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_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, entry, "struct vnode *", "char *",
72 "unsigned long");
73 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, 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 (error == 0) {
166 /*
167 * If we are auditing the kernel pathname, save the user
168 * pathname.
169 */
170 if (cnp->cn_flags & AUDITVNODE1)
171 AUDIT_ARG_UPATH1(td, cnp->cn_pnbuf);
172 if (cnp->cn_flags & AUDITVNODE2)
173 AUDIT_ARG_UPATH2(td, cnp->cn_pnbuf);
174 }
175
176 /*
177 * Don't allow empty pathnames.
178 */
179 if (!error && *cnp->cn_pnbuf == '\0')
180 error = ENOENT;
181
182 if (error) {
183 uma_zfree(namei_zone, cnp->cn_pnbuf);
184 #ifdef DIAGNOSTIC
185 cnp->cn_pnbuf = NULL;
186 cnp->cn_nameptr = NULL;
187 #endif
188 ndp->ni_vp = NULL;
189 return (error);
190 }
191 ndp->ni_loopcnt = 0;
192 #ifdef KTRACE
193 if (KTRPOINT(td, KTR_NAMEI)) {
194 KASSERT(cnp->cn_thread == curthread,
195 ("namei not using curthread"));
196 ktrnamei(cnp->cn_pnbuf);
197 }
198 #endif
199 /*
200 * Get starting point for the translation.
201 */
202 FILEDESC_SLOCK(fdp);
203 ndp->ni_rootdir = fdp->fd_rdir;
204 ndp->ni_topdir = fdp->fd_jdir;
205
206 dp = NULL;
207 if (cnp->cn_pnbuf[0] != '/') {
208 if (ndp->ni_startdir != NULL) {
209 dp = ndp->ni_startdir;
210 error = 0;
211 } else if (ndp->ni_dirfd != AT_FDCWD) {
212 if (cnp->cn_flags & AUDITVNODE1)
213 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
214 if (cnp->cn_flags & AUDITVNODE2)
215 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
216 error = fgetvp(td, ndp->ni_dirfd, &dp);
217 }
218 if (error != 0 || dp != NULL) {
219 FILEDESC_SUNLOCK(fdp);
220 if (error == 0 && dp->v_type != VDIR) {
221 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
222 vrele(dp);
223 VFS_UNLOCK_GIANT(vfslocked);
224 error = ENOTDIR;
225 }
226 }
227 if (error) {
228 uma_zfree(namei_zone, cnp->cn_pnbuf);
229 #ifdef DIAGNOSTIC
230 cnp->cn_pnbuf = NULL;
231 cnp->cn_nameptr = NULL;
232 #endif
233 return (error);
234 }
235 }
236 if (dp == NULL) {
237 dp = fdp->fd_cdir;
238 VREF(dp);
239 FILEDESC_SUNLOCK(fdp);
240 if (ndp->ni_startdir != NULL) {
241 vfslocked = VFS_LOCK_GIANT(ndp->ni_startdir->v_mount);
242 vrele(ndp->ni_startdir);
243 VFS_UNLOCK_GIANT(vfslocked);
244 }
245 }
246 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
247 cnp->cn_flags, 0, 0);
248 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
249 for (;;) {
250 /*
251 * Check if root directory should replace current directory.
252 * Done at start of translation and after symbolic link.
253 */
254 cnp->cn_nameptr = cnp->cn_pnbuf;
255 if (*(cnp->cn_nameptr) == '/') {
256 vrele(dp);
257 VFS_UNLOCK_GIANT(vfslocked);
258 while (*(cnp->cn_nameptr) == '/') {
259 cnp->cn_nameptr++;
260 ndp->ni_pathlen--;
261 }
262 dp = ndp->ni_rootdir;
263 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
264 VREF(dp);
265 }
266 if (vfslocked)
267 ndp->ni_cnd.cn_flags |= GIANTHELD;
268 ndp->ni_startdir = dp;
269 error = lookup(ndp);
270 if (error) {
271 uma_zfree(namei_zone, cnp->cn_pnbuf);
272 #ifdef DIAGNOSTIC
273 cnp->cn_pnbuf = NULL;
274 cnp->cn_nameptr = NULL;
275 #endif
276 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
277 0, 0);
278 return (error);
279 }
280 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
281 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
282 /*
283 * If not a symbolic link, we're done.
284 */
285 if ((cnp->cn_flags & ISSYMLINK) == 0) {
286 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
287 uma_zfree(namei_zone, cnp->cn_pnbuf);
288 #ifdef DIAGNOSTIC
289 cnp->cn_pnbuf = NULL;
290 cnp->cn_nameptr = NULL;
291 #endif
292 } else
293 cnp->cn_flags |= HASBUF;
294
295 if ((cnp->cn_flags & MPSAFE) == 0) {
296 VFS_UNLOCK_GIANT(vfslocked);
297 } else if (vfslocked)
298 ndp->ni_cnd.cn_flags |= GIANTHELD;
299 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
300 0, 0, 0);
301 return (0);
302 }
303 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
304 error = ELOOP;
305 break;
306 }
307 #ifdef MAC
308 if ((cnp->cn_flags & NOMACCHECK) == 0) {
309 error = mac_vnode_check_readlink(td->td_ucred,
310 ndp->ni_vp);
311 if (error)
312 break;
313 }
314 #endif
315 if (ndp->ni_pathlen > 1)
316 cp = uma_zalloc(namei_zone, M_WAITOK);
317 else
318 cp = cnp->cn_pnbuf;
319 aiov.iov_base = cp;
320 aiov.iov_len = MAXPATHLEN;
321 auio.uio_iov = &aiov;
322 auio.uio_iovcnt = 1;
323 auio.uio_offset = 0;
324 auio.uio_rw = UIO_READ;
325 auio.uio_segflg = UIO_SYSSPACE;
326 auio.uio_td = td;
327 auio.uio_resid = MAXPATHLEN;
328 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
329 if (error) {
330 if (ndp->ni_pathlen > 1)
331 uma_zfree(namei_zone, cp);
332 break;
333 }
334 linklen = MAXPATHLEN - auio.uio_resid;
335 if (linklen == 0) {
336 if (ndp->ni_pathlen > 1)
337 uma_zfree(namei_zone, cp);
338 error = ENOENT;
339 break;
340 }
341 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
342 if (ndp->ni_pathlen > 1)
343 uma_zfree(namei_zone, cp);
344 error = ENAMETOOLONG;
345 break;
346 }
347 if (ndp->ni_pathlen > 1) {
348 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
349 uma_zfree(namei_zone, cnp->cn_pnbuf);
350 cnp->cn_pnbuf = cp;
351 } else
352 cnp->cn_pnbuf[linklen] = '\0';
353 ndp->ni_pathlen += linklen;
354 vput(ndp->ni_vp);
355 dp = ndp->ni_dvp;
356 }
357 uma_zfree(namei_zone, cnp->cn_pnbuf);
358 #ifdef DIAGNOSTIC
359 cnp->cn_pnbuf = NULL;
360 cnp->cn_nameptr = NULL;
361 #endif
362 vput(ndp->ni_vp);
363 ndp->ni_vp = NULL;
364 vrele(ndp->ni_dvp);
365 VFS_UNLOCK_GIANT(vfslocked);
366 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
367 return (error);
368 }
369
370 static int
371 compute_cn_lkflags(struct mount *mp, int lkflags)
372 {
373
374 if (mp == NULL ||
375 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
376 lkflags &= ~LK_SHARED;
377 lkflags |= LK_EXCLUSIVE;
378 }
379 return (lkflags);
380 }
381
382 static __inline int
383 needs_exclusive_leaf(struct mount *mp, int flags)
384 {
385
386 /*
387 * Intermediate nodes can use shared locks, we only need to
388 * force an exclusive lock for leaf nodes.
389 */
390 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
391 return (0);
392
393 /* Always use exclusive locks if LOCKSHARED isn't set. */
394 if (!(flags & LOCKSHARED))
395 return (1);
396
397 /*
398 * For lookups during open(), if the mount point supports
399 * extended shared operations, then use a shared lock for the
400 * leaf node, otherwise use an exclusive lock.
401 */
402 if (flags & ISOPEN) {
403 if (mp != NULL &&
404 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
405 return (0);
406 else
407 return (1);
408 }
409
410 /*
411 * Lookup requests outside of open() that specify LOCKSHARED
412 * only need a shared lock on the leaf vnode.
413 */
414 return (0);
415 }
416
417 /*
418 * Search a pathname.
419 * This is a very central and rather complicated routine.
420 *
421 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
422 * The starting directory is taken from ni_startdir. The pathname is
423 * descended until done, or a symbolic link is encountered. The variable
424 * ni_more is clear if the path is completed; it is set to one if a
425 * symbolic link needing interpretation is encountered.
426 *
427 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
428 * whether the name is to be looked up, created, renamed, or deleted.
429 * When CREATE, RENAME, or DELETE is specified, information usable in
430 * creating, renaming, or deleting a directory entry may be calculated.
431 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
432 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
433 * returned unlocked. Otherwise the parent directory is not returned. If
434 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
435 * the target is returned locked, otherwise it is returned unlocked.
436 * When creating or renaming and LOCKPARENT is specified, the target may not
437 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
438 *
439 * Overall outline of lookup:
440 *
441 * dirloop:
442 * identify next component of name at ndp->ni_ptr
443 * handle degenerate case where name is null string
444 * if .. and crossing mount points and on mounted filesys, find parent
445 * call VOP_LOOKUP routine for next component name
446 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
447 * component vnode returned in ni_vp (if it exists), locked.
448 * if result vnode is mounted on and crossing mount points,
449 * find mounted on vnode
450 * if more components of name, do next level at dirloop
451 * return the answer in ni_vp, locked if LOCKLEAF set
452 * if LOCKPARENT set, return locked parent in ni_dvp
453 * if WANTPARENT set, return unlocked parent in ni_dvp
454 */
455 int
456 lookup(struct nameidata *ndp)
457 {
458 char *cp; /* pointer into pathname argument */
459 struct vnode *dp = 0; /* the directory we are searching */
460 struct vnode *tdp; /* saved dp */
461 struct mount *mp; /* mount table entry */
462 struct prison *pr;
463 int docache; /* == 0 do not cache last component */
464 int wantparent; /* 1 => wantparent or lockparent flag */
465 int rdonly; /* lookup read-only flag bit */
466 int error = 0;
467 int dpunlocked = 0; /* dp has already been unlocked */
468 struct componentname *cnp = &ndp->ni_cnd;
469 int vfslocked; /* VFS Giant state for child */
470 int dvfslocked; /* VFS Giant state for parent */
471 int tvfslocked;
472 int lkflags_save;
473 int ni_dvp_unlocked;
474
475 /*
476 * Setup: break out flag bits into variables.
477 */
478 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
479 vfslocked = 0;
480 ni_dvp_unlocked = 0;
481 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
482 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
483 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
484 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
485 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
486 if (cnp->cn_nameiop == DELETE ||
487 (wantparent && cnp->cn_nameiop != CREATE &&
488 cnp->cn_nameiop != LOOKUP))
489 docache = 0;
490 rdonly = cnp->cn_flags & RDONLY;
491 cnp->cn_flags &= ~ISSYMLINK;
492 ndp->ni_dvp = NULL;
493 /*
494 * We use shared locks until we hit the parent of the last cn then
495 * we adjust based on the requesting flags.
496 */
497 if (lookup_shared)
498 cnp->cn_lkflags = LK_SHARED;
499 else
500 cnp->cn_lkflags = LK_EXCLUSIVE;
501 dp = ndp->ni_startdir;
502 ndp->ni_startdir = NULLVP;
503 vn_lock(dp,
504 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
505
506 dirloop:
507 /*
508 * Search a new directory.
509 *
510 * The last component of the filename is left accessible via
511 * cnp->cn_nameptr for callers that need the name. Callers needing
512 * the name set the SAVENAME flag. When done, they assume
513 * responsibility for freeing the pathname buffer.
514 */
515 cnp->cn_consume = 0;
516 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
517 continue;
518 cnp->cn_namelen = cp - cnp->cn_nameptr;
519 if (cnp->cn_namelen > NAME_MAX) {
520 error = ENAMETOOLONG;
521 goto bad;
522 }
523 #ifdef NAMEI_DIAGNOSTIC
524 { char c = *cp;
525 *cp = '\0';
526 printf("{%s}: ", cnp->cn_nameptr);
527 *cp = c; }
528 #endif
529 ndp->ni_pathlen -= cnp->cn_namelen;
530 ndp->ni_next = cp;
531
532 /*
533 * Replace multiple slashes by a single slash and trailing slashes
534 * by a null. This must be done before VOP_LOOKUP() because some
535 * fs's don't know about trailing slashes. Remember if there were
536 * trailing slashes to handle symlinks, existing non-directories
537 * and non-existing files that won't be directories specially later.
538 */
539 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
540 cp++;
541 ndp->ni_pathlen--;
542 if (*cp == '\0') {
543 *ndp->ni_next = '\0';
544 cnp->cn_flags |= TRAILINGSLASH;
545 }
546 }
547 ndp->ni_next = cp;
548
549 cnp->cn_flags |= MAKEENTRY;
550 if (*cp == '\0' && docache == 0)
551 cnp->cn_flags &= ~MAKEENTRY;
552 if (cnp->cn_namelen == 2 &&
553 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
554 cnp->cn_flags |= ISDOTDOT;
555 else
556 cnp->cn_flags &= ~ISDOTDOT;
557 if (*ndp->ni_next == 0)
558 cnp->cn_flags |= ISLASTCN;
559 else
560 cnp->cn_flags &= ~ISLASTCN;
561
562 if ((cnp->cn_flags & ISLASTCN) != 0 &&
563 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
564 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
565 error = EINVAL;
566 goto bad;
567 }
568
569 /*
570 * Check for degenerate name (e.g. / or "")
571 * which is a way of talking about a directory,
572 * e.g. like "/." or ".".
573 */
574 if (cnp->cn_nameptr[0] == '\0') {
575 if (dp->v_type != VDIR) {
576 error = ENOTDIR;
577 goto bad;
578 }
579 if (cnp->cn_nameiop != LOOKUP) {
580 error = EISDIR;
581 goto bad;
582 }
583 if (wantparent) {
584 ndp->ni_dvp = dp;
585 VREF(dp);
586 }
587 ndp->ni_vp = dp;
588
589 if (cnp->cn_flags & AUDITVNODE1)
590 AUDIT_ARG_VNODE1(dp);
591 else if (cnp->cn_flags & AUDITVNODE2)
592 AUDIT_ARG_VNODE2(dp);
593
594 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
595 VOP_UNLOCK(dp, 0);
596 /* XXX This should probably move to the top of function. */
597 if (cnp->cn_flags & SAVESTART)
598 panic("lookup: SAVESTART");
599 goto success;
600 }
601
602 /*
603 * Handle "..": four special cases.
604 * 1. Return an error if this is the last component of
605 * the name and the operation is DELETE or RENAME.
606 * 2. If at root directory (e.g. after chroot)
607 * or at absolute root directory
608 * then ignore it so can't get out.
609 * 3. If this vnode is the root of a mounted
610 * filesystem, then replace it with the
611 * vnode which was mounted on so we take the
612 * .. in the other filesystem.
613 * 4. If the vnode is the top directory of
614 * the jail or chroot, don't let them out.
615 */
616 if (cnp->cn_flags & ISDOTDOT) {
617 if ((cnp->cn_flags & ISLASTCN) != 0 &&
618 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
619 error = EINVAL;
620 goto bad;
621 }
622 for (;;) {
623 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
624 pr = pr->pr_parent)
625 if (dp == pr->pr_root)
626 break;
627 if (dp == ndp->ni_rootdir ||
628 dp == ndp->ni_topdir ||
629 dp == rootvnode ||
630 pr != NULL ||
631 ((dp->v_vflag & VV_ROOT) != 0 &&
632 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
633 ndp->ni_dvp = dp;
634 ndp->ni_vp = dp;
635 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
636 VREF(dp);
637 goto nextname;
638 }
639 if ((dp->v_vflag & VV_ROOT) == 0)
640 break;
641 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
642 error = ENOENT;
643 goto bad;
644 }
645 tdp = dp;
646 dp = dp->v_mount->mnt_vnodecovered;
647 tvfslocked = dvfslocked;
648 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
649 VREF(dp);
650 vput(tdp);
651 VFS_UNLOCK_GIANT(tvfslocked);
652 vn_lock(dp,
653 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
654 LK_RETRY));
655 }
656 }
657
658 /*
659 * We now have a segment name to search for, and a directory to search.
660 */
661 unionlookup:
662 #ifdef MAC
663 if ((cnp->cn_flags & NOMACCHECK) == 0) {
664 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
665 cnp);
666 if (error)
667 goto bad;
668 }
669 #endif
670 ndp->ni_dvp = dp;
671 ndp->ni_vp = NULL;
672 ASSERT_VOP_LOCKED(dp, "lookup");
673 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
674 /*
675 * If we have a shared lock we may need to upgrade the lock for the
676 * last operation.
677 */
678 if (dp != vp_crossmp &&
679 VOP_ISLOCKED(dp) == LK_SHARED &&
680 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
681 vn_lock(dp, LK_UPGRADE|LK_RETRY);
682 if ((dp->v_iflag & VI_DOOMED) != 0) {
683 error = ENOENT;
684 goto bad;
685 }
686 /*
687 * If we're looking up the last component and we need an exclusive
688 * lock, adjust our lkflags.
689 */
690 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
691 cnp->cn_lkflags = LK_EXCLUSIVE;
692 #ifdef NAMEI_DIAGNOSTIC
693 vprint("lookup in", dp);
694 #endif
695 lkflags_save = cnp->cn_lkflags;
696 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
697 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
698 cnp->cn_lkflags = lkflags_save;
699 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
700 #ifdef NAMEI_DIAGNOSTIC
701 printf("not found\n");
702 #endif
703 if ((error == ENOENT) &&
704 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
705 (dp->v_mount->mnt_flag & MNT_UNION)) {
706 tdp = dp;
707 dp = dp->v_mount->mnt_vnodecovered;
708 tvfslocked = dvfslocked;
709 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
710 VREF(dp);
711 vput(tdp);
712 VFS_UNLOCK_GIANT(tvfslocked);
713 vn_lock(dp,
714 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
715 LK_RETRY));
716 goto unionlookup;
717 }
718
719 if (error != EJUSTRETURN)
720 goto bad;
721 /*
722 * At this point, we know we're at the end of the
723 * pathname. If creating / renaming, we can consider
724 * allowing the file or directory to be created / renamed,
725 * provided we're not on a read-only filesystem.
726 */
727 if (rdonly) {
728 error = EROFS;
729 goto bad;
730 }
731 /* trailing slash only allowed for directories */
732 if ((cnp->cn_flags & TRAILINGSLASH) &&
733 !(cnp->cn_flags & WILLBEDIR)) {
734 error = ENOENT;
735 goto bad;
736 }
737 if ((cnp->cn_flags & LOCKPARENT) == 0)
738 VOP_UNLOCK(dp, 0);
739 /*
740 * We return with ni_vp NULL to indicate that the entry
741 * doesn't currently exist, leaving a pointer to the
742 * (possibly locked) directory vnode in ndp->ni_dvp.
743 */
744 if (cnp->cn_flags & SAVESTART) {
745 ndp->ni_startdir = ndp->ni_dvp;
746 VREF(ndp->ni_startdir);
747 }
748 goto success;
749 } else
750 cnp->cn_lkflags = lkflags_save;
751 #ifdef NAMEI_DIAGNOSTIC
752 printf("found\n");
753 #endif
754 /*
755 * Take into account any additional components consumed by
756 * the underlying filesystem.
757 */
758 if (cnp->cn_consume > 0) {
759 cnp->cn_nameptr += cnp->cn_consume;
760 ndp->ni_next += cnp->cn_consume;
761 ndp->ni_pathlen -= cnp->cn_consume;
762 cnp->cn_consume = 0;
763 }
764
765 dp = ndp->ni_vp;
766 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
767
768 /*
769 * Check to see if the vnode has been mounted on;
770 * if so find the root of the mounted filesystem.
771 */
772 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
773 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
774 if (vfs_busy(mp, 0))
775 continue;
776 vput(dp);
777 VFS_UNLOCK_GIANT(vfslocked);
778 vfslocked = VFS_LOCK_GIANT(mp);
779 if (dp != ndp->ni_dvp)
780 vput(ndp->ni_dvp);
781 else
782 vrele(ndp->ni_dvp);
783 VFS_UNLOCK_GIANT(dvfslocked);
784 dvfslocked = 0;
785 vref(vp_crossmp);
786 ndp->ni_dvp = vp_crossmp;
787 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags),
788 &tdp);
789 vfs_unbusy(mp);
790 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
791 panic("vp_crossmp exclusively locked or reclaimed");
792 if (error) {
793 dpunlocked = 1;
794 goto bad2;
795 }
796 ndp->ni_vp = dp = tdp;
797 }
798
799 /*
800 * Check for symbolic link
801 */
802 if ((dp->v_type == VLNK) &&
803 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
804 *ndp->ni_next == '/')) {
805 cnp->cn_flags |= ISSYMLINK;
806 if (dp->v_iflag & VI_DOOMED) {
807 /*
808 * We can't know whether the directory was mounted with
809 * NOSYMFOLLOW, so we can't follow safely.
810 */
811 error = ENOENT;
812 goto bad2;
813 }
814 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
815 error = EACCES;
816 goto bad2;
817 }
818 /*
819 * Symlink code always expects an unlocked dvp.
820 */
821 if (ndp->ni_dvp != ndp->ni_vp) {
822 VOP_UNLOCK(ndp->ni_dvp, 0);
823 ni_dvp_unlocked = 1;
824 }
825 goto success;
826 }
827
828 nextname:
829 /*
830 * Not a symbolic link that we will follow. Continue with the
831 * next component if there is any; otherwise, we're done.
832 */
833 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
834 ("lookup: invalid path state."));
835 if (*ndp->ni_next == '/') {
836 cnp->cn_nameptr = ndp->ni_next;
837 while (*cnp->cn_nameptr == '/') {
838 cnp->cn_nameptr++;
839 ndp->ni_pathlen--;
840 }
841 if (ndp->ni_dvp != dp)
842 vput(ndp->ni_dvp);
843 else
844 vrele(ndp->ni_dvp);
845 VFS_UNLOCK_GIANT(dvfslocked);
846 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
847 vfslocked = 0;
848 goto dirloop;
849 }
850 /*
851 * If we're processing a path with a trailing slash,
852 * check that the end result is a directory.
853 */
854 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
855 error = ENOTDIR;
856 goto bad2;
857 }
858 /*
859 * Disallow directory write attempts on read-only filesystems.
860 */
861 if (rdonly &&
862 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
863 error = EROFS;
864 goto bad2;
865 }
866 if (cnp->cn_flags & SAVESTART) {
867 ndp->ni_startdir = ndp->ni_dvp;
868 VREF(ndp->ni_startdir);
869 }
870 if (!wantparent) {
871 ni_dvp_unlocked = 2;
872 if (ndp->ni_dvp != dp)
873 vput(ndp->ni_dvp);
874 else
875 vrele(ndp->ni_dvp);
876 VFS_UNLOCK_GIANT(dvfslocked);
877 dvfslocked = 0;
878 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
879 VOP_UNLOCK(ndp->ni_dvp, 0);
880 ni_dvp_unlocked = 1;
881 }
882
883 if (cnp->cn_flags & AUDITVNODE1)
884 AUDIT_ARG_VNODE1(dp);
885 else if (cnp->cn_flags & AUDITVNODE2)
886 AUDIT_ARG_VNODE2(dp);
887
888 if ((cnp->cn_flags & LOCKLEAF) == 0)
889 VOP_UNLOCK(dp, 0);
890 success:
891 /*
892 * Because of lookup_shared we may have the vnode shared locked, but
893 * the caller may want it to be exclusively locked.
894 */
895 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
896 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
897 vn_lock(dp, LK_UPGRADE | LK_RETRY);
898 if (dp->v_iflag & VI_DOOMED) {
899 error = ENOENT;
900 goto bad2;
901 }
902 }
903 if (vfslocked && dvfslocked)
904 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
905 if (vfslocked || dvfslocked)
906 ndp->ni_cnd.cn_flags |= GIANTHELD;
907 return (0);
908
909 bad2:
910 if (ni_dvp_unlocked != 2) {
911 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
912 vput(ndp->ni_dvp);
913 else
914 vrele(ndp->ni_dvp);
915 }
916 bad:
917 if (!dpunlocked)
918 vput(dp);
919 VFS_UNLOCK_GIANT(vfslocked);
920 VFS_UNLOCK_GIANT(dvfslocked);
921 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
922 ndp->ni_vp = NULL;
923 return (error);
924 }
925
926 /*
927 * relookup - lookup a path name component
928 * Used by lookup to re-acquire things.
929 */
930 int
931 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
932 {
933 struct vnode *dp = 0; /* the directory we are searching */
934 int wantparent; /* 1 => wantparent or lockparent flag */
935 int rdonly; /* lookup read-only flag bit */
936 int error = 0;
937
938 KASSERT(cnp->cn_flags & ISLASTCN,
939 ("relookup: Not given last component."));
940 /*
941 * Setup: break out flag bits into variables.
942 */
943 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
944 KASSERT(wantparent, ("relookup: parent not wanted."));
945 rdonly = cnp->cn_flags & RDONLY;
946 cnp->cn_flags &= ~ISSYMLINK;
947 dp = dvp;
948 cnp->cn_lkflags = LK_EXCLUSIVE;
949 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
950
951 /*
952 * Search a new directory.
953 *
954 * The last component of the filename is left accessible via
955 * cnp->cn_nameptr for callers that need the name. Callers needing
956 * the name set the SAVENAME flag. When done, they assume
957 * responsibility for freeing the pathname buffer.
958 */
959 #ifdef NAMEI_DIAGNOSTIC
960 printf("{%s}: ", cnp->cn_nameptr);
961 #endif
962
963 /*
964 * Check for "" which represents the root directory after slash
965 * removal.
966 */
967 if (cnp->cn_nameptr[0] == '\0') {
968 /*
969 * Support only LOOKUP for "/" because lookup()
970 * can't succeed for CREATE, DELETE and RENAME.
971 */
972 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
973 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
974
975 if (!(cnp->cn_flags & LOCKLEAF))
976 VOP_UNLOCK(dp, 0);
977 *vpp = dp;
978 /* XXX This should probably move to the top of function. */
979 if (cnp->cn_flags & SAVESTART)
980 panic("lookup: SAVESTART");
981 return (0);
982 }
983
984 if (cnp->cn_flags & ISDOTDOT)
985 panic ("relookup: lookup on dot-dot");
986
987 /*
988 * We now have a segment name to search for, and a directory to search.
989 */
990 #ifdef NAMEI_DIAGNOSTIC
991 vprint("search in:", dp);
992 #endif
993 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
994 KASSERT(*vpp == NULL, ("leaf should be empty"));
995 if (error != EJUSTRETURN)
996 goto bad;
997 /*
998 * If creating and at end of pathname, then can consider
999 * allowing file to be created.
1000 */
1001 if (rdonly) {
1002 error = EROFS;
1003 goto bad;
1004 }
1005 /* ASSERT(dvp == ndp->ni_startdir) */
1006 if (cnp->cn_flags & SAVESTART)
1007 VREF(dvp);
1008 if ((cnp->cn_flags & LOCKPARENT) == 0)
1009 VOP_UNLOCK(dp, 0);
1010 /*
1011 * We return with ni_vp NULL to indicate that the entry
1012 * doesn't currently exist, leaving a pointer to the
1013 * (possibly locked) directory vnode in ndp->ni_dvp.
1014 */
1015 return (0);
1016 }
1017
1018 dp = *vpp;
1019
1020 /*
1021 * Disallow directory write attempts on read-only filesystems.
1022 */
1023 if (rdonly &&
1024 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1025 if (dvp == dp)
1026 vrele(dvp);
1027 else
1028 vput(dvp);
1029 error = EROFS;
1030 goto bad;
1031 }
1032 /*
1033 * Set the parent lock/ref state to the requested state.
1034 */
1035 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1036 if (wantparent)
1037 VOP_UNLOCK(dvp, 0);
1038 else
1039 vput(dvp);
1040 } else if (!wantparent)
1041 vrele(dvp);
1042 /*
1043 * Check for symbolic link
1044 */
1045 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1046 ("relookup: symlink found.\n"));
1047
1048 /* ASSERT(dvp == ndp->ni_startdir) */
1049 if (cnp->cn_flags & SAVESTART)
1050 VREF(dvp);
1051
1052 if ((cnp->cn_flags & LOCKLEAF) == 0)
1053 VOP_UNLOCK(dp, 0);
1054 return (0);
1055 bad:
1056 vput(dp);
1057 *vpp = NULL;
1058 return (error);
1059 }
1060
1061 /*
1062 * Free data allocated by namei(); see namei(9) for details.
1063 */
1064 void
1065 NDFREE(struct nameidata *ndp, const u_int flags)
1066 {
1067 int unlock_dvp;
1068 int unlock_vp;
1069
1070 unlock_dvp = 0;
1071 unlock_vp = 0;
1072
1073 if (!(flags & NDF_NO_FREE_PNBUF) &&
1074 (ndp->ni_cnd.cn_flags & HASBUF)) {
1075 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1076 ndp->ni_cnd.cn_flags &= ~HASBUF;
1077 }
1078 if (!(flags & NDF_NO_VP_UNLOCK) &&
1079 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1080 unlock_vp = 1;
1081 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1082 if (unlock_vp) {
1083 vput(ndp->ni_vp);
1084 unlock_vp = 0;
1085 } else
1086 vrele(ndp->ni_vp);
1087 ndp->ni_vp = NULL;
1088 }
1089 if (unlock_vp)
1090 VOP_UNLOCK(ndp->ni_vp, 0);
1091 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1092 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1093 ndp->ni_dvp != ndp->ni_vp)
1094 unlock_dvp = 1;
1095 if (!(flags & NDF_NO_DVP_RELE) &&
1096 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1097 if (unlock_dvp) {
1098 vput(ndp->ni_dvp);
1099 unlock_dvp = 0;
1100 } else
1101 vrele(ndp->ni_dvp);
1102 ndp->ni_dvp = NULL;
1103 }
1104 if (unlock_dvp)
1105 VOP_UNLOCK(ndp->ni_dvp, 0);
1106 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1107 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1108 vrele(ndp->ni_startdir);
1109 ndp->ni_startdir = NULL;
1110 }
1111 }
1112
1113 /*
1114 * Determine if there is a suitable alternate filename under the specified
1115 * prefix for the specified path. If the create flag is set, then the
1116 * alternate prefix will be used so long as the parent directory exists.
1117 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1118 * files under /compat/linux for example. The chosen path (whether under
1119 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1120 * to by pathbuf. The caller is responsible for free'ing the buffer from
1121 * the M_TEMP bucket if one is returned.
1122 */
1123 int
1124 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1125 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1126 {
1127 struct nameidata nd, ndroot;
1128 char *ptr, *buf, *cp;
1129 size_t len, sz;
1130 int error;
1131
1132 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1133 *pathbuf = buf;
1134
1135 /* Copy the prefix into the new pathname as a starting point. */
1136 len = strlcpy(buf, prefix, MAXPATHLEN);
1137 if (len >= MAXPATHLEN) {
1138 *pathbuf = NULL;
1139 free(buf, M_TEMP);
1140 return (EINVAL);
1141 }
1142 sz = MAXPATHLEN - len;
1143 ptr = buf + len;
1144
1145 /* Append the filename to the prefix. */
1146 if (pathseg == UIO_SYSSPACE)
1147 error = copystr(path, ptr, sz, &len);
1148 else
1149 error = copyinstr(path, ptr, sz, &len);
1150
1151 if (error) {
1152 *pathbuf = NULL;
1153 free(buf, M_TEMP);
1154 return (error);
1155 }
1156
1157 /* Only use a prefix with absolute pathnames. */
1158 if (*ptr != '/') {
1159 error = EINVAL;
1160 goto keeporig;
1161 }
1162
1163 if (dirfd != AT_FDCWD) {
1164 /*
1165 * We want the original because the "prefix" is
1166 * included in the already opened dirfd.
1167 */
1168 bcopy(ptr, buf, len);
1169 return (0);
1170 }
1171
1172 /*
1173 * We know that there is a / somewhere in this pathname.
1174 * Search backwards for it, to find the file's parent dir
1175 * to see if it exists in the alternate tree. If it does,
1176 * and we want to create a file (cflag is set). We don't
1177 * need to worry about the root comparison in this case.
1178 */
1179
1180 if (create) {
1181 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1182 *cp = '\0';
1183
1184 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1185 error = namei(&nd);
1186 *cp = '/';
1187 if (error != 0)
1188 goto keeporig;
1189 } else {
1190 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1191
1192 error = namei(&nd);
1193 if (error != 0)
1194 goto keeporig;
1195
1196 /*
1197 * We now compare the vnode of the prefix to the one
1198 * vnode asked. If they resolve to be the same, then we
1199 * ignore the match so that the real root gets used.
1200 * This avoids the problem of traversing "../.." to find the
1201 * root directory and never finding it, because "/" resolves
1202 * to the emulation root directory. This is expensive :-(
1203 */
1204 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1205 td);
1206
1207 /* We shouldn't ever get an error from this namei(). */
1208 error = namei(&ndroot);
1209 if (error == 0) {
1210 if (nd.ni_vp == ndroot.ni_vp)
1211 error = ENOENT;
1212
1213 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1214 vrele(ndroot.ni_vp);
1215 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1216 }
1217 }
1218
1219 NDFREE(&nd, NDF_ONLY_PNBUF);
1220 vrele(nd.ni_vp);
1221 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1222
1223 keeporig:
1224 /* If there was an error, use the original path name. */
1225 if (error)
1226 bcopy(ptr, buf, len);
1227 return (error);
1228 }
Cache object: 67d500585f73807385ee60475fc275cb
|