FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_lookup.c
1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_ktrace.h"
41 #include "opt_mac.h"
42 #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/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/vnode.h>
51 #include <sys/mount.h>
52 #include <sys/filedesc.h>
53 #include <sys/proc.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #ifdef KTRACE
57 #include <sys/ktrace.h>
58 #endif
59
60 #include <security/audit/audit.h>
61 #include <security/mac/mac_framework.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 * Placeholder vnode for mp traversal
74 */
75 static struct vnode *vp_crossmp;
76
77 static void
78 nameiinit(void *dummy __unused)
79 {
80 int error;
81
82 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
83 UMA_ALIGN_PTR, 0);
84 error = getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
85 if (error != 0)
86 panic("nameiinit: getnewvnode");
87 vp_crossmp->v_vnlock->lk_flags &= ~LK_NOSHARE;
88 }
89 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
90
91 #ifdef LOOKUP_SHARED
92 static int lookup_shared = 1;
93 #else
94 static int lookup_shared = 0;
95 #endif
96 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
97 "Enables/Disables shared locks for path name translation");
98
99 /*
100 * Convert a pathname into a pointer to a locked vnode.
101 *
102 * The FOLLOW flag is set when symbolic links are to be followed
103 * when they occur at the end of the name translation process.
104 * Symbolic links are always followed for all other pathname
105 * components other than the last.
106 *
107 * The segflg defines whether the name is to be copied from user
108 * space or kernel space.
109 *
110 * Overall outline of namei:
111 *
112 * copy in name
113 * get starting directory
114 * while (!done && !error) {
115 * call lookup to search path.
116 * if symbolic link, massage name in buffer and continue
117 * }
118 */
119 int
120 namei(struct nameidata *ndp)
121 {
122 struct filedesc *fdp; /* pointer to file descriptor state */
123 char *cp; /* pointer into pathname argument */
124 struct vnode *dp; /* the directory we are searching */
125 struct iovec aiov; /* uio for reading symbolic links */
126 struct uio auio;
127 int error, linklen;
128 struct componentname *cnp = &ndp->ni_cnd;
129 struct thread *td = cnp->cn_thread;
130 struct proc *p = td->td_proc;
131 int vfslocked;
132
133 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
134 ("NOT MPSAFE and Giant not held"));
135 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
136 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
137 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
138 ("namei: nameiop contaminated with flags"));
139 KASSERT((cnp->cn_flags & OPMASK) == 0,
140 ("namei: flags contaminated with nameiops"));
141 if (!lookup_shared)
142 cnp->cn_flags &= ~LOCKSHARED;
143 fdp = p->p_fd;
144
145 /*
146 * Get a buffer for the name to be translated, and copy the
147 * name into the buffer.
148 */
149 if ((cnp->cn_flags & HASBUF) == 0)
150 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
151 if (ndp->ni_segflg == UIO_SYSSPACE)
152 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
153 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
154 else
155 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
156 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
157
158 /* If we are auditing the kernel pathname, save the user pathname. */
159 if (cnp->cn_flags & AUDITVNODE1)
160 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH1);
161 if (cnp->cn_flags & AUDITVNODE2)
162 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH2);
163
164 /*
165 * Don't allow empty pathnames.
166 */
167 if (!error && *cnp->cn_pnbuf == '\0')
168 error = ENOENT;
169
170 if (error) {
171 uma_zfree(namei_zone, cnp->cn_pnbuf);
172 #ifdef DIAGNOSTIC
173 cnp->cn_pnbuf = NULL;
174 cnp->cn_nameptr = NULL;
175 #endif
176 ndp->ni_vp = NULL;
177 return (error);
178 }
179 ndp->ni_loopcnt = 0;
180 #ifdef KTRACE
181 if (KTRPOINT(td, KTR_NAMEI)) {
182 KASSERT(cnp->cn_thread == curthread,
183 ("namei not using curthread"));
184 ktrnamei(cnp->cn_pnbuf);
185 }
186 #endif
187
188 /*
189 * Get starting point for the translation.
190 */
191 FILEDESC_SLOCK(fdp);
192 ndp->ni_rootdir = fdp->fd_rdir;
193 ndp->ni_topdir = fdp->fd_jdir;
194
195 dp = fdp->fd_cdir;
196 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
197 VREF(dp);
198 FILEDESC_SUNLOCK(fdp);
199 for (;;) {
200 /*
201 * Check if root directory should replace current directory.
202 * Done at start of translation and after symbolic link.
203 */
204 cnp->cn_nameptr = cnp->cn_pnbuf;
205 if (*(cnp->cn_nameptr) == '/') {
206 vrele(dp);
207 VFS_UNLOCK_GIANT(vfslocked);
208 while (*(cnp->cn_nameptr) == '/') {
209 cnp->cn_nameptr++;
210 ndp->ni_pathlen--;
211 }
212 dp = ndp->ni_rootdir;
213 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
214 VREF(dp);
215 }
216 if (vfslocked)
217 ndp->ni_cnd.cn_flags |= GIANTHELD;
218 ndp->ni_startdir = dp;
219 error = lookup(ndp);
220 if (error) {
221 uma_zfree(namei_zone, cnp->cn_pnbuf);
222 #ifdef DIAGNOSTIC
223 cnp->cn_pnbuf = NULL;
224 cnp->cn_nameptr = NULL;
225 #endif
226 return (error);
227 }
228 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
229 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
230 /*
231 * Check for symbolic link
232 */
233 if ((cnp->cn_flags & ISSYMLINK) == 0) {
234 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
235 uma_zfree(namei_zone, cnp->cn_pnbuf);
236 #ifdef DIAGNOSTIC
237 cnp->cn_pnbuf = NULL;
238 cnp->cn_nameptr = NULL;
239 #endif
240 } else
241 cnp->cn_flags |= HASBUF;
242
243 if ((cnp->cn_flags & MPSAFE) == 0) {
244 VFS_UNLOCK_GIANT(vfslocked);
245 } else if (vfslocked)
246 ndp->ni_cnd.cn_flags |= GIANTHELD;
247 return (0);
248 }
249 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
250 error = ELOOP;
251 break;
252 }
253 #ifdef MAC
254 if ((cnp->cn_flags & NOMACCHECK) == 0) {
255 error = mac_check_vnode_readlink(td->td_ucred,
256 ndp->ni_vp);
257 if (error)
258 break;
259 }
260 #endif
261 if (ndp->ni_pathlen > 1)
262 cp = uma_zalloc(namei_zone, M_WAITOK);
263 else
264 cp = cnp->cn_pnbuf;
265 aiov.iov_base = cp;
266 aiov.iov_len = MAXPATHLEN;
267 auio.uio_iov = &aiov;
268 auio.uio_iovcnt = 1;
269 auio.uio_offset = 0;
270 auio.uio_rw = UIO_READ;
271 auio.uio_segflg = UIO_SYSSPACE;
272 auio.uio_td = (struct thread *)0;
273 auio.uio_resid = MAXPATHLEN;
274 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
275 if (error) {
276 if (ndp->ni_pathlen > 1)
277 uma_zfree(namei_zone, cp);
278 break;
279 }
280 linklen = MAXPATHLEN - auio.uio_resid;
281 if (linklen == 0) {
282 if (ndp->ni_pathlen > 1)
283 uma_zfree(namei_zone, cp);
284 error = ENOENT;
285 break;
286 }
287 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
288 if (ndp->ni_pathlen > 1)
289 uma_zfree(namei_zone, cp);
290 error = ENAMETOOLONG;
291 break;
292 }
293 if (ndp->ni_pathlen > 1) {
294 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
295 uma_zfree(namei_zone, cnp->cn_pnbuf);
296 cnp->cn_pnbuf = cp;
297 } else
298 cnp->cn_pnbuf[linklen] = '\0';
299 ndp->ni_pathlen += linklen;
300 vput(ndp->ni_vp);
301 dp = ndp->ni_dvp;
302 }
303 uma_zfree(namei_zone, cnp->cn_pnbuf);
304 #ifdef DIAGNOSTIC
305 cnp->cn_pnbuf = NULL;
306 cnp->cn_nameptr = NULL;
307 #endif
308 vput(ndp->ni_vp);
309 ndp->ni_vp = NULL;
310 vrele(ndp->ni_dvp);
311 VFS_UNLOCK_GIANT(vfslocked);
312 return (error);
313 }
314
315 static int
316 compute_cn_lkflags(struct mount *mp, int lkflags)
317 {
318 if (mp == NULL ||
319 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
320 lkflags &= ~LK_SHARED;
321 lkflags |= LK_EXCLUSIVE;
322 }
323 return lkflags;
324 }
325
326 /*
327 * Search a pathname.
328 * This is a very central and rather complicated routine.
329 *
330 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
331 * The starting directory is taken from ni_startdir. The pathname is
332 * descended until done, or a symbolic link is encountered. The variable
333 * ni_more is clear if the path is completed; it is set to one if a
334 * symbolic link needing interpretation is encountered.
335 *
336 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
337 * whether the name is to be looked up, created, renamed, or deleted.
338 * When CREATE, RENAME, or DELETE is specified, information usable in
339 * creating, renaming, or deleting a directory entry may be calculated.
340 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
341 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
342 * returned unlocked. Otherwise the parent directory is not returned. If
343 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
344 * the target is returned locked, otherwise it is returned unlocked.
345 * When creating or renaming and LOCKPARENT is specified, the target may not
346 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
347 *
348 * Overall outline of lookup:
349 *
350 * dirloop:
351 * identify next component of name at ndp->ni_ptr
352 * handle degenerate case where name is null string
353 * if .. and crossing mount points and on mounted filesys, find parent
354 * call VOP_LOOKUP routine for next component name
355 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
356 * component vnode returned in ni_vp (if it exists), locked.
357 * if result vnode is mounted on and crossing mount points,
358 * find mounted on vnode
359 * if more components of name, do next level at dirloop
360 * return the answer in ni_vp, locked if LOCKLEAF set
361 * if LOCKPARENT set, return locked parent in ni_dvp
362 * if WANTPARENT set, return unlocked parent in ni_dvp
363 */
364 int
365 lookup(struct nameidata *ndp)
366 {
367 char *cp; /* pointer into pathname argument */
368 struct vnode *dp = 0; /* the directory we are searching */
369 struct vnode *tdp; /* saved dp */
370 struct mount *mp; /* mount table entry */
371 int docache; /* == 0 do not cache last component */
372 int wantparent; /* 1 => wantparent or lockparent flag */
373 int rdonly; /* lookup read-only flag bit */
374 int trailing_slash;
375 int error = 0;
376 int dpunlocked = 0; /* dp has already been unlocked */
377 struct componentname *cnp = &ndp->ni_cnd;
378 struct thread *td = cnp->cn_thread;
379 int vfslocked; /* VFS Giant state for child */
380 int dvfslocked; /* VFS Giant state for parent */
381 int tvfslocked;
382 int lkflags_save;
383
384 /*
385 * Setup: break out flag bits into variables.
386 */
387 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
388 vfslocked = 0;
389 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
390 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
391 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
392 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
393 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
394 if (cnp->cn_nameiop == DELETE ||
395 (wantparent && cnp->cn_nameiop != CREATE &&
396 cnp->cn_nameiop != LOOKUP))
397 docache = 0;
398 rdonly = cnp->cn_flags & RDONLY;
399 cnp->cn_flags &= ~ISSYMLINK;
400 ndp->ni_dvp = NULL;
401 /*
402 * We use shared locks until we hit the parent of the last cn then
403 * we adjust based on the requesting flags.
404 */
405 if (lookup_shared)
406 cnp->cn_lkflags = LK_SHARED;
407 else
408 cnp->cn_lkflags = LK_EXCLUSIVE;
409 dp = ndp->ni_startdir;
410 ndp->ni_startdir = NULLVP;
411 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
412
413 dirloop:
414 /*
415 * Search a new directory.
416 *
417 * The last component of the filename is left accessible via
418 * cnp->cn_nameptr for callers that need the name. Callers needing
419 * the name set the SAVENAME flag. When done, they assume
420 * responsibility for freeing the pathname buffer.
421 */
422 cnp->cn_consume = 0;
423 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
424 continue;
425 cnp->cn_namelen = cp - cnp->cn_nameptr;
426 if (cnp->cn_namelen > NAME_MAX) {
427 error = ENAMETOOLONG;
428 goto bad;
429 }
430 #ifdef NAMEI_DIAGNOSTIC
431 { char c = *cp;
432 *cp = '\0';
433 printf("{%s}: ", cnp->cn_nameptr);
434 *cp = c; }
435 #endif
436 ndp->ni_pathlen -= cnp->cn_namelen;
437 ndp->ni_next = cp;
438
439 /*
440 * Replace multiple slashes by a single slash and trailing slashes
441 * by a null. This must be done before VOP_LOOKUP() because some
442 * fs's don't know about trailing slashes. Remember if there were
443 * trailing slashes to handle symlinks, existing non-directories
444 * and non-existing files that won't be directories specially later.
445 */
446 trailing_slash = 0;
447 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
448 cp++;
449 ndp->ni_pathlen--;
450 if (*cp == '\0') {
451 trailing_slash = 1;
452 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
453 }
454 }
455 ndp->ni_next = cp;
456
457 cnp->cn_flags |= MAKEENTRY;
458 if (*cp == '\0' && docache == 0)
459 cnp->cn_flags &= ~MAKEENTRY;
460 if (cnp->cn_namelen == 2 &&
461 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
462 cnp->cn_flags |= ISDOTDOT;
463 else
464 cnp->cn_flags &= ~ISDOTDOT;
465 if (*ndp->ni_next == 0)
466 cnp->cn_flags |= ISLASTCN;
467 else
468 cnp->cn_flags &= ~ISLASTCN;
469
470
471 /*
472 * Check for degenerate name (e.g. / or "")
473 * which is a way of talking about a directory,
474 * e.g. like "/." or ".".
475 */
476 if (cnp->cn_nameptr[0] == '\0') {
477 if (dp->v_type != VDIR) {
478 error = ENOTDIR;
479 goto bad;
480 }
481 if (cnp->cn_nameiop != LOOKUP) {
482 error = EISDIR;
483 goto bad;
484 }
485 if (wantparent) {
486 ndp->ni_dvp = dp;
487 VREF(dp);
488 }
489 ndp->ni_vp = dp;
490
491 if (cnp->cn_flags & AUDITVNODE1)
492 AUDIT_ARG(vnode, dp, ARG_VNODE1);
493 else if (cnp->cn_flags & AUDITVNODE2)
494 AUDIT_ARG(vnode, dp, ARG_VNODE2);
495
496 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
497 VOP_UNLOCK(dp, 0, td);
498 /* XXX This should probably move to the top of function. */
499 if (cnp->cn_flags & SAVESTART)
500 panic("lookup: SAVESTART");
501 goto success;
502 }
503
504 /*
505 * Handle "..": four special cases.
506 * 1. Return an error if this is the last component of
507 * the name and the operation is DELETE or RENAME.
508 * 2. If at root directory (e.g. after chroot)
509 * or at absolute root directory
510 * then ignore it so can't get out.
511 * 3. If this vnode is the root of a mounted
512 * filesystem, then replace it with the
513 * vnode which was mounted on so we take the
514 * .. in the other filesystem.
515 * 4. If the vnode is the top directory of
516 * the jail or chroot, don't let them out.
517 */
518 if (cnp->cn_flags & ISDOTDOT) {
519 if ((cnp->cn_flags & ISLASTCN) != 0 &&
520 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
521 error = EINVAL;
522 goto bad;
523 }
524 for (;;) {
525 if (dp == ndp->ni_rootdir ||
526 dp == ndp->ni_topdir ||
527 dp == rootvnode ||
528 ((dp->v_vflag & VV_ROOT) != 0 &&
529 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
530 ndp->ni_dvp = dp;
531 ndp->ni_vp = dp;
532 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
533 VREF(dp);
534 goto nextname;
535 }
536 if ((dp->v_vflag & VV_ROOT) == 0)
537 break;
538 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
539 error = EBADF;
540 goto bad;
541 }
542 tdp = dp;
543 dp = dp->v_mount->mnt_vnodecovered;
544 tvfslocked = dvfslocked;
545 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
546 VREF(dp);
547 vput(tdp);
548 VFS_UNLOCK_GIANT(tvfslocked);
549 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
550 }
551 }
552
553 /*
554 * We now have a segment name to search for, and a directory to search.
555 */
556 unionlookup:
557 #ifdef MAC
558 if ((cnp->cn_flags & NOMACCHECK) == 0) {
559 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
560 if (error)
561 goto bad;
562 }
563 #endif
564 ndp->ni_dvp = dp;
565 ndp->ni_vp = NULL;
566 ASSERT_VOP_LOCKED(dp, "lookup");
567 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
568 /*
569 * If we have a shared lock we may need to upgrade the lock for the
570 * last operation.
571 */
572 if (dp != vp_crossmp &&
573 VOP_ISLOCKED(dp, td) == LK_SHARED &&
574 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
575 vn_lock(dp, LK_UPGRADE|LK_RETRY, td);
576 /*
577 * If we're looking up the last component and we need an exclusive
578 * lock, adjust our lkflags.
579 */
580 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
581 (ISLASTCN|LOCKLEAF))
582 cnp->cn_lkflags = LK_EXCLUSIVE;
583 #ifdef NAMEI_DIAGNOSTIC
584 vprint("lookup in", dp);
585 #endif
586 lkflags_save = cnp->cn_lkflags;
587 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
588 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
589 cnp->cn_lkflags = lkflags_save;
590 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
591 #ifdef NAMEI_DIAGNOSTIC
592 printf("not found\n");
593 #endif
594 if ((error == ENOENT) &&
595 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
596 (dp->v_mount->mnt_flag & MNT_UNION)) {
597 tdp = dp;
598 dp = dp->v_mount->mnt_vnodecovered;
599 tvfslocked = dvfslocked;
600 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
601 VREF(dp);
602 vput(tdp);
603 VFS_UNLOCK_GIANT(tvfslocked);
604 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
605 goto unionlookup;
606 }
607
608 if (error != EJUSTRETURN)
609 goto bad;
610 /*
611 * If creating and at end of pathname, then can consider
612 * allowing file to be created.
613 */
614 if (rdonly) {
615 error = EROFS;
616 goto bad;
617 }
618 if (*cp == '\0' && trailing_slash &&
619 !(cnp->cn_flags & WILLBEDIR)) {
620 error = ENOENT;
621 goto bad;
622 }
623 if ((cnp->cn_flags & LOCKPARENT) == 0)
624 VOP_UNLOCK(dp, 0, td);
625 /*
626 * This is a temporary assert to make sure I know what the
627 * behavior here was.
628 */
629 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
630 ("lookup: Unhandled case."));
631 /*
632 * We return with ni_vp NULL to indicate that the entry
633 * doesn't currently exist, leaving a pointer to the
634 * (possibly locked) directory vnode in ndp->ni_dvp.
635 */
636 if (cnp->cn_flags & SAVESTART) {
637 ndp->ni_startdir = ndp->ni_dvp;
638 VREF(ndp->ni_startdir);
639 }
640 goto success;
641 } else
642 cnp->cn_lkflags = lkflags_save;
643 #ifdef NAMEI_DIAGNOSTIC
644 printf("found\n");
645 #endif
646 /*
647 * Take into account any additional components consumed by
648 * the underlying filesystem.
649 */
650 if (cnp->cn_consume > 0) {
651 cnp->cn_nameptr += cnp->cn_consume;
652 ndp->ni_next += cnp->cn_consume;
653 ndp->ni_pathlen -= cnp->cn_consume;
654 cnp->cn_consume = 0;
655 }
656
657 dp = ndp->ni_vp;
658 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
659
660 /*
661 * Check to see if the vnode has been mounted on;
662 * if so find the root of the mounted filesystem.
663 */
664 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
665 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
666 if (vfs_busy(mp, 0, 0, td))
667 continue;
668 vput(dp);
669 VFS_UNLOCK_GIANT(vfslocked);
670 vfslocked = VFS_LOCK_GIANT(mp);
671 if (dp != ndp->ni_dvp)
672 vput(ndp->ni_dvp);
673 else
674 vrele(ndp->ni_dvp);
675 VFS_UNLOCK_GIANT(dvfslocked);
676 dvfslocked = 0;
677 vref(vp_crossmp);
678 ndp->ni_dvp = vp_crossmp;
679 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags), &tdp, td);
680 vfs_unbusy(mp, td);
681 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT, td))
682 panic("vp_crossmp exclusively locked or reclaimed");
683 if (error) {
684 dpunlocked = 1;
685 goto bad2;
686 }
687 ndp->ni_vp = dp = tdp;
688 }
689
690 /*
691 * Check for symbolic link
692 */
693 if ((dp->v_type == VLNK) &&
694 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
695 *ndp->ni_next == '/')) {
696 cnp->cn_flags |= ISSYMLINK;
697 if (dp->v_iflag & VI_DOOMED) {
698 /* We can't know whether the directory was mounted with
699 * NOSYMFOLLOW, so we can't follow safely. */
700 error = EBADF;
701 goto bad2;
702 }
703 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
704 error = EACCES;
705 goto bad2;
706 }
707 /*
708 * Symlink code always expects an unlocked dvp.
709 */
710 if (ndp->ni_dvp != ndp->ni_vp)
711 VOP_UNLOCK(ndp->ni_dvp, 0, td);
712 goto success;
713 }
714
715 /*
716 * Check for bogus trailing slashes.
717 */
718 if (trailing_slash && dp->v_type != VDIR) {
719 error = ENOTDIR;
720 goto bad2;
721 }
722
723 nextname:
724 /*
725 * Not a symbolic link. If more pathname,
726 * continue at next component, else return.
727 */
728 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
729 ("lookup: invalid path state."));
730 if (*ndp->ni_next == '/') {
731 cnp->cn_nameptr = ndp->ni_next;
732 while (*cnp->cn_nameptr == '/') {
733 cnp->cn_nameptr++;
734 ndp->ni_pathlen--;
735 }
736 if (ndp->ni_dvp != dp)
737 vput(ndp->ni_dvp);
738 else
739 vrele(ndp->ni_dvp);
740 VFS_UNLOCK_GIANT(dvfslocked);
741 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
742 vfslocked = 0;
743 goto dirloop;
744 }
745 /*
746 * Disallow directory write attempts on read-only filesystems.
747 */
748 if (rdonly &&
749 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
750 error = EROFS;
751 goto bad2;
752 }
753 if (cnp->cn_flags & SAVESTART) {
754 ndp->ni_startdir = ndp->ni_dvp;
755 VREF(ndp->ni_startdir);
756 }
757 if (!wantparent) {
758 if (ndp->ni_dvp != dp)
759 vput(ndp->ni_dvp);
760 else
761 vrele(ndp->ni_dvp);
762 VFS_UNLOCK_GIANT(dvfslocked);
763 dvfslocked = 0;
764 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
765 VOP_UNLOCK(ndp->ni_dvp, 0, td);
766
767 if (cnp->cn_flags & AUDITVNODE1)
768 AUDIT_ARG(vnode, dp, ARG_VNODE1);
769 else if (cnp->cn_flags & AUDITVNODE2)
770 AUDIT_ARG(vnode, dp, ARG_VNODE2);
771
772 if ((cnp->cn_flags & LOCKLEAF) == 0)
773 VOP_UNLOCK(dp, 0, td);
774 success:
775 /*
776 * Because of lookup_shared we may have the vnode shared locked, but
777 * the caller may want it to be exclusively locked.
778 */
779 if ((cnp->cn_flags & (ISLASTCN | LOCKSHARED | LOCKLEAF)) ==
780 (ISLASTCN | LOCKLEAF) && VOP_ISLOCKED(dp, td) != LK_EXCLUSIVE) {
781 vn_lock(dp, LK_UPGRADE | LK_RETRY, td);
782 }
783 if (vfslocked && dvfslocked)
784 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
785 if (vfslocked || dvfslocked)
786 ndp->ni_cnd.cn_flags |= GIANTHELD;
787 return (0);
788
789 bad2:
790 if (dp != ndp->ni_dvp)
791 vput(ndp->ni_dvp);
792 else
793 vrele(ndp->ni_dvp);
794 bad:
795 if (!dpunlocked)
796 vput(dp);
797 VFS_UNLOCK_GIANT(vfslocked);
798 VFS_UNLOCK_GIANT(dvfslocked);
799 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
800 ndp->ni_vp = NULL;
801 return (error);
802 }
803
804 /*
805 * relookup - lookup a path name component
806 * Used by lookup to re-acquire things.
807 */
808 int
809 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
810 {
811 struct thread *td = cnp->cn_thread;
812 struct vnode *dp = 0; /* the directory we are searching */
813 int wantparent; /* 1 => wantparent or lockparent flag */
814 int rdonly; /* lookup read-only flag bit */
815 int error = 0;
816
817 KASSERT(cnp->cn_flags & ISLASTCN,
818 ("relookup: Not given last component."));
819 /*
820 * Setup: break out flag bits into variables.
821 */
822 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
823 KASSERT(wantparent, ("relookup: parent not wanted."));
824 rdonly = cnp->cn_flags & RDONLY;
825 cnp->cn_flags &= ~ISSYMLINK;
826 dp = dvp;
827 cnp->cn_lkflags = LK_EXCLUSIVE;
828 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
829
830 /*
831 * Search a new directory.
832 *
833 * The last component of the filename is left accessible via
834 * cnp->cn_nameptr for callers that need the name. Callers needing
835 * the name set the SAVENAME flag. When done, they assume
836 * responsibility for freeing the pathname buffer.
837 */
838 #ifdef NAMEI_DIAGNOSTIC
839 printf("{%s}: ", cnp->cn_nameptr);
840 #endif
841
842 /*
843 * Check for degenerate name (e.g. / or "")
844 * which is a way of talking about a directory,
845 * e.g. like "/." or ".".
846 */
847 if (cnp->cn_nameptr[0] == '\0') {
848 if (cnp->cn_nameiop != LOOKUP || wantparent) {
849 error = EISDIR;
850 goto bad;
851 }
852 if (dp->v_type != VDIR) {
853 error = ENOTDIR;
854 goto bad;
855 }
856 if (!(cnp->cn_flags & LOCKLEAF))
857 VOP_UNLOCK(dp, 0, td);
858 *vpp = dp;
859 /* XXX This should probably move to the top of function. */
860 if (cnp->cn_flags & SAVESTART)
861 panic("lookup: SAVESTART");
862 return (0);
863 }
864
865 if (cnp->cn_flags & ISDOTDOT)
866 panic ("relookup: lookup on dot-dot");
867
868 /*
869 * We now have a segment name to search for, and a directory to search.
870 */
871 #ifdef NAMEI_DIAGNOSTIC
872 vprint("search in:", dp);
873 #endif
874 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
875 KASSERT(*vpp == NULL, ("leaf should be empty"));
876 if (error != EJUSTRETURN)
877 goto bad;
878 /*
879 * If creating and at end of pathname, then can consider
880 * allowing file to be created.
881 */
882 if (rdonly) {
883 error = EROFS;
884 goto bad;
885 }
886 /* ASSERT(dvp == ndp->ni_startdir) */
887 if (cnp->cn_flags & SAVESTART)
888 VREF(dvp);
889 if ((cnp->cn_flags & LOCKPARENT) == 0)
890 VOP_UNLOCK(dp, 0, td);
891 /*
892 * This is a temporary assert to make sure I know what the
893 * behavior here was.
894 */
895 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
896 ("relookup: Unhandled case."));
897 /*
898 * We return with ni_vp NULL to indicate that the entry
899 * doesn't currently exist, leaving a pointer to the
900 * (possibly locked) directory vnode in ndp->ni_dvp.
901 */
902 return (0);
903 }
904
905 dp = *vpp;
906
907 /*
908 * Disallow directory write attempts on read-only filesystems.
909 */
910 if (rdonly &&
911 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
912 if (dvp == dp)
913 vrele(dvp);
914 else
915 vput(dvp);
916 error = EROFS;
917 goto bad;
918 }
919 /*
920 * Set the parent lock/ref state to the requested state.
921 */
922 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
923 if (wantparent)
924 VOP_UNLOCK(dvp, 0, td);
925 else
926 vput(dvp);
927 } else if (!wantparent)
928 vrele(dvp);
929 /*
930 * Check for symbolic link
931 */
932 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
933 ("relookup: symlink found.\n"));
934
935 /* ASSERT(dvp == ndp->ni_startdir) */
936 if (cnp->cn_flags & SAVESTART)
937 VREF(dvp);
938
939 if ((cnp->cn_flags & LOCKLEAF) == 0)
940 VOP_UNLOCK(dp, 0, td);
941 return (0);
942 bad:
943 vput(dp);
944 *vpp = NULL;
945 return (error);
946 }
947
948 /*
949 * Free data allocated by namei(); see namei(9) for details.
950 */
951 void
952 NDFREE(struct nameidata *ndp, const u_int flags)
953 {
954 int unlock_dvp;
955 int unlock_vp;
956
957 unlock_dvp = 0;
958 unlock_vp = 0;
959
960 if (!(flags & NDF_NO_FREE_PNBUF) &&
961 (ndp->ni_cnd.cn_flags & HASBUF)) {
962 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
963 ndp->ni_cnd.cn_flags &= ~HASBUF;
964 }
965 if (!(flags & NDF_NO_VP_UNLOCK) &&
966 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
967 unlock_vp = 1;
968 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
969 if (unlock_vp) {
970 vput(ndp->ni_vp);
971 unlock_vp = 0;
972 } else
973 vrele(ndp->ni_vp);
974 ndp->ni_vp = NULL;
975 }
976 if (unlock_vp)
977 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
978 if (!(flags & NDF_NO_DVP_UNLOCK) &&
979 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
980 ndp->ni_dvp != ndp->ni_vp)
981 unlock_dvp = 1;
982 if (!(flags & NDF_NO_DVP_RELE) &&
983 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
984 if (unlock_dvp) {
985 vput(ndp->ni_dvp);
986 unlock_dvp = 0;
987 } else
988 vrele(ndp->ni_dvp);
989 ndp->ni_dvp = NULL;
990 }
991 if (unlock_dvp)
992 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
993 if (!(flags & NDF_NO_STARTDIR_RELE) &&
994 (ndp->ni_cnd.cn_flags & SAVESTART)) {
995 vrele(ndp->ni_startdir);
996 ndp->ni_startdir = NULL;
997 }
998 }
999
1000 /*
1001 * Determine if there is a suitable alternate filename under the specified
1002 * prefix for the specified path. If the create flag is set, then the
1003 * alternate prefix will be used so long as the parent directory exists.
1004 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1005 * files under /compat/linux for example. The chosen path (whether under
1006 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1007 * to by pathbuf. The caller is responsible for free'ing the buffer from
1008 * the M_TEMP bucket if one is returned.
1009 */
1010 int
1011 kern_alternate_path(struct thread *td, const char *prefix, char *path,
1012 enum uio_seg pathseg, char **pathbuf, int create)
1013 {
1014 struct nameidata nd, ndroot;
1015 char *ptr, *buf, *cp;
1016 size_t len, sz;
1017 int error;
1018
1019 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1020 *pathbuf = buf;
1021
1022 /* Copy the prefix into the new pathname as a starting point. */
1023 len = strlcpy(buf, prefix, MAXPATHLEN);
1024 if (len >= MAXPATHLEN) {
1025 *pathbuf = NULL;
1026 free(buf, M_TEMP);
1027 return (EINVAL);
1028 }
1029 sz = MAXPATHLEN - len;
1030 ptr = buf + len;
1031
1032 /* Append the filename to the prefix. */
1033 if (pathseg == UIO_SYSSPACE)
1034 error = copystr(path, ptr, sz, &len);
1035 else
1036 error = copyinstr(path, ptr, sz, &len);
1037
1038 if (error) {
1039 *pathbuf = NULL;
1040 free(buf, M_TEMP);
1041 return (error);
1042 }
1043
1044 /* Only use a prefix with absolute pathnames. */
1045 if (*ptr != '/') {
1046 error = EINVAL;
1047 goto keeporig;
1048 }
1049
1050 /*
1051 * We know that there is a / somewhere in this pathname.
1052 * Search backwards for it, to find the file's parent dir
1053 * to see if it exists in the alternate tree. If it does,
1054 * and we want to create a file (cflag is set). We don't
1055 * need to worry about the root comparison in this case.
1056 */
1057
1058 if (create) {
1059 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1060 *cp = '\0';
1061
1062 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1063 error = namei(&nd);
1064 *cp = '/';
1065 if (error != 0)
1066 goto keeporig;
1067 } else {
1068 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1069
1070 error = namei(&nd);
1071 if (error != 0)
1072 goto keeporig;
1073
1074 /*
1075 * We now compare the vnode of the prefix to the one
1076 * vnode asked. If they resolve to be the same, then we
1077 * ignore the match so that the real root gets used.
1078 * This avoids the problem of traversing "../.." to find the
1079 * root directory and never finding it, because "/" resolves
1080 * to the emulation root directory. This is expensive :-(
1081 */
1082 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1083 td);
1084
1085 /* We shouldn't ever get an error from this namei(). */
1086 error = namei(&ndroot);
1087 if (error == 0) {
1088 if (nd.ni_vp == ndroot.ni_vp)
1089 error = ENOENT;
1090
1091 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1092 vrele(ndroot.ni_vp);
1093 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1094 }
1095 }
1096
1097 NDFREE(&nd, NDF_ONLY_PNBUF);
1098 vrele(nd.ni_vp);
1099 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1100
1101 keeporig:
1102 /* If there was an error, use the original path name. */
1103 if (error)
1104 bcopy(ptr, buf, len);
1105 return (error);
1106 }
Cache object: bc6dc2dc718c070c70e080550e80a9b1
|