xref: /netbsd-src/sys/kern/vfs_getcwd.c (revision e94a5d02693120d4ad9d909e488894e9fcf0eb76)
1 /*	$NetBSD: vfs_getcwd.c,v 1.63 2024/12/07 02:27:38 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2020 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 1.63 2024/12/07 02:27:38 riastradh Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/types.h>
37 
38 #include <sys/dirent.h>
39 #include <sys/file.h>
40 #include <sys/filedesc.h>
41 #include <sys/kauth.h>
42 #include <sys/kernel.h>
43 #include <sys/kmem.h>
44 #include <sys/mount.h>
45 #include <sys/namei.h>
46 #include <sys/proc.h>
47 #include <sys/sdt.h>
48 #include <sys/stat.h>
49 #include <sys/syscallargs.h>
50 #include <sys/systm.h>
51 #include <sys/uio.h>
52 #include <sys/vnode.h>
53 
54 #include <ufs/ufs/dir.h>	/* XXX only for DIRBLKSIZ */
55 
56 
57 /*
58  * Vnode variable naming conventions in this file:
59  *
60  * rvp: the current root we're aiming towards.
61  * lvp, *lvpp: the "lower" vnode
62  * uvp, *uvpp: the "upper" vnode.
63  *
64  * Since all the vnodes we're dealing with are directories, and the
65  * lookups are going *up* in the filesystem rather than *down*, the
66  * usual "pvp" (parent) or "dvp" (directory) naming conventions are
67  * too confusing.
68  */
69 
70 /*
71  * XXX Will infinite loop in certain cases if a directory read reliably
72  *	returns EINVAL on last block.
73  * XXX is EINVAL the right thing to return if a directory is malformed?
74  */
75 
76 /*
77  * XXX Untested vs. mount -o union; probably does the wrong thing.
78  */
79 
80 /*
81  * Find parent vnode of *lvpp, return in *uvpp
82  *
83  * If we care about the name, scan it looking for name of directory
84  * entry pointing at lvp.
85  *
86  * Place the name in the buffer which starts at bufp, immediately
87  * before *bpp, and move bpp backwards to point at the start of it.
88  *
89  * On entry, *lvpp is a locked vnode reference; on exit, it is vput and NULL'ed
90  * On exit, *uvpp is either NULL or is a locked vnode reference.
91  */
92 static int
93 getcwd_scandir(struct vnode *lvp, struct vnode **uvpp, char **bpp,
94     char *bufp, struct lwp *l)
95 {
96 	int     error = 0;
97 	int     eofflag;
98 	off_t   off;
99 	int     tries;
100 	struct uio uio;
101 	struct iovec iov;
102 	char   *dirbuf = NULL;
103 	int	dirbuflen;
104 	ino_t   fileno;
105 	struct vattr va;
106 	struct vnode *uvp = NULL;
107 	kauth_cred_t cred = l->l_cred;
108 	struct componentname cn;
109 	int len, reclen;
110 	tries = 0;
111 
112 	/* Need exclusive for UFS VOP_GETATTR (itimes) & VOP_LOOKUP. */
113 	KASSERT(VOP_ISLOCKED(lvp) == LK_EXCLUSIVE);
114 
115 	/*
116 	 * If we want the filename, get some info we need while the
117 	 * current directory is still locked.
118 	 */
119 	if (bufp != NULL) {
120 		error = VOP_GETATTR(lvp, &va, cred);
121 		if (error) {
122 			VOP_UNLOCK(lvp);
123 			*uvpp = NULL;
124 			return error;
125 		}
126 	}
127 
128 	/*
129 	 * Ok, we have to do it the hard way..
130 	 * Next, get parent vnode using lookup of ..
131 	 */
132 	cn.cn_nameiop = LOOKUP;
133 	cn.cn_flags = ISLASTCN | ISDOTDOT | RDONLY;
134 	cn.cn_cred = cred;
135 	cn.cn_nameptr = "..";
136 	cn.cn_namelen = 2;
137 
138 	/* At this point, lvp is locked  */
139 	error = VOP_LOOKUP(lvp, uvpp, &cn);
140 	VOP_UNLOCK(lvp);
141 	if (error) {
142 		*uvpp = NULL;
143 		return error;
144 	}
145 	uvp = *uvpp;
146 	/* If we don't care about the pathname, we're done */
147 	if (bufp == NULL) {
148 		return 0;
149 	}
150 
151 	fileno = va.va_fileid;
152 
153 	/* I guess UFS_DIRBLKSIZ is a good guess at a good size to use? */
154 	dirbuflen = UFS_DIRBLKSIZ;
155 	if (dirbuflen < va.va_blocksize)
156 		dirbuflen = va.va_blocksize;
157 	dirbuf = kmem_alloc(dirbuflen, KM_SLEEP);
158 
159 	/* Now lvp is unlocked, try to lock uvp */
160 	error = vn_lock(uvp, LK_SHARED);
161 	if (error) {
162 		vrele(uvp);
163 		*uvpp = NULL;
164 		return error;
165 	}
166 
167 #if 0
168 unionread:
169 #endif
170 	off = 0;
171 	do {
172 		/* call VOP_READDIR of parent */
173 		iov.iov_base = dirbuf;
174 		iov.iov_len = dirbuflen;
175 
176 		uio.uio_iov = &iov;
177 		uio.uio_iovcnt = 1;
178 		uio.uio_offset = off;
179 		uio.uio_resid = dirbuflen;
180 		uio.uio_rw = UIO_READ;
181 		UIO_SETUP_SYSSPACE(&uio);
182 
183 		eofflag = 0;
184 
185 		error = VOP_READDIR(uvp, &uio, cred, &eofflag, 0, 0);
186 
187 		off = uio.uio_offset;
188 
189 		/*
190 		 * Try again if NFS tosses its cookies.
191 		 * XXX this can still loop forever if the directory is busted
192 		 * such that the second or subsequent page of it always
193 		 * returns EINVAL
194 		 */
195 		if ((error == EINVAL) && (tries < 3)) {
196 			off = 0;
197 			tries++;
198 			continue;	/* once more, with feeling */
199 		}
200 
201 		if (!error) {
202 			char   *cpos;
203 			struct dirent *dp;
204 
205 			cpos = dirbuf;
206 			tries = 0;
207 
208 			/* scan directory page looking for matching vnode */
209 			for (len = (dirbuflen - uio.uio_resid); len > 0;
210 			    len -= reclen) {
211 				dp = (struct dirent *) cpos;
212 				reclen = dp->d_reclen;
213 
214 				/* check for malformed directory.. */
215 				if (reclen < _DIRENT_MINSIZE(dp) ||
216 				    reclen > len) {
217 					error = SET_ERROR(EINVAL);
218 					goto out;
219 				}
220 				/*
221 				 * XXX should perhaps do VOP_LOOKUP to
222 				 * check that we got back to the right place,
223 				 * but getting the locking games for that
224 				 * right would be heinous.
225 				 */
226 				if ((dp->d_type != DT_WHT) &&
227 				    (dp->d_fileno == fileno)) {
228 					char *bp = *bpp;
229 
230 					bp -= dp->d_namlen;
231 					if (bp <= bufp) {
232 						error = SET_ERROR(ERANGE);
233 						goto out;
234 					}
235 					memcpy(bp, dp->d_name, dp->d_namlen);
236 					error = 0;
237 					*bpp = bp;
238 					goto out;
239 				}
240 				cpos += reclen;
241 			}
242 		} else
243 			goto out;
244 	} while (!eofflag);
245 #if 0
246 	/*
247 	 * Deal with mount -o union, which unions only the
248 	 * root directory of the mount.
249 	 */
250 	if ((uvp->v_vflag & VV_ROOT) &&
251 	    (uvp->v_mount->mnt_flag & MNT_UNION)) {
252 		struct vnode *tvp = uvp;
253 
254 		uvp = uvp->v_mount->mnt_vnodecovered;
255 		vput(tvp);
256 		vref(uvp);
257 		*uvpp = uvp;
258 		vn_lock(uvp, LK_SHARED | LK_RETRY);
259 		goto unionread;
260 	}
261 #endif
262 	error = SET_ERROR(ENOENT);
263 
264 out:
265 	VOP_UNLOCK(uvp);
266 	kmem_free(dirbuf, dirbuflen);
267 	return error;
268 }
269 
270 /*
271  * common routine shared by sys___getcwd() and vn_isunder()
272  */
273 int
274 getcwd_common(struct vnode *lvp, struct vnode *rvp, char **bpp, char *bufp,
275     int limit, int flags, struct lwp *l)
276 {
277 	struct cwdinfo *cwdi = l->l_proc->p_cwdi;
278 	kauth_cred_t cred = l->l_cred;
279 	struct vnode *uvp = NULL;
280 	char *bp = NULL;
281 	int error;
282 	accmode_t accmode = VEXEC;
283 
284 	error = 0;
285 	if (rvp == NULL) {
286 		rvp = cwdi->cwdi_rdir;
287 		if (rvp == NULL)
288 			rvp = rootvnode;
289 	}
290 
291 	vref(rvp);
292 	vref(lvp);
293 
294 	/*
295 	 * Error handling invariant:
296 	 * Before a `goto out':
297 	 *	lvp is either NULL, or held.
298 	 *	uvp is either NULL, or held.
299 	 */
300 
301 	if (bufp)
302 		bp = *bpp;
303 
304 	/*
305 	 * this loop will terminate when one of the following happens:
306 	 *	- we hit the root
307 	 *	- getdirentries or lookup fails
308 	 *	- we run out of space in the buffer.
309 	 */
310 	if (lvp == rvp) {
311 		if (bp)
312 			*(--bp) = '/';
313 		goto out;
314 	}
315 	do {
316 		/*
317 		 * access check here is optional, depending on
318 		 * whether or not caller cares.
319 		 */
320 		int chkaccess = (flags & GETCWD_CHECK_ACCESS);
321 		bool locked = false;
322 
323 		/*
324 		 * step up if we're a covered vnode..
325 		 * check access on the first vnode only.
326 		 */
327 		if (lvp->v_vflag & VV_ROOT) {
328 			vn_lock(lvp, LK_SHARED | LK_RETRY);
329 			if (chkaccess) {
330 				error = VOP_ACCESS(lvp, accmode, cred);
331 				if (error) {
332 					VOP_UNLOCK(lvp);
333 					goto out;
334 				}
335 				chkaccess = 0;
336 			}
337 			while (lvp->v_vflag & VV_ROOT) {
338 				struct vnode *tvp;
339 
340 				if (lvp == rvp) {
341 					VOP_UNLOCK(lvp);
342 					goto out;
343 				}
344 
345 				tvp = lvp->v_mount->mnt_vnodecovered;
346 				/*
347 				 * hodie natus est radici frater
348 				 */
349 				if (tvp == NULL) {
350 					VOP_UNLOCK(lvp);
351 					error = SET_ERROR(ENOENT);
352 					goto out;
353 				}
354 				vref(tvp);
355 				vput(lvp);
356 				lvp = tvp;
357 				if (lvp->v_vflag & VV_ROOT)
358 					vn_lock(lvp, LK_SHARED | LK_RETRY);
359 			}
360 		}
361 
362 		/* Do we need to check access to the directory? */
363 		if (chkaccess && !cache_have_id(lvp)) {
364 			/*
365 			 * Need exclusive for UFS VOP_GETATTR (itimes)
366 			 * & VOP_LOOKUP.
367 			 */
368 			vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
369 			error = VOP_ACCESS(lvp, accmode, cred);
370 			if (error) {
371 				VOP_UNLOCK(lvp);
372 				goto out;
373 			}
374 			chkaccess = 0;
375 			locked = true;
376 		}
377 
378 		/*
379 		 * Look in the name cache; if that fails, look in the
380 		 * directory..
381 		 */
382 		error = cache_revlookup(lvp, &uvp, &bp, bufp, chkaccess,
383 		    accmode);
384 		if (error == -1) {
385 			if (!locked) {
386 				locked = true;
387 				vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
388 			}
389 			if (lvp->v_type != VDIR) {
390 				VOP_UNLOCK(lvp);
391 				error = SET_ERROR(ENOTDIR);
392 				goto out;
393 			}
394 			error = getcwd_scandir(lvp, &uvp, &bp, bufp, l);
395 			/* lvp now unlocked */
396 		} else if (locked) {
397 			VOP_UNLOCK(lvp);
398 		}
399 		if (error)
400 			goto out;
401 #if DIAGNOSTIC
402 		if (bufp && (bp <= bufp)) {
403 			panic("getcwd: oops, went back too far");
404 		}
405 #endif
406 		accmode = VEXEC | VREAD;
407 		if (bp)
408 			*(--bp) = '/';
409 		vrele(lvp);
410 		lvp = uvp;
411 		uvp = NULL;
412 		limit--;
413 	} while ((lvp != rvp) && (limit > 0));
414 
415 out:
416 	if (bpp)
417 		*bpp = bp;
418 	if (uvp)
419 		vrele(uvp);
420 	if (lvp)
421 		vrele(lvp);
422 	vrele(rvp);
423 	return error;
424 }
425 
426 /*
427  * Check if one directory can be found inside another in the directory
428  * hierarchy.
429  *
430  * Intended to be used in chroot, chdir, fchdir, etc., to ensure that
431  * chroot() actually means something.
432  */
433 int
434 vn_isunder(struct vnode *lvp, struct vnode *rvp, struct lwp *l)
435 {
436 	int error;
437 
438 	error = getcwd_common(lvp, rvp, NULL, NULL, MAXPATHLEN / 2, 0, l);
439 
440 	if (!error)
441 		return 1;
442 	else
443 		return 0;
444 }
445 
446 /*
447  * Returns true if proc p1's root directory equal to or under p2's
448  * root directory.
449  *
450  * Intended to be used from ptrace/procfs sorts of things.
451  */
452 
453 int
454 proc_isunder(struct proc *p1, struct lwp *l2)
455 {
456 	struct vnode *r1 = p1->p_cwdi->cwdi_rdir;
457 	struct vnode *r2 = l2->l_proc->p_cwdi->cwdi_rdir;
458 
459 	if (r1 == NULL)
460 		return (r2 == NULL);
461 	else if (r2 == NULL)
462 		return 1;
463 	else
464 		return vn_isunder(r1, r2, l2);
465 }
466 
467 /*
468  * Find pathname of process's current directory.
469  *
470  * Use vfs vnode-to-name reverse cache; if that fails, fall back
471  * to reading directory contents.
472  */
473 
474 int
475 sys___getcwd(struct lwp *l, const struct sys___getcwd_args *uap,
476     register_t *retval)
477 {
478 	/* {
479 		syscallarg(char *) bufp;
480 		syscallarg(size_t) length;
481 	} */
482 
483 	int     error;
484 	char   *path;
485 	char   *bp, *bend;
486 	int     len = SCARG(uap, length);
487 	int	lenused;
488 	struct	cwdinfo *cwdi;
489 
490 	if (len > MAXPATHLEN * 4)
491 		len = MAXPATHLEN * 4;
492 	else if (len < 2)
493 		return SET_ERROR(ERANGE);
494 
495 	path = kmem_alloc(len, KM_SLEEP);
496 	bp = &path[len];
497 	bend = bp;
498 	*(--bp) = '\0';
499 
500 	/*
501 	 * 5th argument here is "max number of vnodes to traverse".
502 	 * Since each entry takes up at least 2 bytes in the output buffer,
503 	 * limit it to N/2 vnodes for an N byte buffer.
504 	 */
505 	cwdi = l->l_proc->p_cwdi;
506 	rw_enter(&cwdi->cwdi_lock, RW_READER);
507 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path,
508 	    len/2, GETCWD_CHECK_ACCESS, l);
509 	rw_exit(&cwdi->cwdi_lock);
510 
511 	if (error)
512 		goto out;
513 	lenused = bend - bp;
514 	*retval = lenused;
515 	/* put the result into user buffer */
516 	error = copyout(bp, SCARG(uap, bufp), lenused);
517 
518 out:
519 	kmem_free(path, len);
520 	return error;
521 }
522 
523 /*
524  * Try to find a pathname for a vnode.  Since there is no mapping vnode ->
525  * parent directory, this needs the namecache to succeed.  Caller holds a
526  * reference to the vnode.
527  */
528 int
529 vnode_to_path(char *path, size_t len, struct vnode *vp, struct lwp *curl,
530     struct proc *p)
531 {
532 	struct proc *curp = curl->l_proc;
533 	int error, lenused, elen;
534 	char *bp, *bend;
535 	struct vnode *dvp;
536 
537 	KASSERT(vrefcnt(vp) > 0);
538 
539 	bp = bend = &path[len];
540 	*(--bp) = '\0';
541 
542 	error = cache_revlookup(vp, &dvp, &bp, path, false, 0);
543 	if (error != 0)
544 		return (error == -1 ? SET_ERROR(ENOENT) : error);
545 
546 	*(--bp) = '/';
547 	error = getcwd_common(dvp, NULL, &bp, path, len / 2,
548 	    GETCWD_CHECK_ACCESS, curl);
549 	vrele(dvp);
550 	if (error != 0)
551 		return error;
552 
553 	/*
554 	 * Strip off emulation path for emulated processes looking at
555 	 * the maps file of a process of the same emulation. (Won't
556 	 * work if /emul/xxx is a symlink..)
557 	 */
558 	if (curp->p_emul == p->p_emul && curp->p_emul->e_path != NULL) {
559 		elen = strlen(curp->p_emul->e_path);
560 		if (!strncmp(bp, curp->p_emul->e_path, elen))
561 			bp = &bp[elen];
562 	}
563 
564 	lenused = bend - bp;
565 
566 	memcpy(path, bp, lenused);
567 	path[lenused] = '\0';
568 
569 	return 0;
570 }
571