FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_lookup.c
1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_ktrace.h"
41 #include "opt_mac.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/namei.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 #include <sys/filedesc.h>
52 #include <sys/proc.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysctl.h>
55 #ifdef KTRACE
56 #include <sys/ktrace.h>
57 #endif
58
59 #include <security/audit/audit.h>
60 #include <security/mac/mac_framework.h>
61
62 #include <vm/uma.h>
63
64 #define NAMEI_DIAGNOSTIC 1
65 #undef NAMEI_DIAGNOSTIC
66
67 /*
68 * Allocation zone for namei
69 */
70 uma_zone_t namei_zone;
71 /*
72 * Placeholder vnode for mp traversal
73 */
74 static struct vnode *vp_crossmp;
75
76 static void
77 nameiinit(void *dummy __unused)
78 {
79 int error;
80
81 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
82 UMA_ALIGN_PTR, 0);
83 error = getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
84 if (error != 0)
85 panic("nameiinit: getnewvnode");
86 vp_crossmp->v_vnlock->lk_flags &= ~LK_NOSHARE;
87 }
88 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
89
90 static int lookup_shared = 0;
91 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
92 "Enables/Disables shared locks for path name translation");
93 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
94
95 /*
96 * Convert a pathname into a pointer to a locked vnode.
97 *
98 * The FOLLOW flag is set when symbolic links are to be followed
99 * when they occur at the end of the name translation process.
100 * Symbolic links are always followed for all other pathname
101 * components other than the last.
102 *
103 * The segflg defines whether the name is to be copied from user
104 * space or kernel space.
105 *
106 * Overall outline of namei:
107 *
108 * copy in name
109 * get starting directory
110 * while (!done && !error) {
111 * call lookup to search path.
112 * if symbolic link, massage name in buffer and continue
113 * }
114 */
115 int
116 namei(struct nameidata *ndp)
117 {
118 struct filedesc *fdp; /* pointer to file descriptor state */
119 char *cp; /* pointer into pathname argument */
120 struct vnode *dp; /* the directory we are searching */
121 struct iovec aiov; /* uio for reading symbolic links */
122 struct uio auio;
123 int error, linklen;
124 struct componentname *cnp = &ndp->ni_cnd;
125 struct thread *td = cnp->cn_thread;
126 struct proc *p = td->td_proc;
127 int vfslocked;
128
129 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
130 ("NOT MPSAFE and Giant not held"));
131 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
132 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
133 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
134 ("namei: nameiop contaminated with flags"));
135 KASSERT((cnp->cn_flags & OPMASK) == 0,
136 ("namei: flags contaminated with nameiops"));
137 if (!lookup_shared)
138 cnp->cn_flags &= ~LOCKSHARED;
139 fdp = p->p_fd;
140
141 /*
142 * Get a buffer for the name to be translated, and copy the
143 * name into the buffer.
144 */
145 if ((cnp->cn_flags & HASBUF) == 0)
146 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
147 if (ndp->ni_segflg == UIO_SYSSPACE)
148 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
149 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
150 else
151 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
152 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
153
154 /* If we are auditing the kernel pathname, save the user pathname. */
155 if (cnp->cn_flags & AUDITVNODE1)
156 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH1);
157 if (cnp->cn_flags & AUDITVNODE2)
158 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH2);
159
160 /*
161 * Don't allow empty pathnames.
162 */
163 if (!error && *cnp->cn_pnbuf == '\0')
164 error = ENOENT;
165
166 if (error) {
167 uma_zfree(namei_zone, cnp->cn_pnbuf);
168 #ifdef DIAGNOSTIC
169 cnp->cn_pnbuf = NULL;
170 cnp->cn_nameptr = NULL;
171 #endif
172 ndp->ni_vp = NULL;
173 return (error);
174 }
175 ndp->ni_loopcnt = 0;
176 #ifdef KTRACE
177 if (KTRPOINT(td, KTR_NAMEI)) {
178 KASSERT(cnp->cn_thread == curthread,
179 ("namei not using curthread"));
180 ktrnamei(cnp->cn_pnbuf);
181 }
182 #endif
183
184 /*
185 * Get starting point for the translation.
186 */
187 FILEDESC_SLOCK(fdp);
188 ndp->ni_rootdir = fdp->fd_rdir;
189 ndp->ni_topdir = fdp->fd_jdir;
190
191 dp = fdp->fd_cdir;
192 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
193 VREF(dp);
194 FILEDESC_SUNLOCK(fdp);
195 for (;;) {
196 /*
197 * Check if root directory should replace current directory.
198 * Done at start of translation and after symbolic link.
199 */
200 cnp->cn_nameptr = cnp->cn_pnbuf;
201 if (*(cnp->cn_nameptr) == '/') {
202 vrele(dp);
203 VFS_UNLOCK_GIANT(vfslocked);
204 while (*(cnp->cn_nameptr) == '/') {
205 cnp->cn_nameptr++;
206 ndp->ni_pathlen--;
207 }
208 dp = ndp->ni_rootdir;
209 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
210 VREF(dp);
211 }
212 if (vfslocked)
213 ndp->ni_cnd.cn_flags |= GIANTHELD;
214 ndp->ni_startdir = dp;
215 error = lookup(ndp);
216 if (error) {
217 uma_zfree(namei_zone, cnp->cn_pnbuf);
218 #ifdef DIAGNOSTIC
219 cnp->cn_pnbuf = NULL;
220 cnp->cn_nameptr = NULL;
221 #endif
222 return (error);
223 }
224 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
225 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
226 /*
227 * Check for symbolic link
228 */
229 if ((cnp->cn_flags & ISSYMLINK) == 0) {
230 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
231 uma_zfree(namei_zone, cnp->cn_pnbuf);
232 #ifdef DIAGNOSTIC
233 cnp->cn_pnbuf = NULL;
234 cnp->cn_nameptr = NULL;
235 #endif
236 } else
237 cnp->cn_flags |= HASBUF;
238
239 if ((cnp->cn_flags & MPSAFE) == 0) {
240 VFS_UNLOCK_GIANT(vfslocked);
241 } else if (vfslocked)
242 ndp->ni_cnd.cn_flags |= GIANTHELD;
243 return (0);
244 }
245 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
246 error = ELOOP;
247 break;
248 }
249 #ifdef MAC
250 if ((cnp->cn_flags & NOMACCHECK) == 0) {
251 error = mac_check_vnode_readlink(td->td_ucred,
252 ndp->ni_vp);
253 if (error)
254 break;
255 }
256 #endif
257 if (ndp->ni_pathlen > 1)
258 cp = uma_zalloc(namei_zone, M_WAITOK);
259 else
260 cp = cnp->cn_pnbuf;
261 aiov.iov_base = cp;
262 aiov.iov_len = MAXPATHLEN;
263 auio.uio_iov = &aiov;
264 auio.uio_iovcnt = 1;
265 auio.uio_offset = 0;
266 auio.uio_rw = UIO_READ;
267 auio.uio_segflg = UIO_SYSSPACE;
268 auio.uio_td = (struct thread *)0;
269 auio.uio_resid = MAXPATHLEN;
270 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
271 if (error) {
272 if (ndp->ni_pathlen > 1)
273 uma_zfree(namei_zone, cp);
274 break;
275 }
276 linklen = MAXPATHLEN - auio.uio_resid;
277 if (linklen == 0) {
278 if (ndp->ni_pathlen > 1)
279 uma_zfree(namei_zone, cp);
280 error = ENOENT;
281 break;
282 }
283 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
284 if (ndp->ni_pathlen > 1)
285 uma_zfree(namei_zone, cp);
286 error = ENAMETOOLONG;
287 break;
288 }
289 if (ndp->ni_pathlen > 1) {
290 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
291 uma_zfree(namei_zone, cnp->cn_pnbuf);
292 cnp->cn_pnbuf = cp;
293 } else
294 cnp->cn_pnbuf[linklen] = '\0';
295 ndp->ni_pathlen += linklen;
296 vput(ndp->ni_vp);
297 dp = ndp->ni_dvp;
298 }
299 uma_zfree(namei_zone, cnp->cn_pnbuf);
300 #ifdef DIAGNOSTIC
301 cnp->cn_pnbuf = NULL;
302 cnp->cn_nameptr = NULL;
303 #endif
304 vput(ndp->ni_vp);
305 ndp->ni_vp = NULL;
306 vrele(ndp->ni_dvp);
307 VFS_UNLOCK_GIANT(vfslocked);
308 return (error);
309 }
310
311 static int
312 compute_cn_lkflags(struct mount *mp, int lkflags)
313 {
314 if (mp == NULL ||
315 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
316 lkflags &= ~LK_SHARED;
317 lkflags |= LK_EXCLUSIVE;
318 }
319 return lkflags;
320 }
321
322 /*
323 * Search a pathname.
324 * This is a very central and rather complicated routine.
325 *
326 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
327 * The starting directory is taken from ni_startdir. The pathname is
328 * descended until done, or a symbolic link is encountered. The variable
329 * ni_more is clear if the path is completed; it is set to one if a
330 * symbolic link needing interpretation is encountered.
331 *
332 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
333 * whether the name is to be looked up, created, renamed, or deleted.
334 * When CREATE, RENAME, or DELETE is specified, information usable in
335 * creating, renaming, or deleting a directory entry may be calculated.
336 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
337 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
338 * returned unlocked. Otherwise the parent directory is not returned. If
339 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
340 * the target is returned locked, otherwise it is returned unlocked.
341 * When creating or renaming and LOCKPARENT is specified, the target may not
342 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
343 *
344 * Overall outline of lookup:
345 *
346 * dirloop:
347 * identify next component of name at ndp->ni_ptr
348 * handle degenerate case where name is null string
349 * if .. and crossing mount points and on mounted filesys, find parent
350 * call VOP_LOOKUP routine for next component name
351 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
352 * component vnode returned in ni_vp (if it exists), locked.
353 * if result vnode is mounted on and crossing mount points,
354 * find mounted on vnode
355 * if more components of name, do next level at dirloop
356 * return the answer in ni_vp, locked if LOCKLEAF set
357 * if LOCKPARENT set, return locked parent in ni_dvp
358 * if WANTPARENT set, return unlocked parent in ni_dvp
359 */
360 int
361 lookup(struct nameidata *ndp)
362 {
363 char *cp; /* pointer into pathname argument */
364 struct vnode *dp = 0; /* the directory we are searching */
365 struct vnode *tdp; /* saved dp */
366 struct mount *mp; /* mount table entry */
367 int docache; /* == 0 do not cache last component */
368 int wantparent; /* 1 => wantparent or lockparent flag */
369 int rdonly; /* lookup read-only flag bit */
370 int trailing_slash;
371 int error = 0;
372 int dpunlocked = 0; /* dp has already been unlocked */
373 struct componentname *cnp = &ndp->ni_cnd;
374 struct thread *td = cnp->cn_thread;
375 int vfslocked; /* VFS Giant state for child */
376 int dvfslocked; /* VFS Giant state for parent */
377 int tvfslocked;
378 int lkflags_save;
379
380 /*
381 * Setup: break out flag bits into variables.
382 */
383 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
384 vfslocked = 0;
385 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
386 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
387 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
388 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
389 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
390 if (cnp->cn_nameiop == DELETE ||
391 (wantparent && cnp->cn_nameiop != CREATE &&
392 cnp->cn_nameiop != LOOKUP))
393 docache = 0;
394 rdonly = cnp->cn_flags & RDONLY;
395 cnp->cn_flags &= ~ISSYMLINK;
396 ndp->ni_dvp = NULL;
397 /*
398 * We use shared locks until we hit the parent of the last cn then
399 * we adjust based on the requesting flags.
400 */
401 if (lookup_shared)
402 cnp->cn_lkflags = LK_SHARED;
403 else
404 cnp->cn_lkflags = LK_EXCLUSIVE;
405 dp = ndp->ni_startdir;
406 ndp->ni_startdir = NULLVP;
407 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
408
409 dirloop:
410 /*
411 * Search a new directory.
412 *
413 * The last component of the filename is left accessible via
414 * cnp->cn_nameptr for callers that need the name. Callers needing
415 * the name set the SAVENAME flag. When done, they assume
416 * responsibility for freeing the pathname buffer.
417 */
418 cnp->cn_consume = 0;
419 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
420 continue;
421 cnp->cn_namelen = cp - cnp->cn_nameptr;
422 if (cnp->cn_namelen > NAME_MAX) {
423 error = ENAMETOOLONG;
424 goto bad;
425 }
426 #ifdef NAMEI_DIAGNOSTIC
427 { char c = *cp;
428 *cp = '\0';
429 printf("{%s}: ", cnp->cn_nameptr);
430 *cp = c; }
431 #endif
432 ndp->ni_pathlen -= cnp->cn_namelen;
433 ndp->ni_next = cp;
434
435 /*
436 * Replace multiple slashes by a single slash and trailing slashes
437 * by a null. This must be done before VOP_LOOKUP() because some
438 * fs's don't know about trailing slashes. Remember if there were
439 * trailing slashes to handle symlinks, existing non-directories
440 * and non-existing files that won't be directories specially later.
441 */
442 trailing_slash = 0;
443 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
444 cp++;
445 ndp->ni_pathlen--;
446 if (*cp == '\0') {
447 trailing_slash = 1;
448 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
449 }
450 }
451 ndp->ni_next = cp;
452
453 cnp->cn_flags |= MAKEENTRY;
454 if (*cp == '\0' && docache == 0)
455 cnp->cn_flags &= ~MAKEENTRY;
456 if (cnp->cn_namelen == 2 &&
457 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
458 cnp->cn_flags |= ISDOTDOT;
459 else
460 cnp->cn_flags &= ~ISDOTDOT;
461 if (*ndp->ni_next == 0)
462 cnp->cn_flags |= ISLASTCN;
463 else
464 cnp->cn_flags &= ~ISLASTCN;
465
466
467 /*
468 * Check for degenerate name (e.g. / or "")
469 * which is a way of talking about a directory,
470 * e.g. like "/." or ".".
471 */
472 if (cnp->cn_nameptr[0] == '\0') {
473 if (dp->v_type != VDIR) {
474 error = ENOTDIR;
475 goto bad;
476 }
477 if (cnp->cn_nameiop != LOOKUP) {
478 error = EISDIR;
479 goto bad;
480 }
481 if (wantparent) {
482 ndp->ni_dvp = dp;
483 VREF(dp);
484 }
485 ndp->ni_vp = dp;
486
487 if (cnp->cn_flags & AUDITVNODE1)
488 AUDIT_ARG(vnode, dp, ARG_VNODE1);
489 else if (cnp->cn_flags & AUDITVNODE2)
490 AUDIT_ARG(vnode, dp, ARG_VNODE2);
491
492 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
493 VOP_UNLOCK(dp, 0, td);
494 /* XXX This should probably move to the top of function. */
495 if (cnp->cn_flags & SAVESTART)
496 panic("lookup: SAVESTART");
497 goto success;
498 }
499
500 /*
501 * Handle "..": four special cases.
502 * 1. Return an error if this is the last component of
503 * the name and the operation is DELETE or RENAME.
504 * 2. If at root directory (e.g. after chroot)
505 * or at absolute root directory
506 * then ignore it so can't get out.
507 * 3. If this vnode is the root of a mounted
508 * filesystem, then replace it with the
509 * vnode which was mounted on so we take the
510 * .. in the other filesystem.
511 * 4. If the vnode is the top directory of
512 * the jail or chroot, don't let them out.
513 */
514 if (cnp->cn_flags & ISDOTDOT) {
515 if ((cnp->cn_flags & ISLASTCN) != 0 &&
516 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
517 error = EINVAL;
518 goto bad;
519 }
520 for (;;) {
521 if (dp == ndp->ni_rootdir ||
522 dp == ndp->ni_topdir ||
523 dp == rootvnode ||
524 ((dp->v_vflag & VV_ROOT) != 0 &&
525 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
526 ndp->ni_dvp = dp;
527 ndp->ni_vp = dp;
528 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
529 VREF(dp);
530 goto nextname;
531 }
532 if ((dp->v_vflag & VV_ROOT) == 0)
533 break;
534 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
535 error = EBADF;
536 goto bad;
537 }
538 tdp = dp;
539 dp = dp->v_mount->mnt_vnodecovered;
540 tvfslocked = dvfslocked;
541 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
542 VREF(dp);
543 vput(tdp);
544 VFS_UNLOCK_GIANT(tvfslocked);
545 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
546 }
547 }
548
549 /*
550 * We now have a segment name to search for, and a directory to search.
551 */
552 unionlookup:
553 #ifdef MAC
554 if ((cnp->cn_flags & NOMACCHECK) == 0) {
555 error = mac_check_vnode_lookup(td->td_ucred, dp, cnp);
556 if (error)
557 goto bad;
558 }
559 #endif
560 ndp->ni_dvp = dp;
561 ndp->ni_vp = NULL;
562 ASSERT_VOP_LOCKED(dp, "lookup");
563 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
564 /*
565 * If we have a shared lock we may need to upgrade the lock for the
566 * last operation.
567 */
568 if (dp != vp_crossmp &&
569 VOP_ISLOCKED(dp, td) == LK_SHARED &&
570 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
571 vn_lock(dp, LK_UPGRADE|LK_RETRY, td);
572 /*
573 * If we're looking up the last component and we need an exclusive
574 * lock, adjust our lkflags.
575 */
576 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
577 (ISLASTCN|LOCKLEAF))
578 cnp->cn_lkflags = LK_EXCLUSIVE;
579 #ifdef NAMEI_DIAGNOSTIC
580 vprint("lookup in", dp);
581 #endif
582 lkflags_save = cnp->cn_lkflags;
583 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
584 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
585 cnp->cn_lkflags = lkflags_save;
586 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
587 #ifdef NAMEI_DIAGNOSTIC
588 printf("not found\n");
589 #endif
590 if ((error == ENOENT) &&
591 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
592 (dp->v_mount->mnt_flag & MNT_UNION)) {
593 tdp = dp;
594 dp = dp->v_mount->mnt_vnodecovered;
595 tvfslocked = dvfslocked;
596 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
597 VREF(dp);
598 vput(tdp);
599 VFS_UNLOCK_GIANT(tvfslocked);
600 vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY), td);
601 goto unionlookup;
602 }
603
604 if (error != EJUSTRETURN)
605 goto bad;
606 /*
607 * If creating and at end of pathname, then can consider
608 * allowing file to be created.
609 */
610 if (rdonly) {
611 error = EROFS;
612 goto bad;
613 }
614 if (*cp == '\0' && trailing_slash &&
615 !(cnp->cn_flags & WILLBEDIR)) {
616 error = ENOENT;
617 goto bad;
618 }
619 if ((cnp->cn_flags & LOCKPARENT) == 0)
620 VOP_UNLOCK(dp, 0, td);
621 /*
622 * This is a temporary assert to make sure I know what the
623 * behavior here was.
624 */
625 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
626 ("lookup: Unhandled case."));
627 /*
628 * We return with ni_vp NULL to indicate that the entry
629 * doesn't currently exist, leaving a pointer to the
630 * (possibly locked) directory vnode in ndp->ni_dvp.
631 */
632 if (cnp->cn_flags & SAVESTART) {
633 ndp->ni_startdir = ndp->ni_dvp;
634 VREF(ndp->ni_startdir);
635 }
636 goto success;
637 } else
638 cnp->cn_lkflags = lkflags_save;
639 #ifdef NAMEI_DIAGNOSTIC
640 printf("found\n");
641 #endif
642 /*
643 * Take into account any additional components consumed by
644 * the underlying filesystem.
645 */
646 if (cnp->cn_consume > 0) {
647 cnp->cn_nameptr += cnp->cn_consume;
648 ndp->ni_next += cnp->cn_consume;
649 ndp->ni_pathlen -= cnp->cn_consume;
650 cnp->cn_consume = 0;
651 }
652
653 dp = ndp->ni_vp;
654 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
655
656 /*
657 * Check to see if the vnode has been mounted on;
658 * if so find the root of the mounted filesystem.
659 */
660 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
661 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
662 if (vfs_busy(mp, 0, 0, td))
663 continue;
664 vput(dp);
665 VFS_UNLOCK_GIANT(vfslocked);
666 vfslocked = VFS_LOCK_GIANT(mp);
667 if (dp != ndp->ni_dvp)
668 vput(ndp->ni_dvp);
669 else
670 vrele(ndp->ni_dvp);
671 VFS_UNLOCK_GIANT(dvfslocked);
672 dvfslocked = 0;
673 vref(vp_crossmp);
674 ndp->ni_dvp = vp_crossmp;
675 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags), &tdp, td);
676 vfs_unbusy(mp, td);
677 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT, td))
678 panic("vp_crossmp exclusively locked or reclaimed");
679 if (error) {
680 dpunlocked = 1;
681 goto bad2;
682 }
683 ndp->ni_vp = dp = tdp;
684 }
685
686 /*
687 * Check for symbolic link
688 */
689 if ((dp->v_type == VLNK) &&
690 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
691 *ndp->ni_next == '/')) {
692 cnp->cn_flags |= ISSYMLINK;
693 if (dp->v_iflag & VI_DOOMED) {
694 /* We can't know whether the directory was mounted with
695 * NOSYMFOLLOW, so we can't follow safely. */
696 error = EBADF;
697 goto bad2;
698 }
699 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
700 error = EACCES;
701 goto bad2;
702 }
703 /*
704 * Symlink code always expects an unlocked dvp.
705 */
706 if (ndp->ni_dvp != ndp->ni_vp)
707 VOP_UNLOCK(ndp->ni_dvp, 0, td);
708 goto success;
709 }
710
711 /*
712 * Check for bogus trailing slashes.
713 */
714 if (trailing_slash && dp->v_type != VDIR) {
715 error = ENOTDIR;
716 goto bad2;
717 }
718
719 nextname:
720 /*
721 * Not a symbolic link. If more pathname,
722 * continue at next component, else return.
723 */
724 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
725 ("lookup: invalid path state."));
726 if (*ndp->ni_next == '/') {
727 cnp->cn_nameptr = ndp->ni_next;
728 while (*cnp->cn_nameptr == '/') {
729 cnp->cn_nameptr++;
730 ndp->ni_pathlen--;
731 }
732 if (ndp->ni_dvp != dp)
733 vput(ndp->ni_dvp);
734 else
735 vrele(ndp->ni_dvp);
736 VFS_UNLOCK_GIANT(dvfslocked);
737 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
738 vfslocked = 0;
739 goto dirloop;
740 }
741 /*
742 * Disallow directory write attempts on read-only filesystems.
743 */
744 if (rdonly &&
745 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
746 error = EROFS;
747 goto bad2;
748 }
749 if (cnp->cn_flags & SAVESTART) {
750 ndp->ni_startdir = ndp->ni_dvp;
751 VREF(ndp->ni_startdir);
752 }
753 if (!wantparent) {
754 if (ndp->ni_dvp != dp)
755 vput(ndp->ni_dvp);
756 else
757 vrele(ndp->ni_dvp);
758 VFS_UNLOCK_GIANT(dvfslocked);
759 dvfslocked = 0;
760 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
761 VOP_UNLOCK(ndp->ni_dvp, 0, td);
762
763 if (cnp->cn_flags & AUDITVNODE1)
764 AUDIT_ARG(vnode, dp, ARG_VNODE1);
765 else if (cnp->cn_flags & AUDITVNODE2)
766 AUDIT_ARG(vnode, dp, ARG_VNODE2);
767
768 if ((cnp->cn_flags & LOCKLEAF) == 0)
769 VOP_UNLOCK(dp, 0, td);
770 success:
771 /*
772 * Because of lookup_shared we may have the vnode shared locked, but
773 * the caller may want it to be exclusively locked.
774 */
775 if ((cnp->cn_flags & (ISLASTCN | LOCKSHARED | LOCKLEAF)) ==
776 (ISLASTCN | LOCKLEAF) && VOP_ISLOCKED(dp, td) != LK_EXCLUSIVE) {
777 vn_lock(dp, LK_UPGRADE | LK_RETRY, td);
778 if (dp->v_iflag & VI_DOOMED) {
779 error = ENOENT;
780 goto bad2;
781 }
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: 31d418d95ed663315b3474dd2386f60a
|