Home | History | Annotate | Line # | Download | only in kern
vfs_getcwd.c revision 1.11
      1 /* $NetBSD: vfs_getcwd.c,v 1.11 1999/12/15 08:36:44 jdolecek 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/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/namei.h>
     42 #include <sys/filedesc.h>
     43 #include <sys/kernel.h>
     44 #include <sys/file.h>
     45 #include <sys/stat.h>
     46 #include <sys/vnode.h>
     47 #include <sys/mount.h>
     48 #include <sys/proc.h>
     49 #include <sys/uio.h>
     50 #include <sys/malloc.h>
     51 #include <sys/dirent.h>
     52 #include <ufs/ufs/dir.h>	/* XXX only for DIRBLKSIZ */
     53 
     54 #include <sys/syscallargs.h>
     55 
     56 static int
     57 getcwd_scandir __P((struct vnode **, struct vnode **,
     58     char **, char *, struct proc *));
     59 static int
     60 getcwd_getcache __P((struct vnode **, struct vnode **,
     61     char **, char *));
     62 static int
     63 getcwd_common __P((struct vnode *, struct vnode *,
     64 		   char **, char *, int, int, struct proc *));
     65 
     66 #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4)
     67 
     68 /*
     69  * Vnode variable naming conventions in this file:
     70  *
     71  * rvp: the current root we're aiming towards.
     72  * lvp, *lvpp: the "lower" vnode
     73  * uvp, *uvpp: the "upper" vnode.
     74  *
     75  * Since all the vnodes we're dealing with are directories, and the
     76  * lookups are going *up* in the filesystem rather than *down*, the
     77  * usual "pvp" (parent) or "dvp" (directory) naming conventions are
     78  * too confusing.
     79  */
     80 
     81 /*
     82  * XXX Will infinite loop in certain cases if a directory read reliably
     83  *	returns EINVAL on last block.
     84  * XXX is EINVAL the right thing to return if a directory is malformed?
     85  */
     86 
     87 /*
     88  * XXX Untested vs. mount -o union; probably does the wrong thing.
     89  */
     90 
     91 /*
     92  * Find parent vnode of *lvpp, return in *uvpp
     93  *
     94  * If we care about the name, scan it looking for name of directory
     95  * entry pointing at lvp.
     96  *
     97  * Place the name in the buffer which starts at bufp, immediately
     98  * before *bpp, and move bpp backwards to point at the start of it.
     99  *
    100  * On entry, *lvpp is a locked vnode reference; on exit, it is vput and NULL'ed
    101  * On exit, *uvpp is either NULL or is a locked vnode reference.
    102  */
    103 static int
    104 getcwd_scandir(lvpp, uvpp, bpp, bufp, p)
    105 	struct vnode **lvpp;
    106 	struct vnode **uvpp;
    107 	char **bpp;
    108 	char *bufp;
    109 	struct proc *p;
    110 {
    111 	int     error = 0;
    112 	int     eofflag;
    113 	off_t   off;
    114 	int     tries;
    115 	struct uio uio;
    116 	struct iovec iov;
    117 	char   *dirbuf = NULL;
    118 	int	dirbuflen;
    119 	ino_t   fileno;
    120 	struct vattr va;
    121 	struct vnode *uvp = NULL;
    122 	struct vnode *lvp = *lvpp;
    123 	struct componentname cn;
    124 	int len, reclen;
    125 	tries = 0;
    126 
    127 	/*
    128 	 * If we want the filename, get some info we need while the
    129 	 * current directory is still locked.
    130 	 */
    131 	if (bufp != NULL) {
    132 		error = VOP_GETATTR(lvp, &va, p->p_ucred, p);
    133 		if (error) {
    134 			vput(lvp);
    135 			*lvpp = NULL;
    136 			*uvpp = NULL;
    137 			return error;
    138 		}
    139 	}
    140 
    141 	/*
    142 	 * Ok, we have to do it the hard way..
    143 	 * Next, get parent vnode using lookup of ..
    144 	 */
    145 	cn.cn_nameiop = LOOKUP;
    146 	cn.cn_flags = ISLASTCN | ISDOTDOT | RDONLY;
    147 	cn.cn_proc = p;
    148 	cn.cn_cred = p->p_ucred;
    149 	cn.cn_pnbuf = NULL;
    150 	cn.cn_nameptr = "..";
    151 	cn.cn_namelen = 2;
    152 	cn.cn_hash = 0;
    153 	cn.cn_consume = 0;
    154 
    155 	/*
    156 	 * At this point, lvp is locked and will be unlocked by the lookup.
    157 	 * On successful return, *uvpp will be locked
    158 	 */
    159 	error = VOP_LOOKUP(lvp, uvpp, &cn);
    160 	if (error) {
    161 		vput(lvp);
    162 		*lvpp = NULL;
    163 		*uvpp = NULL;
    164 		return error;
    165 	}
    166 	uvp = *uvpp;
    167 
    168 	/* If we don't care about the pathname, we're done */
    169 	if (bufp == NULL) {
    170 		vrele(lvp);
    171 		*lvpp = NULL;
    172 		return 0;
    173 	}
    174 
    175 	fileno = va.va_fileid;
    176 
    177 	dirbuflen = DIRBLKSIZ;
    178 	if (dirbuflen < va.va_blocksize)
    179 		dirbuflen = va.va_blocksize;
    180 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
    181 
    182 #if 0
    183 unionread:
    184 #endif
    185 	off = 0;
    186 	do {
    187 		/* call VOP_READDIR of parent */
    188 		iov.iov_base = dirbuf;
    189 		iov.iov_len = dirbuflen;
    190 
    191 		uio.uio_iov = &iov;
    192 		uio.uio_iovcnt = 1;
    193 		uio.uio_offset = off;
    194 		uio.uio_resid = dirbuflen;
    195 		uio.uio_segflg = UIO_SYSSPACE;
    196 		uio.uio_rw = UIO_READ;
    197 		uio.uio_procp = p;
    198 
    199 		eofflag = 0;
    200 
    201 		error = VOP_READDIR(uvp, &uio, p->p_ucred, &eofflag, 0, 0);
    202 
    203 		off = uio.uio_offset;
    204 
    205 		/*
    206 		 * Try again if NFS tosses its cookies.
    207 		 * XXX this can still loop forever if the directory is busted
    208 		 * such that the second or subsequent page of it always
    209 		 * returns EINVAL
    210 		 */
    211 		if ((error == EINVAL) && (tries < 3)) {
    212 			off = 0;
    213 			tries++;
    214 			continue;	/* once more, with feeling */
    215 		}
    216 
    217 		if (!error) {
    218 			char   *cpos;
    219 			struct dirent *dp;
    220 
    221 			cpos = dirbuf;
    222 			tries = 0;
    223 
    224 			/* scan directory page looking for matching vnode */
    225 			for (len = (dirbuflen - uio.uio_resid); len > 0; len -= reclen) {
    226 				dp = (struct dirent *) cpos;
    227 				reclen = dp->d_reclen;
    228 
    229 				/* check for malformed directory.. */
    230 				if (reclen < DIRENT_MINSIZE) {
    231 					error = EINVAL;
    232 					goto out;
    233 				}
    234 				/*
    235 				 * XXX should perhaps do VOP_LOOKUP to
    236 				 * check that we got back to the right place,
    237 				 * but getting the locking games for that
    238 				 * right would be heinous.
    239 				 */
    240 				if ((dp->d_type != DT_WHT) &&
    241 				    (dp->d_fileno == fileno)) {
    242 					char *bp = *bpp;
    243 					bp -= dp->d_namlen;
    244 
    245 					if (bp <= bufp) {
    246 						error = ERANGE;
    247 						goto out;
    248 					}
    249 					memcpy(bp, dp->d_name, dp->d_namlen);
    250 					error = 0;
    251 					*bpp = bp;
    252 					goto out;
    253 				}
    254 				cpos += reclen;
    255 			}
    256 		}
    257 	} while (!eofflag);
    258 #if 0
    259 	/*
    260 	 * Deal with mount -o union, which unions only the
    261 	 * root directory of the mount.
    262 	 */
    263 	if ((uvp->v_flag & VROOT) &&
    264 	    (uvp->v_mount->mnt_flag & MNT_UNION)) {
    265 		struct vnode *tvp = uvp;
    266 		uvp = uvp->v_mount->mnt_vnodecovered;
    267 		vput(tvp);
    268 		VREF(uvp);
    269 		*uvpp = uvp;
    270 		error = vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
    271 		if (error != 0) {
    272 			vrele(uvp);
    273 			*uvpp = uvp = NULL;
    274 			goto out;
    275 		}
    276 		goto unionread;
    277 	}
    278 #endif
    279 	error = ENOENT;
    280 
    281 out:
    282 	vrele(lvp);
    283 	*lvpp = NULL;
    284 	free(dirbuf, M_TEMP);
    285 	return error;
    286 }
    287 
    288 /*
    289  * Look in the vnode-to-name reverse cache to see if
    290  * we can find things the easy way.
    291  *
    292  * XXX vget failure path is untested.
    293  *
    294  * On entry, *lvpp is a locked vnode reference.
    295  * On exit, one of the following is the case:
    296  *	0) Both *lvpp and *uvpp are NULL and failure is returned.
    297  * 	1) *uvpp is NULL, *lvpp remains locked and -1 is returned (cache miss)
    298  *      2) *uvpp is a locked vnode reference, *lvpp is vput and NULL'ed
    299  *	   and 0 is returned (cache hit)
    300  */
    301 
    302 static int
    303 getcwd_getcache(lvpp, uvpp, bpp, bufp)
    304 	struct vnode **lvpp, **uvpp;
    305 	char **bpp;
    306 	char *bufp;
    307 {
    308 	struct vnode *lvp, *uvp = NULL;
    309 	int error;
    310 	int vpid;
    311 
    312 	lvp = *lvpp;
    313 
    314 	/*
    315 	 * This returns 0 on a cache hit, -1 on a clean cache miss,
    316 	 * or an errno on other failure.
    317 	 */
    318 	error = cache_revlookup(lvp, uvpp, bpp, bufp);
    319 	if (error) {
    320 		if (error != -1) {
    321 			vput(lvp);
    322 			*lvpp = NULL;
    323 			*uvpp = NULL;
    324 		}
    325 		return error;
    326 	}
    327 	uvp = *uvpp;
    328 	vpid = uvp->v_id;
    329 
    330 	/*
    331 	 * Since we're going up, we have to release the current lock
    332 	 * before we take the parent lock.
    333 	 */
    334 
    335 	VOP_UNLOCK(lvp, 0);
    336 
    337 	error = vget(uvp, LK_EXCLUSIVE | LK_RETRY);
    338 	if (error != 0)
    339 		*uvpp = NULL;
    340 	/*
    341 	 * Verify that vget succeeded, and check that vnode capability
    342 	 * didn't change while we were waiting for the lock.
    343 	 */
    344 	if (error || (vpid != uvp->v_id)) {
    345 		/*
    346 		 * Oops, we missed.  If the vget failed, or the
    347 		 * capability changed, try to get our lock back; if
    348 		 * that works, tell caller to try things the hard way,
    349 		 * otherwise give up.
    350 		 */
    351 		if (!error) vput(uvp);
    352 		*uvpp = NULL;
    353 
    354 		error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
    355 
    356 		if (!error)
    357 			return -1;
    358 	}
    359 	vrele(lvp);
    360 	*lvpp = NULL;
    361 
    362 	return error;
    363 }
    364 
    365 /*
    366  * common routine shared by sys___getcwd() and vn_isunder()
    367  */
    368 
    369 #define GETCWD_CHECK_ACCESS 0x0001
    370 
    371 static int
    372 getcwd_common (lvp, rvp, bpp, bufp, limit, flags, p)
    373 	struct vnode *lvp;
    374 	struct vnode *rvp;
    375 	char **bpp;
    376 	char *bufp;
    377 	int limit;
    378 	int flags;
    379 	struct proc *p;
    380 {
    381 	struct cwdinfo *cwdi = p->p_cwdi;
    382 	struct vnode *uvp = NULL;
    383 	char *bp = NULL;
    384 	int error;
    385 	int perms = VEXEC;
    386 
    387 	if (rvp == NULL) {
    388 		rvp = cwdi->cwdi_rdir;
    389 		if (rvp == NULL)
    390 			rvp = rootvnode;
    391 	}
    392 
    393 	VREF(rvp);
    394 	VREF(lvp);
    395 
    396 	/*
    397 	 * Error handling invariant:
    398 	 * Before a `goto out':
    399 	 *	lvp is either NULL, or locked and held.
    400 	 *	uvp is either NULL, or locked and held.
    401 	 */
    402 
    403 	error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
    404 	if (error) {
    405 		vrele(lvp);
    406 		lvp = NULL;
    407 		goto out;
    408 	}
    409 	if (bufp)
    410 		bp = *bpp;
    411 	/*
    412 	 * this loop will terminate when one of the following happens:
    413 	 *	- we hit the root
    414 	 *	- getdirentries or lookup fails
    415 	 *	- we run out of space in the buffer.
    416 	 */
    417 	if (lvp == rvp) {
    418 		if (bp)
    419 			*(--bp) = '/';
    420 		goto out;
    421 	}
    422 	do {
    423 		if (lvp->v_type != VDIR) {
    424 			error = ENOTDIR;
    425 			goto out;
    426 		}
    427 
    428 		/*
    429 		 * access check here is optional, depending on
    430 		 * whether or not caller cares.
    431 		 */
    432 		if (flags & GETCWD_CHECK_ACCESS) {
    433 			error = VOP_ACCESS(lvp, perms, p->p_ucred, p);
    434 			if (error)
    435 				goto out;
    436 			perms = VEXEC|VREAD;
    437 		}
    438 
    439 		/*
    440 		 * step up if we're a covered vnode..
    441 		 */
    442 		while (lvp->v_flag & VROOT) {
    443 			struct vnode *tvp;
    444 
    445 			if (lvp == rvp)
    446 				goto out;
    447 
    448 			tvp = lvp;
    449 			lvp = lvp->v_mount->mnt_vnodecovered;
    450 			vput(tvp);
    451 			/*
    452 			 * hodie natus est radici frater
    453 			 */
    454 			if (lvp == NULL) {
    455 				error = ENOENT;
    456 				goto out;
    457 			}
    458 			VREF(lvp);
    459 			error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
    460 			if (error != 0) {
    461 				vrele(lvp);
    462 				lvp = NULL;
    463 				goto out;
    464 			}
    465 		}
    466 		/*
    467 		 * Look in the name cache; if that fails, look in the
    468 		 * directory..
    469 		 */
    470 		error = getcwd_getcache(&lvp, &uvp, &bp, bufp);
    471 		if (error == -1)
    472 			error = getcwd_scandir(&lvp, &uvp, &bp, bufp, p);
    473 		if (error)
    474 			goto out;
    475 #if DIAGNOSTIC
    476 		if (lvp != NULL)
    477 			panic("getcwd: oops, forgot to null lvp");
    478 		if (bufp && (bp <= bufp)) {
    479 			panic("getcwd: oops, went back too far");
    480 		}
    481 #endif
    482 		if (bp)
    483 			*(--bp) = '/';
    484 		lvp = uvp;
    485 		uvp = NULL;
    486 		limit--;
    487 	} while ((lvp != rvp) && (limit > 0));
    488 
    489 out:
    490 	if (bpp)
    491 		*bpp = bp;
    492 	if (uvp)
    493 		vput(uvp);
    494 	if (lvp)
    495 		vput(lvp);
    496 	vrele(rvp);
    497 	return error;
    498 }
    499 
    500 /*
    501  * Check if one directory can be found inside another in the directory
    502  * hierarchy.
    503  *
    504  * Intended to be used in chroot, chdir, fchdir, etc., to ensure that
    505  * chroot() actually means something.
    506  */
    507 int
    508 vn_isunder(lvp, rvp, p)
    509 	struct vnode *lvp;
    510 	struct vnode *rvp;
    511 	struct proc *p;
    512 {
    513 	int error;
    514 
    515 	error = getcwd_common (lvp, rvp, NULL, NULL, MAXPATHLEN/2, 0, p);
    516 
    517 	if (!error)
    518 		return 1;
    519 	else
    520 		return 0;
    521 }
    522 
    523 /*
    524  * Returns true if proc p1's root directory equal to or under p2's
    525  * root directory.
    526  *
    527  * Intended to be used from ptrace/procfs sorts of things.
    528  */
    529 
    530 int
    531 proc_isunder (p1, p2)
    532 	struct proc *p1;
    533 	struct proc *p2;
    534 {
    535 	struct vnode *r1 = p1->p_cwdi->cwdi_rdir;
    536 	struct vnode *r2 = p2->p_cwdi->cwdi_rdir;
    537 
    538 	if (r1 == NULL)
    539 		return (r2 == NULL);
    540 	else if (r2 == NULL)
    541 		return 1;
    542 	else
    543 		return vn_isunder(r1, r2, p2);
    544 }
    545 
    546 /*
    547  * Find pathname of process's current directory.
    548  *
    549  * Use vfs vnode-to-name reverse cache; if that fails, fall back
    550  * to reading directory contents.
    551  */
    552 
    553 int
    554 sys___getcwd(p, v, retval)
    555 	struct proc *p;
    556 	void   *v;
    557 	register_t *retval;
    558 {
    559 	register struct sys___getcwd_args /* {
    560 		syscallarg(char *) bufp;
    561 		syscallarg(size_t) length;
    562 	} */ *uap = v;
    563 
    564 	int     error;
    565 	char   *path;
    566 	char   *bp, *bend;
    567 	int     len = SCARG(uap, length);
    568 	int	lenused;
    569 
    570 	if (len > MAXPATHLEN*4)
    571 		len = MAXPATHLEN*4;
    572 	else if (len < 2)
    573 		return ERANGE;
    574 
    575 	path = (char *)malloc(len, M_TEMP, M_WAITOK);
    576 	if (!path)
    577 		return ENOMEM;
    578 
    579 	bp = &path[len];
    580 	bend = bp;
    581 	*(--bp) = '\0';
    582 
    583 	/*
    584 	 * 5th argument here is "max number of vnodes to traverse".
    585 	 * Since each entry takes up at least 2 bytes in the output buffer,
    586 	 * limit it to N/2 vnodes for an N byte buffer.
    587 	 */
    588 	error = getcwd_common (p->p_cwdi->cwdi_cdir, NULL, &bp, path, len/2,
    589 			       GETCWD_CHECK_ACCESS, p);
    590 
    591 	if (error)
    592 		goto out;
    593 	lenused = bend - bp;
    594 	*retval = lenused;
    595 	/* put the result into user buffer */
    596 	error = copyout(bp, SCARG(uap, bufp), lenused);
    597 
    598 out:
    599 	free(path, M_TEMP);
    600 	return error;
    601 }
    602