthread,
175 ("namei not using curthread"));
176 ktrnamei(cnp->cn_pnbuf);
177 }
178 #endif
179
180 /*
181 * Get starting point for the translation.
182 */
183 FILEDESC_LOCK(fdp);
184 ndp->ni_rootdir = fdp->fd_rdir;
185 ndp->ni_topdir = fdp->fd_jdir;
186
187 dp = fdp->fd_cdir;
188 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
189 VREF(dp);
190 FILEDESC_UNLOCK(fdp);
191 for (;;) {
192 /*
193 * Check if root directory should replace current directory.
194 * Done at start of translation and after symbolic link.
195 */
196 cnp->cn_nameptr = cnp->cn_pnbuf;
197 if (*(cnp->cn_nameptr) == '/') {
198 vrele(dp);
199 VFS_UNLOCK_GIANT(vfslocked);
200 while (*(cnp->cn_nameptr) == '/') {
201 cnp->cn_nameptr++;
202 ndp->ni_pathlen--;
203 }
204 dp = ndp->ni_rootdir;
205 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
206 VREF(dp);
207 }
208 if (vfslocked)
209 ndp->ni_cnd.cn_flags |= GIANTHELD;
210 ndp->ni_startdir = dp;
211 error = lookup(ndp);
212 if (error) {
213 uma_zfree(namei_zone, cnp->cn_pnbuf);
214 #ifdef DIAGNOSTIC
215 cnp->cn_pnbuf = NULL;
216 cnp->cn_nameptr = NULL;
217 #endif
218 return (error);
219 }
220 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
221 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
222 /*
223 * Check for symbolic link
224 */
225 if ((cnp->cn_flags & ISSYMLINK) == 0) {
226 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
227 uma_zfree(namei_zone, cnp->cn_pnbuf);
228 #ifdef DIAGNOSTIC
229 cnp->cn_pnbuf = NULL;
230 cnp->cn_nameptr = NULL;
231 #endif
232 } else
233 cnp->cn_flags |= HASBUF;
234
235 if ((cnp->cn_flags & MPSAFE) == 0) {
236 VFS_UNLOCK_GIANT(vfslocked);
237 } else if (vfslocked)
238 ndp->ni_cnd.cn_flags |= GIANTHELD;
239 return (0);
240 }
241 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
242 error = ELOOP;
243 break;
244 }
245 #ifdef MAC
246 if ((cnp->cn_flags & NOMACCHECK) == 0) {
247 error = mac_check_vnode_readlink(td->td_ucred,
248 ndp->ni_vp);
249 if (error)
250 break;
251 }
252 #endif
253 if (ndp->ni_pathlen > 1)
254 cp = uma_zalloc(namei_zone, M_WAITOK);
255 else
256 cp = cnp->cn_pnbuf;
257 aiov.iov_base = cp;
258 aiov.iov_len = MAXPATHLEN;
259 auio.uio_iov = &aiov;
260 auio.uio_iovcnt = 1;
261 auio.uio_offset = 0;
262 auio.uio_rw = UIO_READ;
263 auio.uio_segflg = UIO_SYSSPACE;
264 auio.uio_td = (struct thread *)0;
265 auio.uio_resid = MAXPATHLEN;
266 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
267 if (error) {
268 if (ndp->ni_pathlen > 1)
269 uma_zfree(namei_zone, cp);
270 break;
271 }
272 linklen = MAXPATHLEN - auio.uio_resid;
273 if (linklen == 0) {
274 if (ndp->ni_pathlen > 1)
275 uma_zfree(namei_zone, cp);
276 error = ENOENT;
277 break;
278 }
279 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
280 if (ndp->ni_pathlen > 1)
281 uma_zfree(namei_zone, cp);
282 error = ENAMETOOLONG;
283 break;
284 }
285 if (ndp->ni_pathlen > 1) {
286 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
287 uma_zfree(namei_zone, cnp->cn_pnbuf);
288 cnp->cn_pnbuf = cp;
289 } else
290 cnp->cn_pnbuf[linklen] = '\0';
291 ndp->ni_pathlen += linklen;
292 vput(ndp->ni_vp);
293 dp = ndp->ni_dvp;
294 }
295 uma_zfree(namei_zone, cnp->cn_pnbuf);
296 #ifdef DIAGNOSTIC
297 cnp->cn_pnbuf = NULL;
298 cnp->cn_nameptr = NULL;
299 #endif
300 vput(ndp->ni_vp);
301 ndp->ni_vp = NULL;
302 vrele(ndp->ni_dvp);
303 VFS_UNLOCK_GIANT(vfslocked);
304 return (error);
305 }
306
307 /*
308 * Search a pathname.
309 * This is a very central and rather complicated routine.
310 *
311 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
312 * The starting directory is taken from ni_startdir. The pathname is
313 * descended until done, or a symbolic link is encountered. The variable
314 * ni_more is clear if the path is completed; it is set to one if a
315 * symbolic link needing interpretation is encountered.
316 *
317 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
318 * whether the name is to be looked up, created, renamed, or deleted.
319 * When CREATE, RENAME, or DELETE is specified, information usable in
320 * creating, renaming, or deleting a directory entry may be calculated.
321 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
322 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
323 * returned unlocked. Otherwise the parent directory is not returned. If
324 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
325 * the target is returned locked, otherwise it is returned unlocked.
326 * When creating or renaming and LOCKPARENT is specified, the target may not
327 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
328 *
329 * Overall outline of lookup:
330 *
331 * dirloop:
332 * identify next component of name at ndp->ni_ptr
333 * handle degenerate case where name is null string
334 * if .. and crossing mount points and on mounted filesys, find parent
335 * call VOP_LOOKUP routine for next component name
336 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
337 * component vnode returned in ni_vp (if it exists), locked.
338 * if result vnode is mounted on and crossing mount points,
339 * find mounted on vnode
340 * if more components of name, do next level at dirloop
341 * return the answer in ni_vp, locked if LOCKLEAF set
342 * if LOCKPARENT set, return locked parent in ni_dvp
343 * if WANTPARENT set, return unlocked parent in ni_dvp
344 */
345 int
346 lookup(ndp)
347 register struct nameidata *ndp;
348 {
349 register char *cp; /* pointer into pathname argument */
350 register struct vnode *dp = 0; /* the directory we are searching */
351 struct vnode *tdp; /* saved dp */
352 struct mount *mp; /* mount table entry */
353 int docache; /* == 0 do not cache last component */
354 int wantparent; /* 1 => wantparent or lockparent flag */
355 int rdonly; /* lookup read-only flag bit */
356 int trailing_slash;
357 int error = 0;
358 int dpunlocked = 0; /* dp has already been unlocked */
359 struct componentname *cnp = &ndp->ni_cnd;
360 struct thread *td = cnp->cn_thread;
361 int vfslocked; /* VFS Giant state for child */
362 int dvfslocked; /* VFS Giant state for parent */
363 int tvfslocked;
364
365 /*
366 * Setup: break out flag bits into variables.
367 */
368 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
369 vfslocked = 0;
370 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
371 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
372 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
373 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
374 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
375 if (cnp->cn_nameiop == DELETE ||
376 (wantparent && cnp->cn_nameiop != CREATE &&
377 cnp->cn_nameiop != LOOKUP))
378 docache = 0;
379 rdonly = cnp->cn_flags & RDONLY;
380 cnp->cn_flags &= ~ISSYMLINK;
381 ndp->ni_dvp = NULL;
382 /*
383 * We use shared locks until we hit the parent of the last cn then
384 * we adjust based on the requesting flags.
385 */
386 if (lookup_shared)
387 cnp->cn_lkflags = LK_SHARED;
388 else
389 cnp->cn_lkflags = LK_EXCLUSIVE;
390 dp = ndp->ni_startdir;
391 ndp->ni_startdir = NULLVP;
392 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
393
394 dirloop:
395 /*
396 * Search a new directory.
397 *
398 * The last component of the filename is left accessible via
399 * cnp->cn_nameptr for callers that need the name. Callers needing
400 * the name set the SAVENAME flag. When done, they assume
401 * responsibility for freeing the pathname buffer.
402 */
403 cnp->cn_consume = 0;
404 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
405 continue;
406 cnp->cn_namelen = cp - cnp->cn_nameptr;
407 if (cnp->cn_namelen > NAME_MAX) {
408 error = ENAMETOOLONG;
409 goto bad;
410 }
411 #ifdef NAMEI_DIAGNOSTIC
412 { char c = *cp;
413 *cp = '\0';
414 printf("{%s}: ", cnp->cn_nameptr);
415 *cp = c; }
416 #endif
417 ndp->ni_pathlen -= cnp->cn_namelen;
418 ndp->ni_next = cp;
419
420 /*
421 * Replace multiple slashes by a single slash and trailing slashes
422 * by a null. This must be done before VOP_LOOKUP() because some
423 * fs's don't know about trailing slashes. Remember if there were
424 * trailing slashes to handle symlinks, existing non-directories
425 * and non-existing files that won't be directories specially later.
426 */
427 trailing_slash = 0;
428 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
429 cp++;
430 ndp->ni_pathlen--;
431 if (*cp == '\0') {
432 trailing_slash = 1;
433 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
434 }
435 }
436 ndp->ni_next = cp;
437
438 cnp->cn_flags |= MAKEENTRY;
439 if (*cp == '\0' && docache == 0)
440 cnp->cn_flags &= ~MAKEENTRY;
441 if (cnp->cn_namelen == 2 &&
442 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
443 cnp->cn_flags |= ISDOTDOT;
444 else
445 cnp->cn_flags &= ~ISDOTDOT;
446 if (*ndp->ni_next == 0)
447 cnp->cn_flags |= ISLASTCN;
448 else
449 cnp->cn_flags &= ~ISLASTCN;
450
451
452 /*
453 * Check for degenerate name (e.g. / or "")
454 * which is a way of talking about a directory,
455 * e.g. like "/." or ".".
456 */
457 if (cnp->cn_nameptr[0] == '\0') {
458 if (dp->v_type != VDIR) {
459 error = ENOTDIR;
460 goto bad;
461 }
462 if (cnp->cn_nameiop != LOOKUP) {
463 error = EISDIR;
464 goto bad;
465 }
466 if (wantparent) {
467 ndp->ni_dvp = dp;
468 VREF(dp);
469 }
470 ndp->ni_vp = dp;
471
472 if (cnp->cn_flags & AUDITVNODE1)
473 AUDIT_ARG(vnode, dp, ARG_VNODE1);
474 else if (cnp->cn_flags & AUDITVNODE2)
475 AUDIT_ARG(vnode, dp, ARG_VNODE2);
476
477 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
478 VOP_UNLOCK(dp, 0, td);
479 /* XXX This should probably move to the top of function. */
480 if (cnp->cn_flags & SAVESTART)
481 panic("lookup: SAVESTART");
482 goto success;
483 }
484
485 /*
486 * Handle "..": four special cases.
487 * 1. Return an error if this is the last component of
488 * the name and the operation is DELETE or RENAME.
489 * 2. If at root directory (e.g. after chroot)
490 * or at absolute root directory
491 * then ignore it so can't get out.
492 * 3. If this vnode is the root of a mounted
493 * filesystem, then replace it with the
494 * vnode which was mounted on so we take the
495 * .. in the other filesystem.
496 * 4. If the vnode is the top directory of
497 * the jail or chroot, don't let them out.
498 */
499 if (cnp->cn_flags & ISDOTDOT) {
500 if ((cnp->cn_flags & ISLASTCN) != 0 &&
501 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
502 error = EINVAL;
503 goto bad;
504 }
505 for (;;) {
506 if (dp == ndp->ni_rootdir ||
507 dp == ndp->ni_topdir ||
508 dp == rootvnode) {
509 ndp->ni_dvp = dp;
510 ndp->ni_vp = dp;
511 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
512 VREF(dp);
513 goto nextname;
514 }
515 if ((dp->v_vflag & VV_ROOT) == 0 ||
516 (cnp->cn_flags & NOCROSSMOUNT))
517 break;
518 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
519 error = EBADF;
520 goto bad;
521 }
522 tdp = dp;
523 dp = dp->v_mount->mnt_vnodecovered;
524 tvfslocked = dvfslocked;
525 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
526 VREF(dp);
527 vput(tdp);
528 VFS_UNLOCK_GIANT(tvfslocked);
529 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
530 }
531 }
532
533 /*
534 * We now have a segment name to search for, and a directory to search.
535 */
536 unionlookup:
537 #ifdef MAC
538 if ((cnp->cn_flags & NOMACCHECK) == 0) {
539 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
540 if (error)
541 goto bad;
542 }
543 #endif
544 ndp->ni_dvp = dp;
545 ndp->ni_vp = NULL;
546 ASSERT_VOP_LOCKED(dp, "lookup");
547 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
548 /*
549 * If we have a shared lock we may need to upgrade the lock for the
550 * last operation.
551 */
552 if (VOP_ISLOCKED(dp, td) == LK_SHARED &&
553 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
554 vn_lock(dp, LK_UPGRADE|LK_RETRY, td);
555 /*
556 * If we're looking up the last component and we need an exclusive
557 * lock, adjust our lkflags.
558 */
559 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
560 (ISLASTCN|LOCKLEAF))
561 cnp->cn_lkflags = LK_EXCLUSIVE;
562 #ifdef NAMEI_DIAGNOSTIC
563 vprint("lookup in", dp);
564 #endif
565 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
566 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
567 #ifdef NAMEI_DIAGNOSTIC
568 printf("not found\n");
569 #endif
570 if ((error == ENOENT) &&
571 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
572 (dp->v_mount->mnt_flag & MNT_UNION)) {
573 tdp = dp;
574 dp = dp->v_mount->mnt_vnodecovered;
575 tvfslocked = dvfslocked;
576 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
577 VREF(dp);
578 vput(tdp);
579 VFS_UNLOCK_GIANT(tvfslocked);
580 vn_lock(dp, cnp->cn_lkflags | LK_RETRY, td);
581 goto unionlookup;
582 }
583
584 if (error != EJUSTRETURN)
585 goto bad;
586 /*
587 * If creating and at end of pathname, then can consider
588 * allowing file to be created.
589 */
590 if (rdonly) {
591 error = EROFS;
592 goto bad;
593 }
594 if (*cp == '\0' && trailing_slash &&
595 !(cnp->cn_flags & WILLBEDIR)) {
596 error = ENOENT;
597 goto bad;
598 }
599 if ((cnp->cn_flags & LOCKPARENT) == 0)
600 VOP_UNLOCK(dp, 0, td);
601 /*
602 * This is a temporary assert to make sure I know what the
603 * behavior here was.
604 */
605 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
606 ("lookup: Unhandled case."));
607 /*
608 * We return with ni_vp NULL to indicate that the entry
609 * doesn't currently exist, leaving a pointer to the
610 * (possibly locked) directory vnode in ndp->ni_dvp.
611 */
612 if (cnp->cn_flags & SAVESTART) {
613 ndp->ni_startdir = ndp->ni_dvp;
614 VREF(ndp->ni_startdir);
615 }
616 goto success;
617 }
618 #ifdef NAMEI_DIAGNOSTIC
619 printf("found\n");
620 #endif
621 /*
622 * Take into account any additional components consumed by
623 * the underlying filesystem.
624 */
625 if (cnp->cn_consume > 0) {
626 cnp->cn_nameptr += cnp->cn_consume;
627 ndp->ni_next += cnp->cn_consume;
628 ndp->ni_pathlen -= cnp->cn_consume;
629 cnp->cn_consume = 0;
630 }
631
632 dp = ndp->ni_vp;
633 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
634
635 /*
636 * Check to see if the vnode has been mounted on;
637 * if so find the root of the mounted filesystem.
638 */
639 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
640 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
641 if (vfs_busy(mp, 0, 0, td))
642 continue;
643 vput(dp);
644 VFS_UNLOCK_GIANT(vfslocked);
645 vfslocked = VFS_LOCK_GIANT(mp);
646 if (dp != ndp->ni_dvp)
647 VOP_UNLOCK(ndp->ni_dvp, 0, td);
648 error = VFS_ROOT(mp, cnp->cn_lkflags, &tdp, td);
649 vfs_unbusy(mp, td);
650 vn_lock(ndp->ni_dvp, cnp->cn_lkflags | LK_RETRY, td);
651 if (error) {
652 dpunlocked = 1;
653 goto bad2;
654 }
655 ndp->ni_vp = dp = tdp;
656 }
657
658 /*
659 * Check for symbolic link
660 */
661 if ((dp->v_type == VLNK) &&
662 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
663 *ndp->ni_next == '/')) {
664 cnp->cn_flags |= ISSYMLINK;
665 if (dp->v_iflag & VI_DOOMED) {
666 /* We can't know whether the directory was mounted with
667 * NOSYMFOLLOW, so we can't follow safely. */
668 error = EBADF;
669 goto bad2;
670 }
671 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
672 error = EACCES;
673 goto bad2;
674 }
675 /*
676 * Symlink code always expects an unlocked dvp.
677 */
678 if (ndp->ni_dvp != ndp->ni_vp)
679 VOP_UNLOCK(ndp->ni_dvp, 0, td);
680 goto success;
681 }
682
683 /*
684 * Check for bogus trailing slashes.
685 */
686 if (trailing_slash && dp->v_type != VDIR) {
687 error = ENOTDIR;
688 goto bad2;
689 }
690
691 nextname:
692 /*
693 * Not a symbolic link. If more pathname,
694 * continue at next component, else return.
695 */
696 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
697 ("lookup: invalid path state."));
698 if (*ndp->ni_next == '/') {
699 cnp->cn_nameptr = ndp->ni_next;
700 while (*cnp->cn_nameptr == '/') {
701 cnp->cn_nameptr++;
702 ndp->ni_pathlen--;
703 }
704 if (ndp->ni_dvp != dp)
705 vput(ndp->ni_dvp);
706 else
707 vrele(ndp->ni_dvp);
708 VFS_UNLOCK_GIANT(dvfslocked);
709 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
710 vfslocked = 0;
711 goto dirloop;
712 }
713 /*
714 * Disallow directory write attempts on read-only filesystems.
715 */
716 if (rdonly &&
717 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
718 error = EROFS;
719 goto bad2;
720 }
721 if (cnp->cn_flags & SAVESTART) {
722 ndp->ni_startdir = ndp->ni_dvp;
723 VREF(ndp->ni_startdir);
724 }
725 if (!wantparent) {
726 if (ndp->ni_dvp != dp)
727 vput(ndp->ni_dvp);
728 else
729 vrele(ndp->ni_dvp);
730 VFS_UNLOCK_GIANT(dvfslocked);
731 dvfslocked = 0;
732 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
733 VOP_UNLOCK(ndp->ni_dvp, 0, td);
734
735 if (cnp->cn_flags & AUDITVNODE1)
736 AUDIT_ARG(vnode, dp, ARG_VNODE1);
737 else if (cnp->cn_flags & AUDITVNODE2)
738 AUDIT_ARG(vnode, dp, ARG_VNODE2);
739
740 if ((cnp->cn_flags & LOCKLEAF) == 0)
741 VOP_UNLOCK(dp, 0, td);
742 success:
743 if (vfslocked && dvfslocked)
744 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
745 if (vfslocked || dvfslocked)
746 ndp->ni_cnd.cn_flags |= GIANTHELD;
747 return (0);
748
749 bad2:
750 if (dp != ndp->ni_dvp)
751 vput(ndp->ni_dvp);
752 else
753 vrele(ndp->ni_dvp);
754 bad:
755 if (!dpunlocked)
756 vput(dp);
757 VFS_UNLOCK_GIANT(vfslocked);
758 VFS_UNLOCK_GIANT(dvfslocked);
759 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
760 ndp->ni_vp = NULL;
761 return (error);
762 }
763
764 /*
765 * relookup - lookup a path name component
766 * Used by lookup to re-aquire things.
767 */
768 int
769 relookup(dvp, vpp, cnp)
770 struct vnode *dvp, **vpp;
771 struct componentname *cnp;
772 {
773 struct thread *td = cnp->cn_thread;
774 struct vnode *dp = 0; /* the directory we are searching */
775 int wantparent; /* 1 => wantparent or lockparent flag */
776 int rdonly; /* lookup read-only flag bit */
777 int error = 0;
778
779 KASSERT(cnp->cn_flags & ISLASTCN,
780 ("relookup: Not given last component."));
781 /*
782 * Setup: break out flag bits into variables.
783 */
784 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
785 KASSERT(wantparent, ("relookup: parent not wanted."));
786 rdonly = cnp->cn_flags & RDONLY;
787 cnp->cn_flags &= ~ISSYMLINK;
788 dp = dvp;
789 cnp->cn_lkflags = LK_EXCLUSIVE;
790 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
791
792 /*
793 * Search a new directory.
794 *
795 * The last component of the filename is left accessible via
796 * cnp->cn_nameptr for callers that need the name. Callers needing
797 * the name set the SAVENAME flag. When done, they assume
798 * responsibility for freeing the pathname buffer.
799 */
800 #ifdef NAMEI_DIAGNOSTIC
801 printf("{%s}: ", cnp->cn_nameptr);
802 #endif
803
804 /*
805 * Check for degenerate name (e.g. / or "")
806 * which is a way of talking about a directory,
807 * e.g. like "/." or ".".
808 */
809 if (cnp->cn_nameptr[0] == '\0') {
810 if (cnp->cn_nameiop != LOOKUP || wantparent) {
811 error = EISDIR;
812 goto bad;
813 }
814 if (dp->v_type != VDIR) {
815 error = ENOTDIR;
816 goto bad;
817 }
818 if (!(cnp->cn_flags & LOCKLEAF))
819 VOP_UNLOCK(dp, 0, td);
820 *vpp = dp;
821 /* XXX This should probably move to the top of function. */
822 if (cnp->cn_flags & SAVESTART)
823 panic("lookup: SAVESTART");
824 return (0);
825 }
826
827 if (cnp->cn_flags & ISDOTDOT)
828 panic ("relookup: lookup on dot-dot");
829
830 /*
831 * We now have a segment name to search for, and a directory to search.
832 */
833 #ifdef NAMEI_DIAGNOSTIC
834 vprint("search in:", dp);
835 #endif
836 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
837 KASSERT(*vpp == NULL, ("leaf should be empty"));
838 if (error != EJUSTRETURN)
839 goto bad;
840 /*
841 * If creating and at end of pathname, then can consider
842 * allowing file to be created.
843 */
844 if (rdonly) {
845 error = EROFS;
846 goto bad;
847 }
848 /* ASSERT(dvp == ndp->ni_startdir) */
849 if (cnp->cn_flags & SAVESTART)
850 VREF(dvp);
851 if ((cnp->cn_flags & LOCKPARENT) == 0)
852 VOP_UNLOCK(dp, 0, td);
853 /*
854 * This is a temporary assert to make sure I know what the
855 * behavior here was.
856 */
857 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
858 ("relookup: Unhandled case."));
859 /*
860 * We return with ni_vp NULL to indicate that the entry
861 * doesn't currently exist, leaving a pointer to the
862 * (possibly locked) directory vnode in ndp->ni_dvp.
863 */
864 return (0);
865 }
866 dp = *vpp;
867
868 /*
869 * Disallow directory write attempts on read-only filesystems.
870 */
871 if (rdonly &&
872 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
873 if (dvp == dp)
874 vrele(dvp);
875 else
876 vput(dvp);
877 error = EROFS;
878 goto bad;
879 }
880 /*
881 * Set the parent lock/ref state to the requested state.
882 */
883 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
884 if (wantparent)
885 VOP_UNLOCK(dvp, 0, td);
886 else
887 vput(dvp);
888 } else if (!wantparent)
889 vrele(dvp);
890 /*
891 * Check for symbolic link
892 */
893 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
894 ("relookup: symlink found.\n"));
895
896 /* ASSERT(dvp == ndp->ni_startdir) */
897 if (cnp->cn_flags & SAVESTART)
898 VREF(dvp);
899
900 if ((cnp->cn_flags & LOCKLEAF) == 0)
901 VOP_UNLOCK(dp, 0, td);
902 return (0);
903 bad:
904 vput(dp);
905 *vpp = NULL;
906 return (error);
907 }
908
909 /*
910 * Free data allocated by namei(); see namei(9) for details.
911 */
912 void
913 NDFREE(ndp, flags)
914 struct nameidata *ndp;
915 const u_int flags;
916 {
917 int unlock_dvp;
918 int unlock_vp;
919
920 unlock_dvp = 0;
921 unlock_vp = 0;
922
923 if (!(flags & NDF_NO_FREE_PNBUF) &&
924 (ndp->ni_cnd.cn_flags & HASBUF)) {
925 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
926 ndp->ni_cnd.cn_flags &= ~HASBUF;
927 }
928 if (!(flags & NDF_NO_VP_UNLOCK) &&
929 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
930 unlock_vp = 1;
931 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
932 if (unlock_vp) {
933 vput(ndp->ni_vp);
934 unlock_vp = 0;
935 } else
936 vrele(ndp->ni_vp);
937 ndp->ni_vp = NULL;
938 }
939 if (unlock_vp)
940 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
941 if (!(flags & NDF_NO_DVP_UNLOCK) &&
942 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
943 ndp->ni_dvp != ndp->ni_vp)
944 unlock_dvp = 1;
945 if (!(flags & NDF_NO_DVP_RELE) &&
946 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
947 if (unlock_dvp) {
948 vput(ndp->ni_dvp);
949 unlock_dvp = 0;
950 } else
951 vrele(ndp->ni_dvp);
952 ndp->ni_dvp = NULL;
953 }
954 if (unlock_dvp)
955 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
956 if (!(flags & NDF_NO_STARTDIR_RELE) &&
957 (ndp->ni_cnd.cn_flags & SAVESTART)) {
958 vrele(ndp->ni_startdir);
959 ndp->ni_startdir = NULL;
960 }
961 }
962
963 /*
964 * Determine if there is a suitable alternate filename under the specified
965 * prefix for the specified path. If the create flag is set, then the
966 * alternate prefix will be used so long as the parent directory exists.
967 * This is used by the various compatiblity ABIs so that Linux binaries prefer
968 * files under /compat/linux for example. The chosen path (whether under
969 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
970 * to by pathbuf. The caller is responsible for free'ing the buffer from
971 * the M_TEMP bucket if one is returned.
972 */
973 int
974 kern_alternate_path(struct thread *td, const char *prefix, char *path,
975 enum uio_seg pathseg, char **pathbuf, int create)
976 {
977 struct nameidata nd, ndroot;
978 char *ptr, *buf, *cp;
979 size_t len, sz;
980 int error;
981
982 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
983 *pathbuf = buf;
984
985 /* Copy the prefix into the new pathname as a starting point. */
986 len = strlcpy(buf, prefix, MAXPATHLEN);
987 if (len >= MAXPATHLEN) {
988 *pathbuf = NULL;
989 free(buf, M_TEMP);
990 return (EINVAL);
991 }
992 sz = MAXPATHLEN - len;
993 ptr = buf + len;
994
995 /* Append the filename to the prefix. */
996 if (pathseg == UIO_SYSSPACE)
997 error = copystr(path, ptr, sz, &len);
998 else
999 error = copyinstr(path, ptr, sz, &len);
1000
1001 if (error) {
1002 *pathbuf = NULL;
1003 free(buf, M_TEMP);
1004 return (error);
1005 }
1006
1007 /* Only use a prefix with absolute pathnames. */
1008 if (*ptr != '/') {
1009 error = EINVAL;
1010 goto keeporig;
1011 }
1012
1013 /*
1014 * We know that there is a / somewhere in this pathname.
1015 * Search backwards for it, to find the file's parent dir
1016 * to see if it exists in the alternate tree. If it does,
1017 * and we want to create a file (cflag is set). We don't
1018 * need to worry about the root comparison in this case.
1019 */
1020
1021 if (create) {
1022 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1023 *cp = '\0';
1024
1025 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1026 error = namei(&nd);
1027 *cp = '/';
1028 if (error != 0)
1029 goto keeporig;
1030 } else {
1031 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1032
1033 error = namei(&nd);
1034 if (error != 0)
1035 goto keeporig;
1036
1037 /*
1038 * We now compare the vnode of the prefix to the one
1039 * vnode asked. If they resolve to be the same, then we
1040 * ignore the match so that the real root gets used.
1041 * This avoids the problem of traversing "../.." to find the
1042 * root directory and never finding it, because "/" resolves
1043 * to the emulation root directory. This is expensive :-(
1044 */
1045 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1046 td);
1047
1048 /* We shouldn't ever get an error from this namei(). */
1049 error = namei(&ndroot);
1050 if (error == 0) {
1051 if (nd.ni_vp == ndroot.ni_vp)
1052 error = ENOENT;
1053
1054 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1055 vrele(ndroot.ni_vp);
1056 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1057 }
1058 }
1059
1060 NDFREE(&nd, NDF_ONLY_PNBUF);
1061 vrele(nd.ni_vp);
1062 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1063
1064 keeporig:
1065 /* If there was an error, use the original path name. */
1066 if (error)
1067 bcopy(ptr, buf, len);
1068 return (error);
1069 }
Cache object: a8f7aaf7df4c8eaae881dbd102eaef5a
|