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/8.0/sys/kern/vfs_lookup.c 201679 2010-01-06 21:45:30Z simon $");
39
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/lock.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/sdt.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #ifdef KTRACE
59 #include <sys/ktrace.h>
60 #endif
61
62 #include <security/audit/audit.h>
63 #include <security/mac/mac_framework.h>
64
65 #include <vm/uma.h>
66
67 #define NAMEI_DIAGNOSTIC 1
68 #undef NAMEI_DIAGNOSTIC
69
70 SDT_PROVIDER_DECLARE(vfs);
71 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
72 "unsigned long");
73 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
74
75 /*
76 * Allocation zone for namei
77 */
78 uma_zone_t namei_zone;
79 /*
80 * Placeholder vnode for mp traversal
81 */
82 static struct vnode *vp_crossmp;
83
84 static void
85 nameiinit(void *dummy __unused)
86 {
87 int error;
88
89 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
90 UMA_ALIGN_PTR, 0);
91 error = getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
92 if (error != 0)
93 panic("nameiinit: getnewvnode");
94 VN_LOCK_ASHARE(vp_crossmp);
95 }
96 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
97
98 static int lookup_shared = 1;
99 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
100 "Enables/Disables shared locks for path name translation");
101 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
102
103 /*
104 * Convert a pathname into a pointer to a locked vnode.
105 *
106 * The FOLLOW flag is set when symbolic links are to be followed
107 * when they occur at the end of the name translation process.
108 * Symbolic links are always followed for all other pathname
109 * components other than the last.
110 *
111 * The segflg defines whether the name is to be copied from user
112 * space or kernel space.
113 *
114 * Overall outline of namei:
115 *
116 * copy in name
117 * get starting directory
118 * while (!done && !error) {
119 * call lookup to search path.
120 * if symbolic link, massage name in buffer and continue
121 * }
122 */
123 int
124 namei(struct nameidata *ndp)
125 {
126 struct filedesc *fdp; /* pointer to file descriptor state */
127 char *cp; /* pointer into pathname argument */
128 struct vnode *dp; /* the directory we are searching */
129 struct iovec aiov; /* uio for reading symbolic links */
130 struct uio auio;
131 int error, linklen;
132 struct componentname *cnp = &ndp->ni_cnd;
133 struct thread *td = cnp->cn_thread;
134 struct proc *p = td->td_proc;
135 int vfslocked;
136
137 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
138 ("NOT MPSAFE and Giant not held"));
139 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
140 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
141 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
142 ("namei: nameiop contaminated with flags"));
143 KASSERT((cnp->cn_flags & OPMASK) == 0,
144 ("namei: flags contaminated with nameiops"));
145 if (!lookup_shared)
146 cnp->cn_flags &= ~LOCKSHARED;
147 fdp = p->p_fd;
148
149 /* We will set this ourselves if we need it. */
150 cnp->cn_flags &= ~TRAILINGSLASH;
151
152 /*
153 * Get a buffer for the name to be translated, and copy the
154 * name into the buffer.
155 */
156 if ((cnp->cn_flags & HASBUF) == 0)
157 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
158 if (ndp->ni_segflg == UIO_SYSSPACE)
159 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
160 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
161 else
162 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
163 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
164
165 /* If we are auditing the kernel pathname, save the user pathname. */
166 if (cnp->cn_flags & AUDITVNODE1)
167 AUDIT_ARG_UPATH1(td, cnp->cn_pnbuf);
168 if (cnp->cn_flags & AUDITVNODE2)
169 AUDIT_ARG_UPATH2(td, cnp->cn_pnbuf);
170
171 /*
172 * Don't allow empty pathnames.
173 */
174 if (!error && *cnp->cn_pnbuf == '\0')
175 error = ENOENT;
176
177 if (error) {
178 uma_zfree(namei_zone, cnp->cn_pnbuf);
179 #ifdef DIAGNOSTIC
180 cnp->cn_pnbuf = NULL;
181 cnp->cn_nameptr = NULL;
182 #endif
183 ndp->ni_vp = NULL;
184 return (error);
185 }
186 ndp->ni_loopcnt = 0;
187 #ifdef KTRACE
188 if (KTRPOINT(td, KTR_NAMEI)) {
189 KASSERT(cnp->cn_thread == curthread,
190 ("namei not using curthread"));
191 ktrnamei(cnp->cn_pnbuf);
192 }
193 #endif
194 /*
195 * Get starting point for the translation.
196 */
197 FILEDESC_SLOCK(fdp);
198 ndp->ni_rootdir = fdp->fd_rdir;
199 ndp->ni_topdir = fdp->fd_jdir;
200
201 dp = NULL;
202 if (cnp->cn_pnbuf[0] != '/') {
203 if (ndp->ni_startdir != NULL) {
204 dp = ndp->ni_startdir;
205 error = 0;
206 } else if (ndp->ni_dirfd != AT_FDCWD) {
207 if (cnp->cn_flags & AUDITVNODE1)
208 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
209 if (cnp->cn_flags & AUDITVNODE2)
210 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
211 error = fgetvp(td, ndp->ni_dirfd, &dp);
212 }
213 if (error != 0 || dp != NULL) {
214 FILEDESC_SUNLOCK(fdp);
215 if (error == 0 && dp->v_type != VDIR) {
216 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
217 vrele(dp);
218 VFS_UNLOCK_GIANT(vfslocked);
219 error = ENOTDIR;
220 }
221 }
222 if (error) {
223 uma_zfree(namei_zone, cnp->cn_pnbuf);
224 #ifdef DIAGNOSTIC
225 cnp->cn_pnbuf = NULL;
226 cnp->cn_nameptr = NULL;
227 #endif
228 return (error);
229 }
230 }
231 if (dp == NULL) {
232 dp = fdp->fd_cdir;
233 VREF(dp);
234 FILEDESC_SUNLOCK(fdp);
235 if (ndp->ni_startdir != NULL) {
236 vfslocked = VFS_LOCK_GIANT(ndp->ni_startdir->v_mount);
237 vrele(ndp->ni_startdir);
238 VFS_UNLOCK_GIANT(vfslocked);
239 }
240 }
241 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
242 cnp->cn_flags, 0, 0);
243 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
244 for (;;) {
245 /*
246 * Check if root directory should replace current directory.
247 * Done at start of translation and after symbolic link.
248 */
249 cnp->cn_nameptr = cnp->cn_pnbuf;
250 if (*(cnp->cn_nameptr) == '/') {
251 vrele(dp);
252 VFS_UNLOCK_GIANT(vfslocked);
253 while (*(cnp->cn_nameptr) == '/') {
254 cnp->cn_nameptr++;
255 ndp->ni_pathlen--;
256 }
257 dp = ndp->ni_rootdir;
258 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
259 VREF(dp);
260 }
261 if (vfslocked)
262 ndp->ni_cnd.cn_flags |= GIANTHELD;
263 ndp->ni_startdir = dp;
264 error = lookup(ndp);
265 if (error) {
266 uma_zfree(namei_zone, cnp->cn_pnbuf);
267 #ifdef DIAGNOSTIC
268 cnp->cn_pnbuf = NULL;
269 cnp->cn_nameptr = NULL;
270 #endif
271 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
272 0, 0);
273 return (error);
274 }
275 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
276 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
277 /*
278 * If not a symbolic link, we're done.
279 */
280 if ((cnp->cn_flags & ISSYMLINK) == 0) {
281 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
282 uma_zfree(namei_zone, cnp->cn_pnbuf);
283 #ifdef DIAGNOSTIC
284 cnp->cn_pnbuf = NULL;
285 cnp->cn_nameptr = NULL;
286 #endif
287 } else
288 cnp->cn_flags |= HASBUF;
289
290 if ((cnp->cn_flags & MPSAFE) == 0) {
291 VFS_UNLOCK_GIANT(vfslocked);
292 } else if (vfslocked)
293 ndp->ni_cnd.cn_flags |= GIANTHELD;
294 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
295 0, 0, 0);
296 return (0);
297 }
298 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
299 error = ELOOP;
300 break;
301 }
302 #ifdef MAC
303 if ((cnp->cn_flags & NOMACCHECK) == 0) {
304 error = mac_vnode_check_readlink(td->td_ucred,
305 ndp->ni_vp);
306 if (error)
307 break;
308 }
309 #endif
310 if (ndp->ni_pathlen > 1)
311 cp = uma_zalloc(namei_zone, M_WAITOK);
312 else
313 cp = cnp->cn_pnbuf;
314 aiov.iov_base = cp;
315 aiov.iov_len = MAXPATHLEN;
316 auio.uio_iov = &aiov;
317 auio.uio_iovcnt = 1;
318 auio.uio_offset = 0;
319 auio.uio_rw = UIO_READ;
320 auio.uio_segflg = UIO_SYSSPACE;
321 auio.uio_td = (struct thread *)0;
322 auio.uio_resid = MAXPATHLEN;
323 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
324 if (error) {
325 if (ndp->ni_pathlen > 1)
326 uma_zfree(namei_zone, cp);
327 break;
328 }
329 linklen = MAXPATHLEN - auio.uio_resid;
330 if (linklen == 0) {
331 if (ndp->ni_pathlen > 1)
332 uma_zfree(namei_zone, cp);
333 error = ENOENT;
334 break;
335 }
336 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
337 if (ndp->ni_pathlen > 1)
338 uma_zfree(namei_zone, cp);
339 error = ENAMETOOLONG;
340 break;
341 }
342 if (ndp->ni_pathlen > 1) {
343 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
344 uma_zfree(namei_zone, cnp->cn_pnbuf);
345 cnp->cn_pnbuf = cp;
346 } else
347 cnp->cn_pnbuf[linklen] = '\0';
348 ndp->ni_pathlen += linklen;
349 vput(ndp->ni_vp);
350 dp = ndp->ni_dvp;
351 }
352 uma_zfree(namei_zone, cnp->cn_pnbuf);
353 #ifdef DIAGNOSTIC
354 cnp->cn_pnbuf = NULL;
355 cnp->cn_nameptr = NULL;
356 #endif
357 vput(ndp->ni_vp);
358 ndp->ni_vp = NULL;
359 vrele(ndp->ni_dvp);
360 VFS_UNLOCK_GIANT(vfslocked);
361 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
362 return (error);
363 }
364
365 static int
366 compute_cn_lkflags(struct mount *mp, int lkflags)
367 {
368
369 if (mp == NULL ||
370 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
371 lkflags &= ~LK_SHARED;
372 lkflags |= LK_EXCLUSIVE;
373 }
374 return (lkflags);
375 }
376
377 static __inline int
378 needs_exclusive_leaf(struct mount *mp, int flags)
379 {
380
381 /*
382 * Intermediate nodes can use shared locks, we only need to
383 * force an exclusive lock for leaf nodes.
384 */
385 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
386 return (0);
387
388 /* Always use exclusive locks if LOCKSHARED isn't set. */
389 if (!(flags & LOCKSHARED))
390 return (1);
391
392 /*
393 * For lookups during open(), if the mount point supports
394 * extended shared operations, then use a shared lock for the
395 * leaf node, otherwise use an exclusive lock.
396 */
397 if (flags & ISOPEN) {
398 if (mp != NULL &&
399 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
400 return (0);
401 else
402 return (1);
403 }
404
405 /*
406 * Lookup requests outside of open() that specify LOCKSHARED
407 * only need a shared lock on the leaf vnode.
408 */
409 return (0);
410 }
411
412 /*
413 * Search a pathname.
414 * This is a very central and rather complicated routine.
415 *
416 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
417 * The starting directory is taken from ni_startdir. The pathname is
418 * descended until done, or a symbolic link is encountered. The variable
419 * ni_more is clear if the path is completed; it is set to one if a
420 * symbolic link needing interpretation is encountered.
421 *
422 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
423 * whether the name is to be looked up, created, renamed, or deleted.
424 * When CREATE, RENAME, or DELETE is specified, information usable in
425 * creating, renaming, or deleting a directory entry may be calculated.
426 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
427 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
428 * returned unlocked. Otherwise the parent directory is not returned. If
429 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
430 * the target is returned locked, otherwise it is returned unlocked.
431 * When creating or renaming and LOCKPARENT is specified, the target may not
432 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
433 *
434 * Overall outline of lookup:
435 *
436 * dirloop:
437 * identify next component of name at ndp->ni_ptr
438 * handle degenerate case where name is null string
439 * if .. and crossing mount points and on mounted filesys, find parent
440 * call VOP_LOOKUP routine for next component name
441 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
442 * component vnode returned in ni_vp (if it exists), locked.
443 * if result vnode is mounted on and crossing mount points,
444 * find mounted on vnode
445 * if more components of name, do next level at dirloop
446 * return the answer in ni_vp, locked if LOCKLEAF set
447 * if LOCKPARENT set, return locked parent in ni_dvp
448 * if WANTPARENT set, return unlocked parent in ni_dvp
449 */
450 int
451 lookup(struct nameidata *ndp)
452 {
453 char *cp; /* pointer into pathname argument */
454 struct vnode *dp = 0; /* the directory we are searching */
455 struct vnode *tdp; /* saved dp */
456 struct mount *mp; /* mount table entry */
457 struct prison *pr;
458 int docache; /* == 0 do not cache last component */
459 int wantparent; /* 1 => wantparent or lockparent flag */
460 int rdonly; /* lookup read-only flag bit */
461 int error = 0;
462 int dpunlocked = 0; /* dp has already been unlocked */
463 struct componentname *cnp = &ndp->ni_cnd;
464 int vfslocked; /* VFS Giant state for child */
465 int dvfslocked; /* VFS Giant state for parent */
466 int tvfslocked;
467 int lkflags_save;
468
469 /*
470 * Setup: break out flag bits into variables.
471 */
472 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
473 vfslocked = 0;
474 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
475 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
476 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
477 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
478 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
479 if (cnp->cn_nameiop == DELETE ||
480 (wantparent && cnp->cn_nameiop != CREATE &&
481 cnp->cn_nameiop != LOOKUP))
482 docache = 0;
483 rdonly = cnp->cn_flags & RDONLY;
484 cnp->cn_flags &= ~ISSYMLINK;
485 ndp->ni_dvp = NULL;
486 /*
487 * We use shared locks until we hit the parent of the last cn then
488 * we adjust based on the requesting flags.
489 */
490 if (lookup_shared)
491 cnp->cn_lkflags = LK_SHARED;
492 else
493 cnp->cn_lkflags = LK_EXCLUSIVE;
494 dp = ndp->ni_startdir;
495 ndp->ni_startdir = NULLVP;
496 vn_lock(dp,
497 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
498
499 dirloop:
500 /*
501 * Search a new directory.
502 *
503 * The last component of the filename is left accessible via
504 * cnp->cn_nameptr for callers that need the name. Callers needing
505 * the name set the SAVENAME flag. When done, they assume
506 * responsibility for freeing the pathname buffer.
507 */
508 cnp->cn_consume = 0;
509 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
510 continue;
511 cnp->cn_namelen = cp - cnp->cn_nameptr;
512 if (cnp->cn_namelen > NAME_MAX) {
513 error = ENAMETOOLONG;
514 goto bad;
515 }
516 #ifdef NAMEI_DIAGNOSTIC
517 { char c = *cp;
518 *cp = '\0';
519 printf("{%s}: ", cnp->cn_nameptr);
520 *cp = c; }
521 #endif
522 ndp->ni_pathlen -= cnp->cn_namelen;
523 ndp->ni_next = cp;
524
525 /*
526 * Replace multiple slashes by a single slash and trailing slashes
527 * by a null. This must be done before VOP_LOOKUP() because some
528 * fs's don't know about trailing slashes. Remember if there were
529 * trailing slashes to handle symlinks, existing non-directories
530 * and non-existing files that won't be directories specially later.
531 */
532 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
533 cp++;
534 ndp->ni_pathlen--;
535 if (*cp == '\0') {
536 *ndp->ni_next = '\0';
537 cnp->cn_flags |= TRAILINGSLASH;
538 }
539 }
540 ndp->ni_next = cp;
541
542 cnp->cn_flags |= MAKEENTRY;
543 if (*cp == '\0' && docache == 0)
544 cnp->cn_flags &= ~MAKEENTRY;
545 if (cnp->cn_namelen == 2 &&
546 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
547 cnp->cn_flags |= ISDOTDOT;
548 else
549 cnp->cn_flags &= ~ISDOTDOT;
550 if (*ndp->ni_next == 0)
551 cnp->cn_flags |= ISLASTCN;
552 else
553 cnp->cn_flags &= ~ISLASTCN;
554
555 if ((cnp->cn_flags & ISLASTCN) != 0 &&
556 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
557 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
558 error = EINVAL;
559 goto bad;
560 }
561
562 /*
563 * Check for degenerate name (e.g. / or "")
564 * which is a way of talking about a directory,
565 * e.g. like "/." or ".".
566 */
567 if (cnp->cn_nameptr[0] == '\0') {
568 if (dp->v_type != VDIR) {
569 error = ENOTDIR;
570 goto bad;
571 }
572 if (cnp->cn_nameiop != LOOKUP) {
573 error = EISDIR;
574 goto bad;
575 }
576 if (wantparent) {
577 ndp->ni_dvp = dp;
578 VREF(dp);
579 }
580 ndp->ni_vp = dp;
581
582 if (cnp->cn_flags & AUDITVNODE1)
583 AUDIT_ARG_VNODE1(dp);
584 else if (cnp->cn_flags & AUDITVNODE2)
585 AUDIT_ARG_VNODE2(dp);
586
587 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
588 VOP_UNLOCK(dp, 0);
589 /* XXX This should probably move to the top of function. */
590 if (cnp->cn_flags & SAVESTART)
591 panic("lookup: SAVESTART");
592 goto success;
593 }
594
595 /*
596 * Handle "..": four special cases.
597 * 1. Return an error if this is the last component of
598 * the name and the operation is DELETE or RENAME.
599 * 2. If at root directory (e.g. after chroot)
600 * or at absolute root directory
601 * then ignore it so can't get out.
602 * 3. If this vnode is the root of a mounted
603 * filesystem, then replace it with the
604 * vnode which was mounted on so we take the
605 * .. in the other filesystem.
606 * 4. If the vnode is the top directory of
607 * the jail or chroot, don't let them out.
608 */
609 if (cnp->cn_flags & ISDOTDOT) {
610 if ((cnp->cn_flags & ISLASTCN) != 0 &&
611 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
612 error = EINVAL;
613 goto bad;
614 }
615 for (;;) {
616 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
617 pr = pr->pr_parent)
618 if (dp == pr->pr_root)
619 break;
620 if (dp == ndp->ni_rootdir ||
621 dp == ndp->ni_topdir ||
622 dp == rootvnode ||
623 pr != NULL ||
624 ((dp->v_vflag & VV_ROOT) != 0 &&
625 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
626 ndp->ni_dvp = dp;
627 ndp->ni_vp = dp;
628 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
629 VREF(dp);
630 goto nextname;
631 }
632 if ((dp->v_vflag & VV_ROOT) == 0)
633 break;
634 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
635 error = ENOENT;
636 goto bad;
637 }
638 tdp = dp;
639 dp = dp->v_mount->mnt_vnodecovered;
640 tvfslocked = dvfslocked;
641 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
642 VREF(dp);
643 vput(tdp);
644 VFS_UNLOCK_GIANT(tvfslocked);
645 vn_lock(dp,
646 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
647 LK_RETRY));
648 }
649 }
650
651 /*
652 * We now have a segment name to search for, and a directory to search.
653 */
654 unionlookup:
655 #ifdef MAC
656 if ((cnp->cn_flags & NOMACCHECK) == 0) {
657 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
658 cnp);
659 if (error)
660 goto bad;
661 }
662 #endif
663 ndp->ni_dvp = dp;
664 ndp->ni_vp = NULL;
665 ASSERT_VOP_LOCKED(dp, "lookup");
666 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
667 /*
668 * If we have a shared lock we may need to upgrade the lock for the
669 * last operation.
670 */
671 if (dp != vp_crossmp &&
672 VOP_ISLOCKED(dp) == LK_SHARED &&
673 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
674 vn_lock(dp, LK_UPGRADE|LK_RETRY);
675 /*
676 * If we're looking up the last component and we need an exclusive
677 * lock, adjust our lkflags.
678 */
679 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
680 cnp->cn_lkflags = LK_EXCLUSIVE;
681 #ifdef NAMEI_DIAGNOSTIC
682 vprint("lookup in", dp);
683 #endif
684 lkflags_save = cnp->cn_lkflags;
685 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
686 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
687 cnp->cn_lkflags = lkflags_save;
688 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
689 #ifdef NAMEI_DIAGNOSTIC
690 printf("not found\n");
691 #endif
692 if ((error == ENOENT) &&
693 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
694 (dp->v_mount->mnt_flag & MNT_UNION)) {
695 tdp = dp;
696 dp = dp->v_mount->mnt_vnodecovered;
697 tvfslocked = dvfslocked;
698 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
699 VREF(dp);
700 vput(tdp);
701 VFS_UNLOCK_GIANT(tvfslocked);
702 vn_lock(dp,
703 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
704 LK_RETRY));
705 goto unionlookup;
706 }
707
708 if (error != EJUSTRETURN)
709 goto bad;
710 /*
711 * At this point, we know we're at the end of the
712 * pathname. If creating / renaming, we can consider
713 * allowing the file or directory to be created / renamed,
714 * provided we're not on a read-only filesystem.
715 */
716 if (rdonly) {
717 error = EROFS;
718 goto bad;
719 }
720 /* trailing slash only allowed for directories */
721 if ((cnp->cn_flags & TRAILINGSLASH) &&
722 !(cnp->cn_flags & WILLBEDIR)) {
723 error = ENOENT;
724 goto bad;
725 }
726 if ((cnp->cn_flags & LOCKPARENT) == 0)
727 VOP_UNLOCK(dp, 0);
728 /*
729 * We return with ni_vp NULL to indicate that the entry
730 * doesn't currently exist, leaving a pointer to the
731 * (possibly locked) directory vnode in ndp->ni_dvp.
732 */
733 if (cnp->cn_flags & SAVESTART) {
734 ndp->ni_startdir = ndp->ni_dvp;
735 VREF(ndp->ni_startdir);
736 }
737 goto success;
738 } else
739 cnp->cn_lkflags = lkflags_save;
740 #ifdef NAMEI_DIAGNOSTIC
741 printf("found\n");
742 #endif
743 /*
744 * Take into account any additional components consumed by
745 * the underlying filesystem.
746 */
747 if (cnp->cn_consume > 0) {
748 cnp->cn_nameptr += cnp->cn_consume;
749 ndp->ni_next += cnp->cn_consume;
750 ndp->ni_pathlen -= cnp->cn_consume;
751 cnp->cn_consume = 0;
752 }
753
754 dp = ndp->ni_vp;
755 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
756
757 /*
758 * Check to see if the vnode has been mounted on;
759 * if so find the root of the mounted filesystem.
760 */
761 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
762 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
763 if (vfs_busy(mp, 0))
764 continue;
765 vput(dp);
766 VFS_UNLOCK_GIANT(vfslocked);
767 vfslocked = VFS_LOCK_GIANT(mp);
768 if (dp != ndp->ni_dvp)
769 vput(ndp->ni_dvp);
770 else
771 vrele(ndp->ni_dvp);
772 VFS_UNLOCK_GIANT(dvfslocked);
773 dvfslocked = 0;
774 vref(vp_crossmp);
775 ndp->ni_dvp = vp_crossmp;
776 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags),
777 &tdp);
778 vfs_unbusy(mp);
779 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
780 panic("vp_crossmp exclusively locked or reclaimed");
781 if (error) {
782 dpunlocked = 1;
783 goto bad2;
784 }
785 ndp->ni_vp = dp = tdp;
786 }
787
788 /*
789 * Check for symbolic link
790 */
791 if ((dp->v_type == VLNK) &&
792 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
793 *ndp->ni_next == '/')) {
794 cnp->cn_flags |= ISSYMLINK;
795 if (dp->v_iflag & VI_DOOMED) {
796 /*
797 * We can't know whether the directory was mounted with
798 * NOSYMFOLLOW, so we can't follow safely.
799 */
800 error = ENOENT;
801 goto bad2;
802 }
803 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
804 error = EACCES;
805 goto bad2;
806 }
807 /*
808 * Symlink code always expects an unlocked dvp.
809 */
810 if (ndp->ni_dvp != ndp->ni_vp)
811 VOP_UNLOCK(ndp->ni_dvp, 0);
812 goto success;
813 }
814
815 nextname:
816 /*
817 * Not a symbolic link that we will follow. Continue with the
818 * next component if there is any; otherwise, we're done.
819 */
820 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
821 ("lookup: invalid path state."));
822 if (*ndp->ni_next == '/') {
823 cnp->cn_nameptr = ndp->ni_next;
824 while (*cnp->cn_nameptr == '/') {
825 cnp->cn_nameptr++;
826 ndp->ni_pathlen--;
827 }
828 if (ndp->ni_dvp != dp)
829 vput(ndp->ni_dvp);
830 else
831 vrele(ndp->ni_dvp);
832 VFS_UNLOCK_GIANT(dvfslocked);
833 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
834 vfslocked = 0;
835 goto dirloop;
836 }
837 /*
838 * If we're processing a path with a trailing slash,
839 * check that the end result is a directory.
840 */
841 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
842 error = ENOTDIR;
843 goto bad2;
844 }
845 /*
846 * Disallow directory write attempts on read-only filesystems.
847 */
848 if (rdonly &&
849 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
850 error = EROFS;
851 goto bad2;
852 }
853 if (cnp->cn_flags & SAVESTART) {
854 ndp->ni_startdir = ndp->ni_dvp;
855 VREF(ndp->ni_startdir);
856 }
857 if (!wantparent) {
858 if (ndp->ni_dvp != dp)
859 vput(ndp->ni_dvp);
860 else
861 vrele(ndp->ni_dvp);
862 VFS_UNLOCK_GIANT(dvfslocked);
863 dvfslocked = 0;
864 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
865 VOP_UNLOCK(ndp->ni_dvp, 0);
866
867 if (cnp->cn_flags & AUDITVNODE1)
868 AUDIT_ARG_VNODE1(dp);
869 else if (cnp->cn_flags & AUDITVNODE2)
870 AUDIT_ARG_VNODE2(dp);
871
872 if ((cnp->cn_flags & LOCKLEAF) == 0)
873 VOP_UNLOCK(dp, 0);
874 success:
875 /*
876 * Because of lookup_shared we may have the vnode shared locked, but
877 * the caller may want it to be exclusively locked.
878 */
879 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
880 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
881 vn_lock(dp, LK_UPGRADE | LK_RETRY);
882 if (dp->v_iflag & VI_DOOMED) {
883 error = ENOENT;
884 goto bad2;
885 }
886 }
887 if (vfslocked && dvfslocked)
888 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
889 if (vfslocked || dvfslocked)
890 ndp->ni_cnd.cn_flags |= GIANTHELD;
891 return (0);
892
893 bad2:
894 if (dp != ndp->ni_dvp)
895 vput(ndp->ni_dvp);
896 else
897 vrele(ndp->ni_dvp);
898 bad:
899 if (!dpunlocked)
900 vput(dp);
901 VFS_UNLOCK_GIANT(vfslocked);
902 VFS_UNLOCK_GIANT(dvfslocked);
903 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
904 ndp->ni_vp = NULL;
905 return (error);
906 }
907
908 /*
909 * relookup - lookup a path name component
910 * Used by lookup to re-acquire things.
911 */
912 int
913 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
914 {
915 struct vnode *dp = 0; /* the directory we are searching */
916 int wantparent; /* 1 => wantparent or lockparent flag */
917 int rdonly; /* lookup read-only flag bit */
918 int error = 0;
919
920 KASSERT(cnp->cn_flags & ISLASTCN,
921 ("relookup: Not given last component."));
922 /*
923 * Setup: break out flag bits into variables.
924 */
925 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
926 KASSERT(wantparent, ("relookup: parent not wanted."));
927 rdonly = cnp->cn_flags & RDONLY;
928 cnp->cn_flags &= ~ISSYMLINK;
929 dp = dvp;
930 cnp->cn_lkflags = LK_EXCLUSIVE;
931 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
932
933 /*
934 * Search a new directory.
935 *
936 * The last component of the filename is left accessible via
937 * cnp->cn_nameptr for callers that need the name. Callers needing
938 * the name set the SAVENAME flag. When done, they assume
939 * responsibility for freeing the pathname buffer.
940 */
941 #ifdef NAMEI_DIAGNOSTIC
942 printf("{%s}: ", cnp->cn_nameptr);
943 #endif
944
945 /*
946 * Check for degenerate name (e.g. / or "")
947 * which is a way of talking about a directory,
948 * e.g. like "/." or ".".
949 */
950 if (cnp->cn_nameptr[0] == '\0') {
951 if (cnp->cn_nameiop != LOOKUP || wantparent) {
952 error = EISDIR;
953 goto bad;
954 }
955 if (dp->v_type != VDIR) {
956 error = ENOTDIR;
957 goto bad;
958 }
959 if (!(cnp->cn_flags & LOCKLEAF))
960 VOP_UNLOCK(dp, 0);
961 *vpp = dp;
962 /* XXX This should probably move to the top of function. */
963 if (cnp->cn_flags & SAVESTART)
964 panic("lookup: SAVESTART");
965 return (0);
966 }
967
968 if (cnp->cn_flags & ISDOTDOT)
969 panic ("relookup: lookup on dot-dot");
970
971 /*
972 * We now have a segment name to search for, and a directory to search.
973 */
974 #ifdef NAMEI_DIAGNOSTIC
975 vprint("search in:", dp);
976 #endif
977 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
978 KASSERT(*vpp == NULL, ("leaf should be empty"));
979 if (error != EJUSTRETURN)
980 goto bad;
981 /*
982 * If creating and at end of pathname, then can consider
983 * allowing file to be created.
984 */
985 if (rdonly) {
986 error = EROFS;
987 goto bad;
988 }
989 /* ASSERT(dvp == ndp->ni_startdir) */
990 if (cnp->cn_flags & SAVESTART)
991 VREF(dvp);
992 if ((cnp->cn_flags & LOCKPARENT) == 0)
993 VOP_UNLOCK(dp, 0);
994 /*
995 * We return with ni_vp NULL to indicate that the entry
996 * doesn't currently exist, leaving a pointer to the
997 * (possibly locked) directory vnode in ndp->ni_dvp.
998 */
999 return (0);
1000 }
1001
1002 dp = *vpp;
1003
1004 /*
1005 * Disallow directory write attempts on read-only filesystems.
1006 */
1007 if (rdonly &&
1008 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1009 if (dvp == dp)
1010 vrele(dvp);
1011 else
1012 vput(dvp);
1013 error = EROFS;
1014 goto bad;
1015 }
1016 /*
1017 * Set the parent lock/ref state to the requested state.
1018 */
1019 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1020 if (wantparent)
1021 VOP_UNLOCK(dvp, 0);
1022 else
1023 vput(dvp);
1024 } else if (!wantparent)
1025 vrele(dvp);
1026 /*
1027 * Check for symbolic link
1028 */
1029 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1030 ("relookup: symlink found.\n"));
1031
1032 /* ASSERT(dvp == ndp->ni_startdir) */
1033 if (cnp->cn_flags & SAVESTART)
1034 VREF(dvp);
1035
1036 if ((cnp->cn_flags & LOCKLEAF) == 0)
1037 VOP_UNLOCK(dp, 0);
1038 return (0);
1039 bad:
1040 vput(dp);
1041 *vpp = NULL;
1042 return (error);
1043 }
1044
1045 /*
1046 * Free data allocated by namei(); see namei(9) for details.
1047 */
1048 void
1049 NDFREE(struct nameidata *ndp, const u_int flags)
1050 {
1051 int unlock_dvp;
1052 int unlock_vp;
1053
1054 unlock_dvp = 0;
1055 unlock_vp = 0;
1056
1057 if (!(flags & NDF_NO_FREE_PNBUF) &&
1058 (ndp->ni_cnd.cn_flags & HASBUF)) {
1059 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1060 ndp->ni_cnd.cn_flags &= ~HASBUF;
1061 }
1062 if (!(flags & NDF_NO_VP_UNLOCK) &&
1063 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1064 unlock_vp = 1;
1065 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1066 if (unlock_vp) {
1067 vput(ndp->ni_vp);
1068 unlock_vp = 0;
1069 } else
1070 vrele(ndp->ni_vp);
1071 ndp->ni_vp = NULL;
1072 }
1073 if (unlock_vp)
1074 VOP_UNLOCK(ndp->ni_vp, 0);
1075 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1076 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1077 ndp->ni_dvp != ndp->ni_vp)
1078 unlock_dvp = 1;
1079 if (!(flags & NDF_NO_DVP_RELE) &&
1080 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1081 if (unlock_dvp) {
1082 vput(ndp->ni_dvp);
1083 unlock_dvp = 0;
1084 } else
1085 vrele(ndp->ni_dvp);
1086 ndp->ni_dvp = NULL;
1087 }
1088 if (unlock_dvp)
1089 VOP_UNLOCK(ndp->ni_dvp, 0);
1090 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1091 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1092 vrele(ndp->ni_startdir);
1093 ndp->ni_startdir = NULL;
1094 }
1095 }
1096
1097 /*
1098 * Determine if there is a suitable alternate filename under the specified
1099 * prefix for the specified path. If the create flag is set, then the
1100 * alternate prefix will be used so long as the parent directory exists.
1101 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1102 * files under /compat/linux for example. The chosen path (whether under
1103 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1104 * to by pathbuf. The caller is responsible for free'ing the buffer from
1105 * the M_TEMP bucket if one is returned.
1106 */
1107 int
1108 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1109 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1110 {
1111 struct nameidata nd, ndroot;
1112 char *ptr, *buf, *cp;
1113 size_t len, sz;
1114 int error;
1115
1116 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1117 *pathbuf = buf;
1118
1119 /* Copy the prefix into the new pathname as a starting point. */
1120 len = strlcpy(buf, prefix, MAXPATHLEN);
1121 if (len >= MAXPATHLEN) {
1122 *pathbuf = NULL;
1123 free(buf, M_TEMP);
1124 return (EINVAL);
1125 }
1126 sz = MAXPATHLEN - len;
1127 ptr = buf + len;
1128
1129 /* Append the filename to the prefix. */
1130 if (pathseg == UIO_SYSSPACE)
1131 error = copystr(path, ptr, sz, &len);
1132 else
1133 error = copyinstr(path, ptr, sz, &len);
1134
1135 if (error) {
1136 *pathbuf = NULL;
1137 free(buf, M_TEMP);
1138 return (error);
1139 }
1140
1141 /* Only use a prefix with absolute pathnames. */
1142 if (*ptr != '/') {
1143 error = EINVAL;
1144 goto keeporig;
1145 }
1146
1147 if (dirfd != AT_FDCWD) {
1148 /*
1149 * We want the original because the "prefix" is
1150 * included in the already opened dirfd.
1151 */
1152 bcopy(ptr, buf, len);
1153 return (0);
1154 }
1155
1156 /*
1157 * We know that there is a / somewhere in this pathname.
1158 * Search backwards for it, to find the file's parent dir
1159 * to see if it exists in the alternate tree. If it does,
1160 * and we want to create a file (cflag is set). We don't
1161 * need to worry about the root comparison in this case.
1162 */
1163
1164 if (create) {
1165 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1166 *cp = '\0';
1167
1168 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1169 error = namei(&nd);
1170 *cp = '/';
1171 if (error != 0)
1172 goto keeporig;
1173 } else {
1174 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1175
1176 error = namei(&nd);
1177 if (error != 0)
1178 goto keeporig;
1179
1180 /*
1181 * We now compare the vnode of the prefix to the one
1182 * vnode asked. If they resolve to be the same, then we
1183 * ignore the match so that the real root gets used.
1184 * This avoids the problem of traversing "../.." to find the
1185 * root directory and never finding it, because "/" resolves
1186 * to the emulation root directory. This is expensive :-(
1187 */
1188 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1189 td);
1190
1191 /* We shouldn't ever get an error from this namei(). */
1192 error = namei(&ndroot);
1193 if (error == 0) {
1194 if (nd.ni_vp == ndroot.ni_vp)
1195 error = ENOENT;
1196
1197 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1198 vrele(ndroot.ni_vp);
1199 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1200 }
1201 }
1202
1203 NDFREE(&nd, NDF_ONLY_PNBUF);
1204 vrele(nd.ni_vp);
1205 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1206
1207 keeporig:
1208 /* If there was an error, use the original path name. */
1209 if (error)
1210 bcopy(ptr, buf, len);
1211 return (error);
1212 }
Cache object: 06025d54177b3c124289a014a83df4be
|