1 /* $OpenBSD: ext2fs_lookup.c,v 1.23 2007/06/17 20:15:25 jasper Exp $ */
2 /* $NetBSD: ext2fs_lookup.c,v 1.16 2000/08/03 20:29:26 thorpej Exp $ */
3
4 /*
5 * Modified for NetBSD 1.2E
6 * May 1997, Manuel Bouyer
7 * Laboratoire d'informatique de Paris VI
8 */
9 /*
10 * modified for Lites 1.1
11 *
12 * Aug 1995, Godmar Back (gback@cs.utah.edu)
13 * University of Utah, Department of Computer Science
14 */
15 /*
16 * Copyright (c) 1989, 1993
17 * The Regents of the University of California. All rights reserved.
18 * (c) UNIX System Laboratories, Inc.
19 * All or some portions of this file are derived from material licensed
20 * to the University of California by American Telephone and Telegraph
21 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
22 * the permission of UNIX System Laboratories, Inc.
23 *
24 * Redistribution and use in source and binary forms, with or without
25 * modification, are permitted provided that the following conditions
26 * are met:
27 * 1. Redistributions of source code must retain the above copyright
28 * notice, this list of conditions and the following disclaimer.
29 * 2. Redistributions in binary form must reproduce the above copyright
30 * notice, this list of conditions and the following disclaimer in the
31 * documentation and/or other materials provided with the distribution.
32 * 3. Neither the name of the University nor the names of its contributors
33 * may be used to endorse or promote products derived from this software
34 * without specific prior written permission.
35 *
36 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
37 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
39 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
40 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
41 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
42 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
43 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
44 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
45 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
46 * SUCH DAMAGE.
47 *
48 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
49 */
50
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/namei.h>
54 #include <sys/buf.h>
55 #include <sys/file.h>
56 #include <sys/mount.h>
57 #include <sys/vnode.h>
58 #include <sys/malloc.h>
59 #include <sys/dirent.h>
60
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/inode.h>
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/ufs_extern.h>
65
66 #include <ufs/ext2fs/ext2fs_extern.h>
67 #include <ufs/ext2fs/ext2fs_dir.h>
68 #include <ufs/ext2fs/ext2fs.h>
69
70 extern int dirchk;
71
72 static void ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
73 struct dirent *ffsdir);
74 static int ext2fs_dirbadentry(struct vnode *dp,
75 struct ext2fs_direct *de,
76 int entryoffsetinblock);
77
78 /*
79 * the problem that is tackled below is the fact that FFS
80 * includes the terminating zero on disk while EXT2FS doesn't
81 * this implies that we need to introduce some padding.
82 * For instance, a filename "sbin" has normally a reclen 12
83 * in EXT2, but 16 in FFS.
84 * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
85 * If it wasn't for that, the complete ufs code for directories would
86 * have worked w/o changes (except for the difference in DIRBLKSIZ)
87 */
88 static void
89 ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir)
90 {
91 memset(ffsdir, 0, sizeof(struct dirent));
92 ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
93 ffsdir->d_namlen = e2dir->e2d_namlen;
94
95 ffsdir->d_type = DT_UNKNOWN; /* don't know more here */
96 #ifdef DIAGNOSTIC
97 /*
98 * XXX Rigth now this can't happen, but if one day
99 * MAXNAMLEN != E2FS_MAXNAMLEN we should handle this more gracefully !
100 */
101 /* XXX: e2d_namlen is to small for such comparison
102 if (e2dir->e2d_namlen > MAXNAMLEN)
103 panic("ext2fs: e2dir->e2d_namlen");
104 */
105 #endif
106 strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
107
108 /* Godmar thinks: since e2dir->e2d_reclen can be big and means
109 nothing anyway, we compute our own reclen according to what
110 we think is right
111 */
112 ffsdir->d_reclen = DIRENT_SIZE(ffsdir);
113 }
114
115 /*
116 * Vnode op for reading directories.
117 *
118 * Convert the on-disk entries to <sys/dirent.h> entries.
119 * the problem is that the conversion will blow up some entries by four bytes,
120 * so it can't be done in place. This is too bad. Right now the conversion is
121 * done entry by entry, the converted entry is sent via uiomove.
122 *
123 * XXX allocate a buffer, convert as many entries as possible, then send
124 * the whole buffer to uiomove
125 */
126 int
127 ext2fs_readdir(void *v)
128 {
129 struct vop_readdir_args *ap = v;
130 struct uio *uio = ap->a_uio;
131 int error;
132 size_t e2fs_count, readcnt, entries;
133 struct vnode *vp = ap->a_vp;
134 struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
135
136 struct ext2fs_direct *dp;
137 struct dirent dstd;
138 struct uio auio;
139 struct iovec aiov;
140 caddr_t dirbuf;
141 off_t off = uio->uio_offset;
142 u_long *cookies = NULL;
143 int nc = 0, ncookies = 0;
144 int e2d_reclen;
145
146 if (vp->v_type != VDIR)
147 return (ENOTDIR);
148
149 e2fs_count = uio->uio_resid;
150 entries = (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize - 1);
151
152 /* Make sure we don't return partial entries. */
153 if (e2fs_count <= entries)
154 return (EINVAL);
155
156 e2fs_count -= entries;
157 auio = *uio;
158 auio.uio_iov = &aiov;
159 auio.uio_iovcnt = 1;
160 auio.uio_segflg = UIO_SYSSPACE;
161 aiov.iov_len = e2fs_count;
162 auio.uio_resid = e2fs_count;
163 MALLOC(dirbuf, caddr_t, e2fs_count, M_TEMP, M_WAITOK);
164 if (ap->a_ncookies) {
165 nc = ncookies = e2fs_count / 16;
166 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
167 *ap->a_cookies = cookies;
168 }
169 memset(dirbuf, 0, e2fs_count);
170 aiov.iov_base = dirbuf;
171
172 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
173 if (error == 0) {
174 readcnt = e2fs_count - auio.uio_resid;
175 for (dp = (struct ext2fs_direct *)dirbuf;
176 (char *)dp < (char *)dirbuf + readcnt; ) {
177 e2d_reclen = fs2h16(dp->e2d_reclen);
178 if (e2d_reclen == 0) {
179 error = EIO;
180 break;
181 }
182 ext2fs_dirconv2ffs(dp, &dstd);
183 if(dstd.d_reclen > uio->uio_resid) {
184 break;
185 }
186 if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
187 break;
188 }
189 off = off + e2d_reclen;
190 if (cookies != NULL) {
191 *cookies++ = off;
192 if (--ncookies <= 0){
193 break; /* out of cookies */
194 }
195 }
196 /* advance dp */
197 dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
198 }
199 /* we need to correct uio_offset */
200 uio->uio_offset = off;
201 }
202 FREE(dirbuf, M_TEMP);
203 *ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
204 if (ap->a_ncookies) {
205 if (error) {
206 free(*ap->a_cookies, M_TEMP);
207 *ap->a_ncookies = 0;
208 *ap->a_cookies = NULL;
209 } else
210 *ap->a_ncookies = nc - ncookies;
211 }
212 return (error);
213 }
214
215 /*
216 * Convert a component of a pathname into a pointer to a locked inode.
217 * This is a very central and rather complicated routine.
218 * If the file system is not maintained in a strict tree hierarchy,
219 * this can result in a deadlock situation (see comments in code below).
220 *
221 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
222 * on whether the name is to be looked up, created, renamed, or deleted.
223 * When CREATE, RENAME, or DELETE is specified, information usable in
224 * creating, renaming, or deleting a directory entry may be calculated.
225 * If flag has LOCKPARENT or'ed into it and the target of the pathname
226 * exists, lookup returns both the target and its parent directory locked.
227 * When creating or renaming and LOCKPARENT is specified, the target may
228 * not be ".". When deleting and LOCKPARENT is specified, the target may
229 * be "."., but the caller must check to ensure it does an vrele and vput
230 * instead of two vputs.
231 *
232 * Overall outline of ext2fs_lookup:
233 *
234 * check accessibility of directory
235 * look for name in cache, if found, then if at end of path
236 * and deleting or creating, drop it, else return name
237 * search for name in directory, to found or notfound
238 * notfound:
239 * if creating, return locked directory, leaving info on available slots
240 * else return error
241 * found:
242 * if at end of path and deleting, return information to allow delete
243 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
244 * inode and return info to allow rewrite
245 * if not at end, add name to cache; if at end and neither creating
246 * nor deleting, add name to cache
247 */
248 int
249 ext2fs_lookup(void *v)
250 {
251 struct vop_lookup_args *ap = v;
252 struct vnode *vdp; /* vnode for directory being searched */
253 struct inode *dp; /* inode for directory being searched */
254 struct buf *bp; /* a buffer of directory entries */
255 struct ext2fs_direct *ep; /* the current directory entry */
256 int entryoffsetinblock; /* offset of ep in bp's buffer */
257 enum {NONE, COMPACT, FOUND} slotstatus;
258 doff_t slotoffset; /* offset of area with free space */
259 int slotsize; /* size of area at slotoffset */
260 int slotfreespace; /* amount of space free in slot */
261 int slotneeded; /* size of the entry we're seeking */
262 int numdirpasses; /* strategy for directory search */
263 doff_t endsearch; /* offset to end directory search */
264 doff_t prevoff; /* prev entry dp->i_offset */
265 struct vnode *pdp; /* saved dp during symlink work */
266 struct vnode *tdp; /* returned by VFS_VGET */
267 doff_t enduseful; /* pointer past last used dir slot */
268 u_long bmask; /* block offset mask */
269 int lockparent; /* 1 => lockparent flag is set */
270 int wantparent; /* 1 => wantparent or lockparent flag */
271 int namlen, error;
272 struct vnode **vpp = ap->a_vpp;
273 struct componentname *cnp = ap->a_cnp;
274 struct ucred *cred = cnp->cn_cred;
275 int flags = cnp->cn_flags;
276 int nameiop = cnp->cn_nameiop;
277 struct proc *p = cnp->cn_proc;
278 int dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
279
280 bp = NULL;
281 slotoffset = -1;
282 *vpp = NULL;
283 vdp = ap->a_dvp;
284 dp = VTOI(vdp);
285 lockparent = flags & LOCKPARENT;
286 wantparent = flags & (LOCKPARENT|WANTPARENT);
287 /*
288 * Check accessiblity of directory.
289 */
290 if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
291 return (error);
292
293 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
294 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
295 return (EROFS);
296
297 /*
298 * We now have a segment name to search for, and a directory to search.
299 *
300 * Before tediously performing a linear scan of the directory,
301 * check the name cache to see if the directory/name pair
302 * we are looking for is known already.
303 */
304 if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
305 return (error);
306
307 /*
308 * Suppress search for slots unless creating
309 * file and at end of pathname, in which case
310 * we watch for a place to put the new file in
311 * case it doesn't already exist.
312 */
313 slotstatus = FOUND;
314 slotfreespace = slotsize = slotneeded = 0;
315 if ((nameiop == CREATE || nameiop == RENAME) &&
316 (flags & ISLASTCN)) {
317 slotstatus = NONE;
318 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
319 }
320
321 /*
322 * If there is cached information on a previous search of
323 * this directory, pick up where we last left off.
324 * We cache only lookups as these are the most common
325 * and have the greatest payoff. Caching CREATE has little
326 * benefit as it usually must search the entire directory
327 * to determine that the entry does not exist. Caching the
328 * location of the last DELETE or RENAME has not reduced
329 * profiling time and hence has been removed in the interest
330 * of simplicity.
331 */
332 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
333 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
334 dp->i_diroff >ext2fs_size(dp)) {
335 entryoffsetinblock = 0;
336 dp->i_offset = 0;
337 numdirpasses = 1;
338 } else {
339 dp->i_offset = dp->i_diroff;
340 if ((entryoffsetinblock = dp->i_offset & bmask) &&
341 (error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
342 NULL, &bp)))
343 return (error);
344 numdirpasses = 2;
345 }
346 prevoff = dp->i_offset;
347 endsearch = roundup(ext2fs_size(dp), dirblksize);
348 enduseful = 0;
349
350 searchloop:
351 while (dp->i_offset < endsearch) {
352 /*
353 * If necessary, get the next directory block.
354 */
355 if ((dp->i_offset & bmask) == 0) {
356 if (bp != NULL)
357 brelse(bp);
358 error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
359 NULL, &bp);
360 if (error != 0)
361 return (error);
362 entryoffsetinblock = 0;
363 }
364 /*
365 * If still looking for a slot, and at a dirblksize
366 * boundary, have to start looking for free space again.
367 */
368 if (slotstatus == NONE &&
369 (entryoffsetinblock & (dirblksize - 1)) == 0) {
370 slotoffset = -1;
371 slotfreespace = 0;
372 }
373 /*
374 * Get pointer to next entry.
375 * Full validation checks are slow, so we only check
376 * enough to insure forward progress through the
377 * directory. Complete checks can be run by patching
378 * "dirchk" to be true.
379 */
380 ep = (struct ext2fs_direct *)
381 ((char *)bp->b_data + entryoffsetinblock);
382 if (ep->e2d_reclen == 0 ||
383 (dirchk &&
384 ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
385 int i;
386 ufs_dirbad(dp, dp->i_offset, "mangled entry");
387 i = dirblksize -
388 (entryoffsetinblock & (dirblksize - 1));
389 dp->i_offset += i;
390 entryoffsetinblock += i;
391 continue;
392 }
393
394 /*
395 * If an appropriate sized slot has not yet been found,
396 * check to see if one is available. Also accumulate space
397 * in the current block so that we can determine if
398 * compaction is viable.
399 */
400 if (slotstatus != FOUND) {
401 int size = fs2h16(ep->e2d_reclen);
402
403 if (ep->e2d_ino != 0)
404 size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
405 if (size > 0) {
406 if (size >= slotneeded) {
407 slotstatus = FOUND;
408 slotoffset = dp->i_offset;
409 slotsize = fs2h16(ep->e2d_reclen);
410 } else if (slotstatus == NONE) {
411 slotfreespace += size;
412 if (slotoffset == -1)
413 slotoffset = dp->i_offset;
414 if (slotfreespace >= slotneeded) {
415 slotstatus = COMPACT;
416 slotsize = dp->i_offset +
417 fs2h16(ep->e2d_reclen) - slotoffset;
418 }
419 }
420 }
421 }
422
423 /*
424 * Check for a name match.
425 */
426 if (ep->e2d_ino) {
427 namlen = ep->e2d_namlen;
428 if (namlen == cnp->cn_namelen &&
429 !memcmp(cnp->cn_nameptr, ep->e2d_name,
430 (unsigned)namlen)) {
431 /*
432 * Save directory entry's inode number and
433 * reclen in ndp->ni_ufs area, and release
434 * directory buffer.
435 */
436 dp->i_ino = fs2h32(ep->e2d_ino);
437 dp->i_reclen = fs2h16(ep->e2d_reclen);
438 brelse(bp);
439 goto found;
440 }
441 }
442 prevoff = dp->i_offset;
443 dp->i_offset += fs2h16(ep->e2d_reclen);
444 entryoffsetinblock += fs2h16(ep->e2d_reclen);
445 if (ep->e2d_ino)
446 enduseful = dp->i_offset;
447 }
448 /* notfound: */
449 /*
450 * If we started in the middle of the directory and failed
451 * to find our target, we must check the beginning as well.
452 */
453 if (numdirpasses == 2) {
454 numdirpasses--;
455 dp->i_offset = 0;
456 endsearch = dp->i_diroff;
457 goto searchloop;
458 }
459 if (bp != NULL)
460 brelse(bp);
461 /*
462 * If creating, and at end of pathname and current
463 * directory has not been removed, then can consider
464 * allowing file to be created.
465 */
466 if ((nameiop == CREATE || nameiop == RENAME) &&
467 (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
468 /*
469 * Creation of files on a read-only mounted file system
470 * is pointless, so don't proceed any further.
471 */
472 if (vdp->v_mount->mnt_flag & MNT_RDONLY)
473 return (EROFS);
474 /*
475 * Access for write is interpreted as allowing
476 * creation of files in the directory.
477 */
478 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
479 return (error);
480 /*
481 * Return an indication of where the new directory
482 * entry should be put. If we didn't find a slot,
483 * then set dp->i_count to 0 indicating
484 * that the new slot belongs at the end of the
485 * directory. If we found a slot, then the new entry
486 * can be put in the range from dp->i_offset to
487 * dp->i_offset + dp->i_count.
488 */
489 if (slotstatus == NONE) {
490 dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
491 dp->i_count = 0;
492 enduseful = dp->i_offset;
493 } else {
494 dp->i_offset = slotoffset;
495 dp->i_count = slotsize;
496 if (enduseful < slotoffset + slotsize)
497 enduseful = slotoffset + slotsize;
498 }
499 dp->i_endoff = roundup(enduseful, dirblksize);
500 dp->i_flag |= IN_CHANGE | IN_UPDATE;
501 /*
502 * We return with the directory locked, so that
503 * the parameters we set up above will still be
504 * valid if we actually decide to do a direnter().
505 * We return ni_vp == NULL to indicate that the entry
506 * does not currently exist; we leave a pointer to
507 * the (locked) directory inode in ndp->ni_dvp.
508 * The pathname buffer is saved so that the name
509 * can be obtained later.
510 *
511 * NB - if the directory is unlocked, then this
512 * information cannot be used.
513 */
514 cnp->cn_flags |= SAVENAME;
515 if (!lockparent) {
516 VOP_UNLOCK(vdp, 0, p);
517 cnp->cn_flags |= PDIRUNLOCK;
518 }
519 return (EJUSTRETURN);
520 }
521 /*
522 * Insert name into cache (as non-existent) if appropriate.
523 */
524 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
525 cache_enter(vdp, *vpp, cnp);
526 return (ENOENT);
527
528 found:
529 /*
530 * Check that directory length properly reflects presence
531 * of this entry.
532 */
533 if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
534 > ext2fs_size(dp)) {
535 ufs_dirbad(dp, dp->i_offset, "i_size too small");
536 error = ext2fs_setsize(dp,
537 entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
538 if (error) {
539 brelse(bp);
540 return(error);
541 }
542 dp->i_flag |= IN_CHANGE | IN_UPDATE;
543 }
544
545 /*
546 * Found component in pathname.
547 * If the final component of path name, save information
548 * in the cache as to where the entry was found.
549 */
550 if ((flags & ISLASTCN) && nameiop == LOOKUP)
551 dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
552
553 /*
554 * If deleting, and at end of pathname, return
555 * parameters which can be used to remove file.
556 * If the wantparent flag isn't set, we return only
557 * the directory (in ndp->ni_dvp), otherwise we go
558 * on and lock the inode, being careful with ".".
559 */
560 if (nameiop == DELETE && (flags & ISLASTCN)) {
561 /*
562 * Write access to directory required to delete files.
563 */
564 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
565 return (error);
566 /*
567 * Return pointer to current entry in dp->i_offset,
568 * and distance past previous entry (if there
569 * is a previous entry in this block) in dp->i_count.
570 * Save directory inode pointer in ndp->ni_dvp for dirremove().
571 */
572 if ((dp->i_offset & (dirblksize - 1)) == 0)
573 dp->i_count = 0;
574 else
575 dp->i_count = dp->i_offset - prevoff;
576 if (dp->i_number == dp->i_ino) {
577 VREF(vdp);
578 *vpp = vdp;
579 return (0);
580 }
581 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
582 return (error);
583 /*
584 * If directory is "sticky", then user must own
585 * the directory, or the file in it, else she
586 * may not delete it (unless she's root). This
587 * implements append-only directories.
588 */
589 if ((dp->i_e2fs_mode & ISVTX) &&
590 cred->cr_uid != 0 &&
591 cred->cr_uid != dp->i_e2fs_uid &&
592 VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
593 vput(tdp);
594 return (EPERM);
595 }
596 *vpp = tdp;
597 if (!lockparent) {
598 VOP_UNLOCK(vdp, 0, p);
599 cnp->cn_flags |= PDIRUNLOCK;
600 }
601 return (0);
602 }
603
604 /*
605 * If rewriting (RENAME), return the inode and the
606 * information required to rewrite the present directory
607 * Must get inode of directory entry to verify it's a
608 * regular file, or empty directory.
609 */
610 if (nameiop == RENAME && wantparent &&
611 (flags & ISLASTCN)) {
612 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
613 return (error);
614 /*
615 * Careful about locking second inode.
616 * This can only occur if the target is ".".
617 */
618 if (dp->i_number == dp->i_ino)
619 return (EISDIR);
620 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
621 return (error);
622 *vpp = tdp;
623 cnp->cn_flags |= SAVENAME;
624 if (!lockparent) {
625 VOP_UNLOCK(vdp, 0, p);
626 cnp->cn_flags |= PDIRUNLOCK;
627 }
628 return (0);
629 }
630
631 /*
632 * Step through the translation in the name. We do not `vput' the
633 * directory because we may need it again if a symbolic link
634 * is relative to the current directory. Instead we save it
635 * unlocked as "pdp". We must get the target inode before unlocking
636 * the directory to insure that the inode will not be removed
637 * before we get it. We prevent deadlock by always fetching
638 * inodes from the root, moving down the directory tree. Thus
639 * when following backward pointers ".." we must unlock the
640 * parent directory before getting the requested directory.
641 * There is a potential race condition here if both the current
642 * and parent directories are removed before the VFS_VGET for the
643 * inode associated with ".." returns. We hope that this occurs
644 * infrequently since we cannot avoid this race condition without
645 * implementing a sophisticated deadlock detection algorithm.
646 * Note also that this simple deadlock detection scheme will not
647 * work if the file system has any hard links other than ".."
648 * that point backwards in the directory structure.
649 */
650 pdp = vdp;
651 if (flags & ISDOTDOT) {
652 VOP_UNLOCK(pdp, 0, p); /* race to get the inode */
653 cnp->cn_flags |= PDIRUNLOCK;
654 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
655 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
656 cnp->cn_flags &= ~PDIRUNLOCK;
657 return (error);
658 }
659 if (lockparent && (flags & ISLASTCN)) {
660 if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
661 vput(tdp);
662 return (error);
663 }
664 cnp->cn_flags &= ~PDIRUNLOCK;
665 }
666 *vpp = tdp;
667 } else if (dp->i_number == dp->i_ino) {
668 VREF(vdp); /* we want ourself, ie "." */
669 *vpp = vdp;
670 } else {
671 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
672 return (error);
673 if (!lockparent || !(flags & ISLASTCN)) {
674 VOP_UNLOCK(pdp, 0, p);
675 cnp->cn_flags |= PDIRUNLOCK;
676 }
677 *vpp = tdp;
678 }
679
680 /*
681 * Insert name into cache if appropriate.
682 */
683 if (cnp->cn_flags & MAKEENTRY)
684 cache_enter(vdp, *vpp, cnp);
685 return (0);
686 }
687
688 /*
689 * Do consistency checking on a directory entry:
690 * record length must be multiple of 4
691 * entry must fit in rest of its dirblksize block
692 * record must be large enough to contain entry
693 * name is not longer than MAXNAMLEN
694 * name must be as long as advertised, and null terminated
695 */
696 /*
697 * changed so that it confirms to ext2fs_check_dir_entry
698 */
699 static int
700 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
701 int entryoffsetinblock)
702 {
703 int dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
704
705 char * error_msg = NULL;
706 int reclen = fs2h16(de->e2d_reclen);
707 int namlen = de->e2d_namlen;
708
709 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
710 error_msg = "rec_len is smaller than minimal";
711 else if (reclen % 4 != 0)
712 error_msg = "rec_len % 4 != 0";
713 else if (reclen < EXT2FS_DIRSIZ(namlen))
714 error_msg = "reclen is too small for name_len";
715 else if (entryoffsetinblock + reclen > dirblksize)
716 error_msg = "directory entry across blocks";
717 else if (fs2h32(de->e2d_ino) >
718 VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
719 error_msg = "inode out of bounds";
720
721 if (error_msg != NULL) {
722 printf( "bad directory entry: %s\n"
723 "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
724 error_msg, entryoffsetinblock,
725 (unsigned long) fs2h32(de->e2d_ino),
726 reclen, namlen);
727 panic("ext2fs_dirbadentry");
728 }
729 return error_msg == NULL ? 0 : 1;
730 }
731
732 /*
733 * Write a directory entry after a call to namei, using the parameters
734 * that it left in nameidata. The argument ip is the inode which the new
735 * directory entry will refer to. Dvp is a pointer to the directory to
736 * be written, which was left locked by namei. Remaining parameters
737 * (dp->i_offset, dp->i_count) indicate how the space for the new
738 * entry is to be obtained.
739 */
740 int
741 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
742 struct componentname *cnp)
743 {
744 struct ext2fs_direct *ep, *nep;
745 struct inode *dp;
746 struct buf *bp;
747 struct ext2fs_direct newdir;
748 struct iovec aiov;
749 struct uio auio;
750 u_int dsize;
751 int error, loc, newentrysize, spacefree;
752 char *dirbuf;
753 int dirblksize = ip->i_e2fs->e2fs_bsize;
754
755
756 #ifdef DIAGNOSTIC
757 if ((cnp->cn_flags & SAVENAME) == 0)
758 panic("direnter: missing name");
759 #endif
760 dp = VTOI(dvp);
761 newdir.e2d_ino = h2fs32(ip->i_number);
762 newdir.e2d_namlen = cnp->cn_namelen;
763 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
764 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
765 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
766 } else {
767 newdir.e2d_type = 0;
768 };
769 memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
770 newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
771 if (dp->i_count == 0) {
772 /*
773 * If dp->i_count is 0, then namei could find no
774 * space in the directory. Here, dp->i_offset will
775 * be on a directory block boundary and we will write the
776 * new entry into a fresh block.
777 */
778 if (dp->i_offset & (dirblksize - 1))
779 panic("ext2fs_direnter: newblk");
780 auio.uio_offset = dp->i_offset;
781 newdir.e2d_reclen = h2fs16(dirblksize);
782 auio.uio_resid = newentrysize;
783 aiov.iov_len = newentrysize;
784 aiov.iov_base = (caddr_t)&newdir;
785 auio.uio_iov = &aiov;
786 auio.uio_iovcnt = 1;
787 auio.uio_rw = UIO_WRITE;
788 auio.uio_segflg = UIO_SYSSPACE;
789 auio.uio_procp = (struct proc *)0;
790 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
791 if (dirblksize >
792 VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
793 /* XXX should grow with balloc() */
794 panic("ext2fs_direnter: frag size");
795 else if (!error) {
796 error = ext2fs_setsize(dp,
797 roundup(ext2fs_size(dp), dirblksize));
798 if (error)
799 return (error);
800 dp->i_flag |= IN_CHANGE;
801 }
802 return (error);
803 }
804
805 /*
806 * If dp->i_count is non-zero, then namei found space
807 * for the new entry in the range dp->i_offset to
808 * dp->i_offset + dp->i_count in the directory.
809 * To use this space, we may have to compact the entries located
810 * there, by copying them together towards the beginning of the
811 * block, leaving the free space in one usable chunk at the end.
812 */
813
814 /*
815 * Get the block containing the space for the new directory entry.
816 */
817 if ((error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, &dirbuf, &bp))
818 != 0)
819 return (error);
820 /*
821 * Find space for the new entry. In the simple case, the entry at
822 * offset base will have the space. If it does not, then namei
823 * arranged that compacting the region dp->i_offset to
824 * dp->i_offset + dp->i_count would yield the
825 * space.
826 */
827 ep = (struct ext2fs_direct *)dirbuf;
828 dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
829 spacefree = fs2h16(ep->e2d_reclen) - dsize;
830 for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) {
831 nep = (struct ext2fs_direct *)(dirbuf + loc);
832 if (ep->e2d_ino) {
833 /* trim the existing slot */
834 ep->e2d_reclen = h2fs16(dsize);
835 ep = (struct ext2fs_direct *)((char *)ep + dsize);
836 } else {
837 /* overwrite; nothing there; header is ours */
838 spacefree += dsize;
839 }
840 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
841 spacefree += fs2h16(nep->e2d_reclen) - dsize;
842 loc += fs2h16(nep->e2d_reclen);
843 memcpy((caddr_t)ep, (caddr_t)nep, dsize);
844 }
845 /*
846 * Update the pointer fields in the previous entry (if any),
847 * copy in the new entry, and write out the block.
848 */
849 if (ep->e2d_ino == 0) {
850 #ifdef DIAGNOSTIC
851 if (spacefree + dsize < newentrysize)
852 panic("ext2fs_direnter: compact1");
853 #endif
854 newdir.e2d_reclen = h2fs16(spacefree + dsize);
855 } else {
856 #ifdef DIAGNOSTIC
857 if (spacefree < newentrysize) {
858 printf("ext2fs_direnter: compact2 %u %u",
859 (u_int)spacefree, (u_int)newentrysize);
860 panic("ext2fs_direnter: compact2");
861 }
862 #endif
863 newdir.e2d_reclen = h2fs16(spacefree);
864 ep->e2d_reclen = h2fs16(dsize);
865 ep = (struct ext2fs_direct *)((char *)ep + dsize);
866 }
867 memcpy((caddr_t)ep, (caddr_t)&newdir, (u_int)newentrysize);
868 error = VOP_BWRITE(bp);
869 dp->i_flag |= IN_CHANGE | IN_UPDATE;
870 if (!error && dp->i_endoff && dp->i_endoff < ext2fs_size(dp))
871 error = ext2fs_truncate(dp, (off_t)dp->i_endoff, IO_SYNC,
872 cnp->cn_cred);
873 return (error);
874 }
875
876 /*
877 * Remove a directory entry after a call to namei, using
878 * the parameters which it left in nameidata. The entry
879 * dp->i_offset contains the offset into the directory of the
880 * entry to be eliminated. The dp->i_count field contains the
881 * size of the previous record in the directory. If this
882 * is 0, the first entry is being deleted, so we need only
883 * zero the inode number to mark the entry as free. If the
884 * entry is not the first in the directory, we must reclaim
885 * the space of the now empty record by adding the record size
886 * to the size of the previous entry.
887 */
888 int
889 ext2fs_dirremove(struct vnode *dvp, struct componentname *cnp)
890 {
891 struct inode *dp;
892 struct ext2fs_direct *ep;
893 struct buf *bp;
894 int error;
895
896 dp = VTOI(dvp);
897 if (dp->i_count == 0) {
898 /*
899 * First entry in block: set d_ino to zero.
900 */
901 error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep,
902 &bp);
903 if (error != 0)
904 return (error);
905 ep->e2d_ino = 0;
906 error = VOP_BWRITE(bp);
907 dp->i_flag |= IN_CHANGE | IN_UPDATE;
908 return (error);
909 }
910 /*
911 * Collapse new free space into previous entry.
912 */
913 error = ext2fs_bufatoff(dp, (off_t)(dp->i_offset - dp->i_count),
914 (char **)&ep, &bp);
915 if (error != 0)
916 return (error);
917 ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen);
918 error = VOP_BWRITE(bp);
919 dp->i_flag |= IN_CHANGE | IN_UPDATE;
920 return (error);
921 }
922
923 /*
924 * Rewrite an existing directory entry to point at the inode
925 * supplied. The parameters describing the directory entry are
926 * set up by a call to namei.
927 */
928 int
929 ext2fs_dirrewrite(struct inode *dp, struct inode *ip,
930 struct componentname *cnp)
931 {
932 struct buf *bp;
933 struct ext2fs_direct *ep;
934 int error;
935
936 error = ext2fs_bufatoff(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
937 if (error != 0)
938 return (error);
939 ep->e2d_ino = h2fs32(ip->i_number);
940 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
941 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
942 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
943 } else {
944 ep->e2d_type = 0;
945 }
946 error = VOP_BWRITE(bp);
947 dp->i_flag |= IN_CHANGE | IN_UPDATE;
948 return (error);
949 }
950
951 /*
952 * Check if a directory is empty or not.
953 * Inode supplied must be locked.
954 *
955 * Using a struct dirtemplate here is not precisely
956 * what we want, but better than using a struct ext2fs_direct.
957 *
958 * NB: does not handle corrupted directories.
959 */
960 int
961 ext2fs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
962 {
963 off_t off;
964 struct ext2fs_dirtemplate dbuf;
965 struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
966 int error, namlen;
967 size_t count;
968
969 #define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
970
971 for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
972 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
973 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
974 /*
975 * Since we read MINDIRSIZ, residual must
976 * be 0 unless we're at end of file.
977 */
978 if (error || count != 0)
979 return (0);
980 /* avoid infinite loops */
981 if (dp->e2d_reclen == 0)
982 return (0);
983 /* skip empty entries */
984 if (dp->e2d_ino == 0)
985 continue;
986 /* accept only "." and ".." */
987 namlen = dp->e2d_namlen;
988 if (namlen > 2)
989 return (0);
990 if (dp->e2d_name[0] != '.')
991 return (0);
992 /*
993 * At this point namlen must be 1 or 2.
994 * 1 implies ".", 2 implies ".." if second
995 * char is also "."
996 */
997 if (namlen == 1)
998 continue;
999 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1000 continue;
1001 return (0);
1002 }
1003 return (1);
1004 }
1005
1006 /*
1007 * Check if source directory is in the path of the target directory.
1008 * Target is supplied locked, source is unlocked.
1009 * The target is always vput before returning.
1010 */
1011 int
1012 ext2fs_checkpath(struct inode *source, struct inode *target,
1013 struct ucred *cred)
1014 {
1015 struct vnode *vp;
1016 int error, rootino, namlen;
1017 struct ext2fs_dirtemplate dirbuf;
1018 u_int32_t ino;
1019
1020 vp = ITOV(target);
1021 if (target->i_number == source->i_number) {
1022 error = EEXIST;
1023 goto out;
1024 }
1025 rootino = ROOTINO;
1026 error = 0;
1027 if (target->i_number == rootino)
1028 goto out;
1029
1030 for (;;) {
1031 if (vp->v_type != VDIR) {
1032 error = ENOTDIR;
1033 break;
1034 }
1035 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1036 sizeof (struct ext2fs_dirtemplate), (off_t)0,
1037 UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1038 (struct proc *)0);
1039 if (error != 0)
1040 break;
1041 namlen = dirbuf.dotdot_namlen;
1042 if (namlen != 2 ||
1043 dirbuf.dotdot_name[0] != '.' ||
1044 dirbuf.dotdot_name[1] != '.') {
1045 error = ENOTDIR;
1046 break;
1047 }
1048 ino = fs2h32(dirbuf.dotdot_ino);
1049 if (ino == source->i_number) {
1050 error = EINVAL;
1051 break;
1052 }
1053 if (ino == rootino)
1054 break;
1055 vput(vp);
1056 error = VFS_VGET(vp->v_mount, ino, &vp);
1057 if (error != 0) {
1058 vp = NULL;
1059 break;
1060 }
1061 }
1062
1063 out:
1064 if (error == ENOTDIR) {
1065 printf("checkpath: .. not a directory\n");
1066 panic("checkpath");
1067 }
1068 if (vp != NULL)
1069 vput(vp);
1070 return (error);
1071 }