vfs_getcwd.c revision 1.15.10.1 1 /* $NetBSD: vfs_getcwd.c,v 1.15.10.1 2004/03/15 05:22:25 jmc 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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c,v 1.15.10.1 2004/03/15 05:22:25 jmc Exp $");
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/namei.h>
45 #include <sys/filedesc.h>
46 #include <sys/kernel.h>
47 #include <sys/file.h>
48 #include <sys/stat.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 #include <sys/proc.h>
52 #include <sys/uio.h>
53 #include <sys/malloc.h>
54 #include <sys/dirent.h>
55 #include <ufs/ufs/dir.h> /* XXX only for DIRBLKSIZ */
56
57 #include <sys/syscallargs.h>
58
59 static int
60 getcwd_scandir __P((struct vnode **, struct vnode **,
61 char **, char *, struct proc *));
62 static int
63 getcwd_getcache __P((struct vnode **, struct vnode **,
64 char **, char *));
65 int
66 getcwd_common __P((struct vnode *, struct vnode *,
67 char **, char *, int, int, struct proc *));
68
69 #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4)
70
71 /*
72 * Vnode variable naming conventions in this file:
73 *
74 * rvp: the current root we're aiming towards.
75 * lvp, *lvpp: the "lower" vnode
76 * uvp, *uvpp: the "upper" vnode.
77 *
78 * Since all the vnodes we're dealing with are directories, and the
79 * lookups are going *up* in the filesystem rather than *down*, the
80 * usual "pvp" (parent) or "dvp" (directory) naming conventions are
81 * too confusing.
82 */
83
84 /*
85 * XXX Will infinite loop in certain cases if a directory read reliably
86 * returns EINVAL on last block.
87 * XXX is EINVAL the right thing to return if a directory is malformed?
88 */
89
90 /*
91 * XXX Untested vs. mount -o union; probably does the wrong thing.
92 */
93
94 /*
95 * Find parent vnode of *lvpp, return in *uvpp
96 *
97 * If we care about the name, scan it looking for name of directory
98 * entry pointing at lvp.
99 *
100 * Place the name in the buffer which starts at bufp, immediately
101 * before *bpp, and move bpp backwards to point at the start of it.
102 *
103 * On entry, *lvpp is a locked vnode reference; on exit, it is vput and NULL'ed
104 * On exit, *uvpp is either NULL or is a locked vnode reference.
105 */
106 static int
107 getcwd_scandir(lvpp, uvpp, bpp, bufp, p)
108 struct vnode **lvpp;
109 struct vnode **uvpp;
110 char **bpp;
111 char *bufp;
112 struct proc *p;
113 {
114 int error = 0;
115 int eofflag;
116 off_t off;
117 int tries;
118 struct uio uio;
119 struct iovec iov;
120 char *dirbuf = NULL;
121 int dirbuflen;
122 ino_t fileno;
123 struct vattr va;
124 struct vnode *uvp = NULL;
125 struct vnode *lvp = *lvpp;
126 struct componentname cn;
127 int len, reclen;
128 tries = 0;
129
130 /*
131 * If we want the filename, get some info we need while the
132 * current directory is still locked.
133 */
134 if (bufp != NULL) {
135 error = VOP_GETATTR(lvp, &va, p->p_ucred, p);
136 if (error) {
137 vput(lvp);
138 *lvpp = NULL;
139 *uvpp = NULL;
140 return error;
141 }
142 }
143
144 /*
145 * Ok, we have to do it the hard way..
146 * Next, get parent vnode using lookup of ..
147 */
148 cn.cn_nameiop = LOOKUP;
149 cn.cn_flags = ISLASTCN | ISDOTDOT | RDONLY;
150 cn.cn_proc = p;
151 cn.cn_cred = p->p_ucred;
152 cn.cn_pnbuf = NULL;
153 cn.cn_nameptr = "..";
154 cn.cn_namelen = 2;
155 cn.cn_hash = 0;
156 cn.cn_consume = 0;
157
158 /*
159 * At this point, lvp is locked and will be unlocked by the lookup.
160 * On successful return, *uvpp will be locked
161 */
162 error = VOP_LOOKUP(lvp, uvpp, &cn);
163 if (error) {
164 vput(lvp);
165 *lvpp = NULL;
166 *uvpp = NULL;
167 return error;
168 }
169 uvp = *uvpp;
170
171 /* If we don't care about the pathname, we're done */
172 if (bufp == NULL) {
173 vrele(lvp);
174 *lvpp = NULL;
175 return 0;
176 }
177
178 fileno = va.va_fileid;
179
180 dirbuflen = DIRBLKSIZ;
181 if (dirbuflen < va.va_blocksize)
182 dirbuflen = va.va_blocksize;
183 dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
184
185 #if 0
186 unionread:
187 #endif
188 off = 0;
189 do {
190 /* call VOP_READDIR of parent */
191 iov.iov_base = dirbuf;
192 iov.iov_len = dirbuflen;
193
194 uio.uio_iov = &iov;
195 uio.uio_iovcnt = 1;
196 uio.uio_offset = off;
197 uio.uio_resid = dirbuflen;
198 uio.uio_segflg = UIO_SYSSPACE;
199 uio.uio_rw = UIO_READ;
200 uio.uio_procp = p;
201
202 eofflag = 0;
203
204 error = VOP_READDIR(uvp, &uio, p->p_ucred, &eofflag, 0, 0);
205
206 off = uio.uio_offset;
207
208 /*
209 * Try again if NFS tosses its cookies.
210 * XXX this can still loop forever if the directory is busted
211 * such that the second or subsequent page of it always
212 * returns EINVAL
213 */
214 if ((error == EINVAL) && (tries < 3)) {
215 off = 0;
216 tries++;
217 continue; /* once more, with feeling */
218 }
219
220 if (!error) {
221 char *cpos;
222 struct dirent *dp;
223
224 cpos = dirbuf;
225 tries = 0;
226
227 /* scan directory page looking for matching vnode */
228 for (len = (dirbuflen - uio.uio_resid); len > 0; len -= reclen) {
229 dp = (struct dirent *) cpos;
230 reclen = dp->d_reclen;
231
232 /* check for malformed directory.. */
233 if (reclen < DIRENT_MINSIZE) {
234 error = EINVAL;
235 goto out;
236 }
237 /*
238 * XXX should perhaps do VOP_LOOKUP to
239 * check that we got back to the right place,
240 * but getting the locking games for that
241 * right would be heinous.
242 */
243 if ((dp->d_type != DT_WHT) &&
244 (dp->d_fileno == fileno)) {
245 char *bp = *bpp;
246 bp -= dp->d_namlen;
247
248 if (bp <= bufp) {
249 error = ERANGE;
250 goto out;
251 }
252 memcpy(bp, dp->d_name, dp->d_namlen);
253 error = 0;
254 *bpp = bp;
255 goto out;
256 }
257 cpos += reclen;
258 }
259 } else
260 goto out;
261 } while (!eofflag);
262 #if 0
263 /*
264 * Deal with mount -o union, which unions only the
265 * root directory of the mount.
266 */
267 if ((uvp->v_flag & VROOT) &&
268 (uvp->v_mount->mnt_flag & MNT_UNION)) {
269 struct vnode *tvp = uvp;
270 uvp = uvp->v_mount->mnt_vnodecovered;
271 vput(tvp);
272 VREF(uvp);
273 *uvpp = uvp;
274 error = vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
275 if (error != 0) {
276 vrele(uvp);
277 *uvpp = uvp = NULL;
278 goto out;
279 }
280 goto unionread;
281 }
282 #endif
283 error = ENOENT;
284
285 out:
286 vrele(lvp);
287 *lvpp = NULL;
288 free(dirbuf, M_TEMP);
289 return error;
290 }
291
292 /*
293 * Look in the vnode-to-name reverse cache to see if
294 * we can find things the easy way.
295 *
296 * XXX vget failure path is untested.
297 *
298 * On entry, *lvpp is a locked vnode reference.
299 * On exit, one of the following is the case:
300 * 0) Both *lvpp and *uvpp are NULL and failure is returned.
301 * 1) *uvpp is NULL, *lvpp remains locked and -1 is returned (cache miss)
302 * 2) *uvpp is a locked vnode reference, *lvpp is vput and NULL'ed
303 * and 0 is returned (cache hit)
304 */
305
306 static int
307 getcwd_getcache(lvpp, uvpp, bpp, bufp)
308 struct vnode **lvpp, **uvpp;
309 char **bpp;
310 char *bufp;
311 {
312 struct vnode *lvp, *uvp = NULL;
313 char *obp = *bpp;
314 int error;
315 int vpid;
316
317 lvp = *lvpp;
318
319 /*
320 * This returns 0 on a cache hit, -1 on a clean cache miss,
321 * or an errno on other failure.
322 */
323 error = cache_revlookup(lvp, uvpp, bpp, bufp);
324 if (error) {
325 if (error != -1) {
326 vput(lvp);
327 *lvpp = NULL;
328 *uvpp = NULL;
329 }
330 return error;
331 }
332 uvp = *uvpp;
333 vpid = uvp->v_id;
334
335 /*
336 * Since we're going up, we have to release the current lock
337 * before we take the parent lock.
338 */
339
340 VOP_UNLOCK(lvp, 0);
341
342 error = vget(uvp, LK_EXCLUSIVE | LK_RETRY);
343 if (error != 0)
344 *uvpp = NULL;
345 /*
346 * Verify that vget succeeded, and check that vnode capability
347 * didn't change while we were waiting for the lock.
348 */
349 if (error || (vpid != uvp->v_id)) {
350 /*
351 * Oops, we missed. If the vget failed, or the
352 * capability changed, try to get our lock back; if
353 * that works, tell caller to try things the hard way,
354 * otherwise give up.
355 */
356 if (!error) vput(uvp);
357 *uvpp = NULL;
358 *bpp = obp;
359
360 error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
361
362 if (!error)
363 return -1;
364 }
365 vrele(lvp);
366 *lvpp = NULL;
367
368 return error;
369 }
370
371 /*
372 * common routine shared by sys___getcwd() and vn_isunder()
373 */
374
375 #define GETCWD_CHECK_ACCESS 0x0001
376
377 int
378 getcwd_common (lvp, rvp, bpp, bufp, limit, flags, p)
379 struct vnode *lvp;
380 struct vnode *rvp;
381 char **bpp;
382 char *bufp;
383 int limit;
384 int flags;
385 struct proc *p;
386 {
387 struct cwdinfo *cwdi = p->p_cwdi;
388 struct vnode *uvp = NULL;
389 char *bp = NULL;
390 int error;
391 int perms = VEXEC;
392
393 if (rvp == NULL) {
394 rvp = cwdi->cwdi_rdir;
395 if (rvp == NULL)
396 rvp = rootvnode;
397 }
398
399 VREF(rvp);
400 VREF(lvp);
401
402 /*
403 * Error handling invariant:
404 * Before a `goto out':
405 * lvp is either NULL, or locked and held.
406 * uvp is either NULL, or locked and held.
407 */
408
409 error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
410 if (error) {
411 vrele(lvp);
412 lvp = NULL;
413 goto out;
414 }
415 if (bufp)
416 bp = *bpp;
417 /*
418 * this loop will terminate when one of the following happens:
419 * - we hit the root
420 * - getdirentries or lookup fails
421 * - we run out of space in the buffer.
422 */
423 if (lvp == rvp) {
424 if (bp)
425 *(--bp) = '/';
426 goto out;
427 }
428 do {
429 if (lvp->v_type != VDIR) {
430 error = ENOTDIR;
431 goto out;
432 }
433
434 /*
435 * access check here is optional, depending on
436 * whether or not caller cares.
437 */
438 if (flags & GETCWD_CHECK_ACCESS) {
439 error = VOP_ACCESS(lvp, perms, p->p_ucred, p);
440 if (error)
441 goto out;
442 perms = VEXEC|VREAD;
443 }
444
445 /*
446 * step up if we're a covered vnode..
447 */
448 while (lvp->v_flag & VROOT) {
449 struct vnode *tvp;
450
451 if (lvp == rvp)
452 goto out;
453
454 tvp = lvp;
455 lvp = lvp->v_mount->mnt_vnodecovered;
456 vput(tvp);
457 /*
458 * hodie natus est radici frater
459 */
460 if (lvp == NULL) {
461 error = ENOENT;
462 goto out;
463 }
464 VREF(lvp);
465 error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
466 if (error != 0) {
467 vrele(lvp);
468 lvp = NULL;
469 goto out;
470 }
471 }
472 /*
473 * Look in the name cache; if that fails, look in the
474 * directory..
475 */
476 error = getcwd_getcache(&lvp, &uvp, &bp, bufp);
477 if (error == -1)
478 error = getcwd_scandir(&lvp, &uvp, &bp, bufp, p);
479 if (error)
480 goto out;
481 #if DIAGNOSTIC
482 if (lvp != NULL)
483 panic("getcwd: oops, forgot to null lvp");
484 if (bufp && (bp <= bufp)) {
485 panic("getcwd: oops, went back too far");
486 }
487 #endif
488 if (bp)
489 *(--bp) = '/';
490 lvp = uvp;
491 uvp = NULL;
492 limit--;
493 } while ((lvp != rvp) && (limit > 0));
494
495 out:
496 if (bpp)
497 *bpp = bp;
498 if (uvp)
499 vput(uvp);
500 if (lvp)
501 vput(lvp);
502 vrele(rvp);
503 return error;
504 }
505
506 /*
507 * Check if one directory can be found inside another in the directory
508 * hierarchy.
509 *
510 * Intended to be used in chroot, chdir, fchdir, etc., to ensure that
511 * chroot() actually means something.
512 */
513 int
514 vn_isunder(lvp, rvp, p)
515 struct vnode *lvp;
516 struct vnode *rvp;
517 struct proc *p;
518 {
519 int error;
520
521 error = getcwd_common (lvp, rvp, NULL, NULL, MAXPATHLEN/2, 0, p);
522
523 if (!error)
524 return 1;
525 else
526 return 0;
527 }
528
529 /*
530 * Returns true if proc p1's root directory equal to or under p2's
531 * root directory.
532 *
533 * Intended to be used from ptrace/procfs sorts of things.
534 */
535
536 int
537 proc_isunder (p1, p2)
538 struct proc *p1;
539 struct proc *p2;
540 {
541 struct vnode *r1 = p1->p_cwdi->cwdi_rdir;
542 struct vnode *r2 = p2->p_cwdi->cwdi_rdir;
543
544 if (r1 == NULL)
545 return (r2 == NULL);
546 else if (r2 == NULL)
547 return 1;
548 else
549 return vn_isunder(r1, r2, p2);
550 }
551
552 /*
553 * Find pathname of process's current directory.
554 *
555 * Use vfs vnode-to-name reverse cache; if that fails, fall back
556 * to reading directory contents.
557 */
558
559 int
560 sys___getcwd(p, v, retval)
561 struct proc *p;
562 void *v;
563 register_t *retval;
564 {
565 struct sys___getcwd_args /* {
566 syscallarg(char *) bufp;
567 syscallarg(size_t) length;
568 } */ *uap = v;
569
570 int error;
571 char *path;
572 char *bp, *bend;
573 int len = SCARG(uap, length);
574 int lenused;
575
576 if (len > MAXPATHLEN*4)
577 len = MAXPATHLEN*4;
578 else if (len < 2)
579 return ERANGE;
580
581 path = (char *)malloc(len, M_TEMP, M_WAITOK);
582 if (!path)
583 return ENOMEM;
584
585 bp = &path[len];
586 bend = bp;
587 *(--bp) = '\0';
588
589 /*
590 * 5th argument here is "max number of vnodes to traverse".
591 * Since each entry takes up at least 2 bytes in the output buffer,
592 * limit it to N/2 vnodes for an N byte buffer.
593 */
594 error = getcwd_common (p->p_cwdi->cwdi_cdir, NULL, &bp, path, len/2,
595 GETCWD_CHECK_ACCESS, p);
596
597 if (error)
598 goto out;
599 lenused = bend - bp;
600 *retval = lenused;
601 /* put the result into user buffer */
602 error = copyout(bp, SCARG(uap, bufp), lenused);
603
604 out:
605 free(path, M_TEMP);
606 return error;
607 }
608