1/* $NetBSD: vfs_getcwd.c,v 1.50 2014/02/07 15:29:22 hannken Exp $ */
2
3/*-
4 * Copyright (c) 1999 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Bill Sommerfeld.
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 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32#include <sys/cdefs.h>
33__KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c,v 1.50 2014/02/07 15:29:22 hannken Exp $");
34
35#include <sys/param.h>
36#include <sys/systm.h>
37#include <sys/namei.h>
38#include <sys/filedesc.h>
39#include <sys/kernel.h>
40#include <sys/file.h>
41#include <sys/stat.h>
42#include <sys/vnode.h>
43#include <sys/mount.h>
44#include <sys/proc.h>
45#include <sys/uio.h>
46#include <sys/kmem.h>
47#include <sys/dirent.h>
48#include <sys/kauth.h>
49
50#include <ufs/ufs/dir.h> /* XXX only for DIRBLKSIZ */
51
52#include <sys/syscallargs.h>
53
54/*
55 * Vnode variable naming conventions in this file:
56 *
57 * rvp: the current root we're aiming towards.
58 * lvp, *lvpp: the "lower" vnode
59 * uvp, *uvpp: the "upper" vnode.
60 *
61 * Since all the vnodes we're dealing with are directories, and the
62 * lookups are going *up* in the filesystem rather than *down*, the
63 * usual "pvp" (parent) or "dvp" (directory) naming conventions are
64 * too confusing.
65 */
66
67/*
68 * XXX Will infinite loop in certain cases if a directory read reliably
69 * returns EINVAL on last block.
70 * XXX is EINVAL the right thing to return if a directory is malformed?
71 */
72
73/*
74 * XXX Untested vs. mount -o union; probably does the wrong thing.
75 */
76
77/*
78 * Find parent vnode of *lvpp, return in *uvpp
79 *
80 * If we care about the name, scan it looking for name of directory
81 * entry pointing at lvp.
82 *
83 * Place the name in the buffer which starts at bufp, immediately
84 * before *bpp, and move bpp backwards to point at the start of it.
85 *
86 * On entry, *lvpp is a locked vnode reference; on exit, it is vput and NULL'ed
87 * On exit, *uvpp is either NULL or is a locked vnode reference.
88 */
89static int
90getcwd_scandir(struct vnode **lvpp, struct vnode **uvpp, char **bpp,
91 char *bufp, struct lwp *l)
92{
93 int error = 0;
94 int eofflag;
95 off_t off;
96 int tries;
97 struct uio uio;
98 struct iovec iov;
99 char *dirbuf = NULL;
100 int dirbuflen;
101 ino_t fileno;
102 struct vattr va;
103 struct vnode *uvp = NULL;
104 struct vnode *lvp = *lvpp;
105 kauth_cred_t cred = l->l_cred;
106 struct componentname cn;
107 int len, reclen;
108 tries = 0;
109
110 /*
111 * If we want the filename, get some info we need while the
112 * current directory is still locked.
113 */
114 if (bufp != NULL) {
115 error = VOP_GETATTR(lvp, &va, cred);
116 if (error) {
117 vput(lvp);
118 *lvpp = NULL;
119 *uvpp = NULL;
120 return error;
121 }
122 }
123
124 /*
125 * Ok, we have to do it the hard way..
126 * Next, get parent vnode using lookup of ..
127 */
128 cn.cn_nameiop = LOOKUP;
129 cn.cn_flags = ISLASTCN | ISDOTDOT | RDONLY;
130 cn.cn_cred = cred;
131 cn.cn_nameptr = "..";
132 cn.cn_namelen = 2;
133 cn.cn_consume = 0;
134
135 /* At this point, lvp is locked */
136 error = VOP_LOOKUP(lvp, uvpp, &cn);
137 vput(lvp);
138 if (error) {
139 *lvpp = NULL;
140 *uvpp = NULL;
141 return error;
142 }
143 uvp = *uvpp;
144 /* Now lvp is unlocked, try to lock uvp */
145 error = vn_lock(uvp, LK_EXCLUSIVE);
146 if (error) {
147 *lvpp = NULL;
148 *uvpp = NULL;
149 return error;
150 }
151
152 /* If we don't care about the pathname, we're done */
153 if (bufp == NULL) {
154 *lvpp = NULL;
155 return 0;
156 }
157
158 fileno = va.va_fileid;
159
160 /* I guess UFS_DIRBLKSIZ is a good guess at a good size to use? */
161 dirbuflen = UFS_DIRBLKSIZ;
162 if (dirbuflen < va.va_blocksize)
163 dirbuflen = va.va_blocksize;
164 dirbuf = kmem_alloc(dirbuflen, KM_SLEEP);
165
166#if 0
167unionread:
168#endif
169 off = 0;
170 do {
171 /* call VOP_READDIR of parent */
172 iov.iov_base = dirbuf;
173 iov.iov_len = dirbuflen;
174
175 uio.uio_iov = &iov;
176 uio.uio_iovcnt = 1;
177 uio.uio_offset = off;
178 uio.uio_resid = dirbuflen;
179 uio.uio_rw = UIO_READ;
180 UIO_SETUP_SYSSPACE(&uio);
181
182 eofflag = 0;
183
184 error = VOP_READDIR(uvp, &uio, cred, &eofflag, 0, 0);
185
186 off = uio.uio_offset;
187
188 /*
189 * Try again if NFS tosses its cookies.
190 * XXX this can still loop forever if the directory is busted
191 * such that the second or subsequent page of it always
192 * returns EINVAL
193 */
194 if ((error == EINVAL) && (tries < 3)) {
195 off = 0;
196 tries++;
197 continue; /* once more, with feeling */
198 }
199
200 if (!error) {
201 char *cpos;
202 struct dirent *dp;
203
204 cpos = dirbuf;
205 tries = 0;
206
207 /* scan directory page looking for matching vnode */
208 for (len = (dirbuflen - uio.uio_resid); len > 0;
209 len -= reclen) {
210 dp = (struct dirent *) cpos;
211 reclen = dp->d_reclen;
212
213 /* check for malformed directory.. */
214 if (reclen < _DIRENT_MINSIZE(dp)) {
215 error = EINVAL;
216 goto out;
217 }
218 /*
219 * XXX should perhaps do VOP_LOOKUP to
220 * check that we got back to the right place,
221 * but getting the locking games for that
222 * right would be heinous.
223 */
224 if ((dp->d_type != DT_WHT) &&
225 (dp->d_fileno == fileno)) {
226 char *bp = *bpp;
227
228 bp -= dp->d_namlen;
229 if (bp <= bufp) {
230 error = ERANGE;
231 goto out;
232 }
233 memcpy(bp, dp->d_name, dp->d_namlen);
234 error = 0;
235 *bpp = bp;
236 goto out;
237 }
238 cpos += reclen;
239 }
240 } else
241 goto out;
242 } while (!eofflag);
243#if 0
244 /*
245 * Deal with mount -o union, which unions only the
246 * root directory of the mount.
247 */
248 if ((uvp->v_vflag & VV_ROOT) &&
249 (uvp->v_mount->mnt_flag & MNT_UNION)) {
250 struct vnode *tvp = uvp;
251
252 uvp = uvp->v_mount->mnt_vnodecovered;
253 vput(tvp);
254 vref(uvp);
255 *uvpp = uvp;
256 vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
257 goto unionread;
258 }
259#endif
260 error = ENOENT;
261
262out:
263 *lvpp = NULL;
264 kmem_free(dirbuf, dirbuflen);
265 return error;
266}
267
268/*
269 * Look in the vnode-to-name reverse cache to see if
270 * we can find things the easy way.
271 *
272 * XXX vget failure path is untested.
273 *
274 * On entry, *lvpp is a locked vnode reference.
275 * On exit, one of the following is the case:
276 * 0) Both *lvpp and *uvpp are NULL and failure is returned.
277 * 1) *uvpp is NULL, *lvpp remains locked and -1 is returned (cache miss)
278 * 2) *uvpp is a locked vnode reference, *lvpp is vput and NULL'ed
279 * and 0 is returned (cache hit)
280 */
281
282static int
283getcwd_getcache(struct vnode **lvpp, struct vnode **uvpp, char **bpp,
284 char *bufp)
285{
286 struct vnode *lvp, *uvp = NULL;
287 int error;
288
289 lvp = *lvpp;
290
291 /*
292 * This returns 0 on a cache hit, -1 on a clean cache miss,
293 * or an errno on other failure.
294 */
295 error = cache_revlookup(lvp, uvpp, bpp, bufp);
296 if (error) {
297 if (error != -1) {
298 vput(lvp);
299 *lvpp = NULL;
300 *uvpp = NULL;
301 }
302 return error;
303 }
304 uvp = *uvpp;
305
306 /*
307 * Since we're going up, we have to release the current lock
308 * before we take the parent lock.
309 */
310
311 VOP_UNLOCK(lvp);
312 vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
313 vrele(lvp);
314 *lvpp = NULL;
315
316 return error;
317}
318
319/*
320 * common routine shared by sys___getcwd() and vn_isunder()
321 */
322
323int
324getcwd_common(struct vnode *lvp, struct vnode *rvp, char **bpp, char *bufp,
325 int limit, int flags, struct lwp *l)
326{
327 struct cwdinfo *cwdi = l->l_proc->p_cwdi;
328 kauth_cred_t cred = l->l_cred;
329 struct vnode *uvp = NULL;
330 char *bp = NULL;
331 int error;
332 int perms = VEXEC;
333
334 error = 0;
335 if (rvp == NULL) {
336 rvp = cwdi->cwdi_rdir;
337 if (rvp == NULL)
338 rvp = rootvnode;
339 }
340
341 vref(rvp);
342 vref(lvp);
343
344 /*
345 * Error handling invariant:
346 * Before a `goto out':
347 * lvp is either NULL, or locked and held.
348 * uvp is either NULL, or locked and held.
349 */
350
351 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
352 if (bufp)
353 bp = *bpp;
354
355 /*
356 * this loop will terminate when one of the following happens:
357 * - we hit the root
358 * - getdirentries or lookup fails
359 * - we run out of space in the buffer.
360 */
361 if (lvp == rvp) {
362 if (bp)
363 *(--bp) = '/';
364 goto out;
365 }
366 do {
367 /*
368 * access check here is optional, depending on
369 * whether or not caller cares.
370 */
371 if (flags & GETCWD_CHECK_ACCESS) {
372 error = VOP_ACCESS(lvp, perms, cred);
373 if (error)
374 goto out;
375 perms = VEXEC|VREAD;
376 }
377
378 /*
379 * step up if we're a covered vnode..
380 */
381 while (lvp->v_vflag & VV_ROOT) {
382 struct vnode *tvp;
383
384 if (lvp == rvp)
385 goto out;
386
387 tvp = lvp;
388 lvp = lvp->v_mount->mnt_vnodecovered;
389 vput(tvp);
390 /*
391 * hodie natus est radici frater
392 */
393 if (lvp == NULL) {
394 error = ENOENT;
395 goto out;
396 }
397 vref(lvp);
398 error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
399 if (error != 0) {
400 vrele(lvp);
401 lvp = NULL;
402 goto out;
403 }
404 }
405 /*
406 * Look in the name cache; if that fails, look in the
407 * directory..
408 */
409 error = getcwd_getcache(&lvp, &uvp, &bp, bufp);
410 if (error == -1) {
411 if (lvp->v_type != VDIR) {
412 error = ENOTDIR;
413 goto out;
414 }
415 error = getcwd_scandir(&lvp, &uvp, &bp, bufp, l);
416 }
417 if (error)
418 goto out;
419#if DIAGNOSTIC
420 if (lvp != NULL)
421 panic("getcwd: oops, forgot to null lvp");
422 if (bufp && (bp <= bufp)) {
423 panic("getcwd: oops, went back too far");
424 }
425#endif
426 if (bp)
427 *(--bp) = '/';
428 lvp = uvp;
429 uvp = NULL;
430 limit--;
431 } while ((lvp != rvp) && (limit > 0));
432
433out:
434 if (bpp)
435 *bpp = bp;
436 if (uvp)
437 vput(uvp);
438 if (lvp)
439 vput(lvp);
440 vrele(rvp);
441 return error;
442}
443
444/*
445 * Check if one directory can be found inside another in the directory
446 * hierarchy.
447 *
448 * Intended to be used in chroot, chdir, fchdir, etc., to ensure that
449 * chroot() actually means something.
450 */
451int
452vn_isunder(struct vnode *lvp, struct vnode *rvp, struct lwp *l)
453{
454 int error;
455
456 error = getcwd_common(lvp, rvp, NULL, NULL, MAXPATHLEN / 2, 0, l);
457
458 if (!error)
459 return 1;
460 else
461 return 0;
462}
463
464/*
465 * Returns true if proc p1's root directory equal to or under p2's
466 * root directory.
467 *
468 * Intended to be used from ptrace/procfs sorts of things.
469 */
470
471int
472proc_isunder(struct proc *p1, struct lwp *l2)
473{
474 struct vnode *r1 = p1->p_cwdi->cwdi_rdir;
475 struct vnode *r2 = l2->l_proc->p_cwdi->cwdi_rdir;
476
477 if (r1 == NULL)
478 return (r2 == NULL);
479 else if (r2 == NULL)
480 return 1;
481 else
482 return vn_isunder(r1, r2, l2);
483}
484
485/*
486 * Find pathname of process's current directory.
487 *
488 * Use vfs vnode-to-name reverse cache; if that fails, fall back
489 * to reading directory contents.
490 */
491
492int
493sys___getcwd(struct lwp *l, const struct sys___getcwd_args *uap, register_t *retval)
494{
495 /* {
496 syscallarg(char *) bufp;
497 syscallarg(size_t) length;
498 } */
499
500 int error;
501 char *path;
502 char *bp, *bend;
503 int len = SCARG(uap, length);
504 int lenused;
505 struct cwdinfo *cwdi;
506
507 if (len > MAXPATHLEN * 4)
508 len = MAXPATHLEN * 4;
509 else if (len < 2)
510 return ERANGE;
511
512 path = kmem_alloc(len, KM_SLEEP);
513 if (!path)
514 return ENOMEM;
515
516 bp = &path[len];
517 bend = bp;
518 *(--bp) = '\0';
519
520 /*
521 * 5th argument here is "max number of vnodes to traverse".
522 * Since each entry takes up at least 2 bytes in the output buffer,
523 * limit it to N/2 vnodes for an N byte buffer.
524 */
525 cwdi = l->l_proc->p_cwdi;
526 rw_enter(&cwdi->cwdi_lock, RW_READER);
527 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path,
528 len/2, GETCWD_CHECK_ACCESS, l);
529 rw_exit(&cwdi->cwdi_lock);
530
531 if (error)
532 goto out;
533 lenused = bend - bp;
534 *retval = lenused;
535 /* put the result into user buffer */
536 error = copyout(bp, SCARG(uap, bufp), lenused);
537
538out:
539 kmem_free(path, len);
540 return error;
541}
542
543/*
544 * Try to find a pathname for a vnode. Since there is no mapping
545 * vnode -> parent directory, this needs the NAMECACHE_ENTER_REVERSE
546 * option to work (to make cache_revlookup succeed). Caller holds a
547 * reference to the vnode.
548 */
549int
550vnode_to_path(char *path, size_t len, struct vnode *vp, struct lwp *curl,
551 struct proc *p)
552{
553 struct proc *curp = curl->l_proc;
554 int error, lenused, elen;
555 char *bp, *bend;
556 struct vnode *dvp;
557
558 KASSERT(vp->v_usecount > 0);
559
560 bp = bend = &path[len];
561 *(--bp) = '\0';
562
563 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
564 if (error != 0)
565 return error;
566 error = cache_revlookup(vp, &dvp, &bp, path);
567 VOP_UNLOCK(vp);
568 if (error != 0)
569 return (error == -1 ? ENOENT : error);
570
571 *(--bp) = '/';
572 error = getcwd_common(dvp, NULL, &bp, path, len / 2,
573 GETCWD_CHECK_ACCESS, curl);
574 vrele(dvp);
575
576 /*
577 * Strip off emulation path for emulated processes looking at
578 * the maps file of a process of the same emulation. (Won't
579 * work if /emul/xxx is a symlink..)
580 */
581 if (curp->p_emul == p->p_emul && curp->p_emul->e_path != NULL) {
582 elen = strlen(curp->p_emul->e_path);
583 if (!strncmp(bp, curp->p_emul->e_path, elen))
584 bp = &bp[elen];
585 }
586
587 lenused = bend - bp;
588
589 memcpy(path, bp, lenused);
590 path[lenused] = 0;
591
592 return 0;
593}
594