1 /* $NetBSD: ext2fs_lookup.c,v 1.28 2003/08/07 16:34:26 agc Exp $ */
2
3 /*
4 * Modified for NetBSD 1.2E
5 * May 1997, Manuel Bouyer
6 * Laboratoire d'informatique de Paris VI
7 */
8 /*
9 * modified for Lites 1.1
10 *
11 * Aug 1995, Godmar Back (gback@cs.utah.edu)
12 * University of Utah, Department of Computer Science
13 */
14 /*
15 * Copyright (c) 1989, 1993
16 * The Regents of the University of California. All rights reserved.
17 * (c) UNIX System Laboratories, Inc.
18 * All or some portions of this file are derived from material licensed
19 * to the University of California by American Telephone and Telegraph
20 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
21 * the permission of UNIX System Laboratories, Inc.
22 *
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
25 * are met:
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 3. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
34 *
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
45 * SUCH DAMAGE.
46 *
47 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
48 */
49
50 #include <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.28 2003/08/07 16:34:26 agc Exp $");
52
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/namei.h>
56 #include <sys/buf.h>
57 #include <sys/file.h>
58 #include <sys/mount.h>
59 #include <sys/vnode.h>
60 #include <sys/malloc.h>
61 #include <sys/dirent.h>
62
63 #include <ufs/ufs/inode.h>
64 #include <ufs/ufs/ufsmount.h>
65 #include <ufs/ufs/ufs_extern.h>
66
67 #include <ufs/ext2fs/ext2fs_extern.h>
68 #include <ufs/ext2fs/ext2fs_dir.h>
69 #include <ufs/ext2fs/ext2fs.h>
70
71 extern int dirchk;
72
73 static void ext2fs_dirconv2ffs __P((struct ext2fs_direct *e2dir,
74 struct dirent *ffsdir));
75 static int ext2fs_dirbadentry __P((struct vnode *dp,
76 struct ext2fs_direct *de,
77 int entryoffsetinblock));
78
79 /*
80 * the problem that is tackled below is the fact that FFS
81 * includes the terminating zero on disk while EXT2FS doesn't
82 * this implies that we need to introduce some padding.
83 * For instance, a filename "sbin" has normally a reclen 12
84 * in EXT2, but 16 in FFS.
85 * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
86 * If it wasn't for that, the complete ufs code for directories would
87 * have worked w/o changes (except for the difference in DIRBLKSIZ)
88 */
89 static void
90 ext2fs_dirconv2ffs( e2dir, ffsdir)
91 struct ext2fs_direct *e2dir;
92 struct dirent *ffsdir;
93 {
94 memset(ffsdir, 0, sizeof(struct dirent));
95 ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
96 ffsdir->d_namlen = e2dir->e2d_namlen;
97
98 ffsdir->d_type = DT_UNKNOWN; /* don't know more here */
99 #ifdef DIAGNOSTIC
100 /*
101 * XXX Right now this can't happen, but if one day
102 * MAXNAMLEN != E2FS_MAXNAMLEN we should handle this more gracefully !
103 */
104 #if 0
105 if (e2dir->e2d_namlen > MAXNAMLEN)
106 panic("ext2fs: e2dir->e2d_namlen");
107 #endif
108 #endif
109 strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
110
111 /* Godmar thinks: since e2dir->e2d_reclen can be big and means
112 nothing anyway, we compute our own reclen according to what
113 we think is right
114 */
115 ffsdir->d_reclen = DIRENT_SIZE(ffsdir);
116 }
117
118 /*
119 * Vnode op for reading directories.
120 *
121 * Convert the on-disk entries to <sys/dirent.h> entries.
122 * the problem is that the conversion will blow up some entries by four bytes,
123 * so it can't be done in place. This is too bad. Right now the conversion is
124 * done entry by entry, the converted entry is sent via uiomove.
125 *
126 * XXX allocate a buffer, convert as many entries as possible, then send
127 * the whole buffer to uiomove
128 */
129 int
130 ext2fs_readdir(v)
131 void *v;
132 {
133 struct vop_readdir_args /* {
134 struct vnode *a_vp;
135 struct uio *a_uio;
136 struct ucred *a_cred;
137 int **a_eofflag;
138 off_t **a_cookies;
139 int ncookies;
140 } */ *ap = v;
141 struct uio *uio = ap->a_uio;
142 int error;
143 size_t e2fs_count, readcnt;
144 struct vnode *vp = ap->a_vp;
145 struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
146
147 struct ext2fs_direct *dp;
148 struct dirent dstd;
149 struct uio auio;
150 struct iovec aiov;
151 caddr_t dirbuf;
152 off_t off = uio->uio_offset;
153 off_t *cookies = NULL;
154 int nc = 0, ncookies = 0;
155 int e2d_reclen;
156
157 if (vp->v_type != VDIR)
158 return (ENOTDIR);
159
160 e2fs_count = uio->uio_resid;
161 /* Make sure we don't return partial entries. */
162 e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
163 if (e2fs_count <= 0)
164 return (EINVAL);
165
166 auio = *uio;
167 auio.uio_iov = &aiov;
168 auio.uio_iovcnt = 1;
169 auio.uio_segflg = UIO_SYSSPACE;
170 aiov.iov_len = e2fs_count;
171 auio.uio_resid = e2fs_count;
172 MALLOC(dirbuf, caddr_t, e2fs_count, M_TEMP, M_WAITOK);
173 if (ap->a_ncookies) {
174 nc = ncookies = e2fs_count / 16;
175 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
176 *ap->a_cookies = cookies;
177 }
178 memset(dirbuf, 0, e2fs_count);
179 aiov.iov_base = dirbuf;
180
181 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
182 if (error == 0) {
183 readcnt = e2fs_count - auio.uio_resid;
184 for (dp = (struct ext2fs_direct *)dirbuf;
185 (char *)dp < (char *)dirbuf + readcnt; ) {
186 e2d_reclen = fs2h16(dp->e2d_reclen);
187 if (e2d_reclen == 0) {
188 error = EIO;
189 break;
190 }
191 ext2fs_dirconv2ffs(dp, &dstd);
192 if(dstd.d_reclen > uio->uio_resid) {
193 break;
194 }
195 if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
196 break;
197 }
198 off = off + e2d_reclen;
199 if (cookies != NULL) {
200 *cookies++ = off;
201 if (--ncookies <= 0){
202 break; /* out of cookies */
203 }
204 }
205 /* advance dp */
206 dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
207 }
208 /* we need to correct uio_offset */
209 uio->uio_offset = off;
210 }
211 FREE(dirbuf, M_TEMP);
212 *ap->a_eofflag = VTOI(ap->a_vp)->i_e2fs_size <= uio->uio_offset;
213 if (ap->a_ncookies) {
214 if (error) {
215 free(*ap->a_cookies, M_TEMP);
216 *ap->a_ncookies = 0;
217 *ap->a_cookies = NULL;
218 } else
219 *ap->a_ncookies = nc - ncookies;
220 }
221 return (error);
222 }
223
224 /*
225 * Convert a component of a pathname into a pointer to a locked inode.
226 * This is a very central and rather complicated routine.
227 * If the file system is not maintained in a strict tree hierarchy,
228 * this can result in a deadlock situation (see comments in code below).
229 *
230 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
231 * on whether the name is to be looked up, created, renamed, or deleted.
232 * When CREATE, RENAME, or DELETE is specified, information usable in
233 * creating, renaming, or deleting a directory entry may be calculated.
234 * If flag has LOCKPARENT or'ed into it and the target of the pathname
235 * exists, lookup returns both the target and its parent directory locked.
236 * When creating or renaming and LOCKPARENT is specified, the target may
237 * not be ".". When deleting and LOCKPARENT is specified, the target may
238 * be "."., but the caller must check to ensure it does an vrele and vput
239 * instead of two vputs.
240 *
241 * Overall outline of ext2fs_lookup:
242 *
243 * check accessibility of directory
244 * look for name in cache, if found, then if at end of path
245 * and deleting or creating, drop it, else return name
246 * search for name in directory, to found or notfound
247 * notfound:
248 * if creating, return locked directory, leaving info on available slots
249 * else return error
250 * found:
251 * if at end of path and deleting, return information to allow delete
252 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
253 * inode and return info to allow rewrite
254 * if not at end, add name to cache; if at end and neither creating
255 * nor deleting, add name to cache
256 */
257 int
258 ext2fs_lookup(v)
259 void *v;
260 {
261 struct vop_lookup_args /* {
262 struct vnode *a_dvp;
263 struct vnode **a_vpp;
264 struct componentname *a_cnp;
265 } */ *ap = v;
266 struct vnode *vdp; /* vnode for directory being searched */
267 struct inode *dp; /* inode for directory being searched */
268 struct buf *bp; /* a buffer of directory entries */
269 struct ext2fs_direct *ep; /* the current directory entry */
270 int entryoffsetinblock; /* offset of ep in bp's buffer */
271 enum {NONE, COMPACT, FOUND} slotstatus;
272 doff_t slotoffset; /* offset of area with free space */
273 int slotsize; /* size of area at slotoffset */
274 int slotfreespace; /* amount of space free in slot */
275 int slotneeded; /* size of the entry we're seeking */
276 int numdirpasses; /* strategy for directory search */
277 doff_t endsearch; /* offset to end directory search */
278 doff_t prevoff; /* prev entry dp->i_offset */
279 struct vnode *pdp; /* saved dp during symlink work */
280 struct vnode *tdp; /* returned by VFS_VGET */
281 doff_t enduseful; /* pointer past last used dir slot */
282 u_long bmask; /* block offset mask */
283 int lockparent; /* 1 => lockparent flag is set */
284 int wantparent; /* 1 => wantparent or lockparent flag */
285 int namlen, error;
286 struct vnode **vpp = ap->a_vpp;
287 struct componentname *cnp = ap->a_cnp;
288 struct ucred *cred = cnp->cn_cred;
289 int flags = cnp->cn_flags;
290 int nameiop = cnp->cn_nameiop;
291 ino_t foundino;
292
293 int dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
294
295 bp = NULL;
296 slotoffset = -1;
297 *vpp = NULL;
298 vdp = ap->a_dvp;
299 dp = VTOI(vdp);
300 lockparent = flags & LOCKPARENT;
301 wantparent = flags & (LOCKPARENT|WANTPARENT);
302 /*
303 * Check accessiblity of directory.
304 */
305 if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
306 return (error);
307
308 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
309 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
310 return (EROFS);
311
312 /*
313 * We now have a segment name to search for, and a directory to search.
314 *
315 * Before tediously performing a linear scan of the directory,
316 * check the name cache to see if the directory/name pair
317 * we are looking for is known already.
318 */
319 if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
320 return (error);
321
322 /*
323 * Suppress search for slots unless creating
324 * file and at end of pathname, in which case
325 * we watch for a place to put the new file in
326 * case it doesn't already exist.
327 */
328 slotstatus = FOUND;
329 slotfreespace = slotsize = slotneeded = 0;
330 if ((nameiop == CREATE || nameiop == RENAME) &&
331 (flags & ISLASTCN)) {
332 slotstatus = NONE;
333 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
334 }
335
336 /*
337 * If there is cached information on a previous search of
338 * this directory, pick up where we last left off.
339 * We cache only lookups as these are the most common
340 * and have the greatest payoff. Caching CREATE has little
341 * benefit as it usually must search the entire directory
342 * to determine that the entry does not exist. Caching the
343 * location of the last DELETE or RENAME has not reduced
344 * profiling time and hence has been removed in the interest
345 * of simplicity.
346 */
347 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
348 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
349 dp->i_diroff > dp->i_e2fs_size) {
350 entryoffsetinblock = 0;
351 dp->i_offset = 0;
352 numdirpasses = 1;
353 } else {
354 dp->i_offset = dp->i_diroff;
355 if ((entryoffsetinblock = dp->i_offset & bmask) &&
356 (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
357 return (error);
358 numdirpasses = 2;
359 }
360 prevoff = dp->i_offset;
361 endsearch = roundup(dp->i_e2fs_size, dirblksize);
362 enduseful = 0;
363
364 searchloop:
365 while (dp->i_offset < endsearch) {
366 /*
367 * If necessary, get the next directory block.
368 */
369 if ((dp->i_offset & bmask) == 0) {
370 if (bp != NULL)
371 brelse(bp);
372 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset,
373 NULL, &bp);
374 if (error != 0)
375 return (error);
376 entryoffsetinblock = 0;
377 }
378 /*
379 * If still looking for a slot, and at a dirblksize
380 * boundary, have to start looking for free space again.
381 */
382 if (slotstatus == NONE &&
383 (entryoffsetinblock & (dirblksize - 1)) == 0) {
384 slotoffset = -1;
385 slotfreespace = 0;
386 }
387 /*
388 * Get pointer to next entry.
389 * Full validation checks are slow, so we only check
390 * enough to insure forward progress through the
391 * directory. Complete checks can be run by patching
392 * "dirchk" to be true.
393 */
394 ep = (struct ext2fs_direct *)
395 ((char *)bp->b_data + entryoffsetinblock);
396 if (ep->e2d_reclen == 0 ||
397 (dirchk &&
398 ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
399 int i;
400 ufs_dirbad(dp, dp->i_offset, "mangled entry");
401 i = dirblksize -
402 (entryoffsetinblock & (dirblksize - 1));
403 dp->i_offset += i;
404 entryoffsetinblock += i;
405 continue;
406 }
407
408 /*
409 * If an appropriate sized slot has not yet been found,
410 * check to see if one is available. Also accumulate space
411 * in the current block so that we can determine if
412 * compaction is viable.
413 */
414 if (slotstatus != FOUND) {
415 int size = fs2h16(ep->e2d_reclen);
416
417 if (ep->e2d_ino != 0)
418 size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
419 if (size > 0) {
420 if (size >= slotneeded) {
421 slotstatus = FOUND;
422 slotoffset = dp->i_offset;
423 slotsize = fs2h16(ep->e2d_reclen);
424 } else if (slotstatus == NONE) {
425 slotfreespace += size;
426 if (slotoffset == -1)
427 slotoffset = dp->i_offset;
428 if (slotfreespace >= slotneeded) {
429 slotstatus = COMPACT;
430 slotsize = dp->i_offset +
431 fs2h16(ep->e2d_reclen) - slotoffset;
432 }
433 }
434 }
435 }
436
437 /*
438 * Check for a name match.
439 */
440 if (ep->e2d_ino) {
441 namlen = ep->e2d_namlen;
442 if (namlen == cnp->cn_namelen &&
443 !memcmp(cnp->cn_nameptr, ep->e2d_name,
444 (unsigned)namlen)) {
445 /*
446 * Save directory entry's inode number and
447 * reclen in ndp->ni_ufs area, and release
448 * directory buffer.
449 */
450 foundino = fs2h32(ep->e2d_ino);
451 dp->i_reclen = fs2h16(ep->e2d_reclen);
452 brelse(bp);
453 goto found;
454 }
455 }
456 prevoff = dp->i_offset;
457 dp->i_offset += fs2h16(ep->e2d_reclen);
458 entryoffsetinblock += fs2h16(ep->e2d_reclen);
459 if (ep->e2d_ino)
460 enduseful = dp->i_offset;
461 }
462 /* notfound: */
463 /*
464 * If we started in the middle of the directory and failed
465 * to find our target, we must check the beginning as well.
466 */
467 if (numdirpasses == 2) {
468 numdirpasses--;
469 dp->i_offset = 0;
470 endsearch = dp->i_diroff;
471 goto searchloop;
472 }
473 if (bp != NULL)
474 brelse(bp);
475 /*
476 * If creating, and at end of pathname and current
477 * directory has not been removed, then can consider
478 * allowing file to be created.
479 */
480 if ((nameiop == CREATE || nameiop == RENAME) &&
481 (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
482 /*
483 * Access for write is interpreted as allowing
484 * creation of files in the directory.
485 */
486 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
487 return (error);
488 /*
489 * Return an indication of where the new directory
490 * entry should be put. If we didn't find a slot,
491 * then set dp->i_count to 0 indicating
492 * that the new slot belongs at the end of the
493 * directory. If we found a slot, then the new entry
494 * can be put in the range from dp->i_offset to
495 * dp->i_offset + dp->i_count.
496 */
497 if (slotstatus == NONE) {
498 dp->i_offset = roundup(dp->i_e2fs_size, dirblksize);
499 dp->i_count = 0;
500 enduseful = dp->i_offset;
501 } else {
502 dp->i_offset = slotoffset;
503 dp->i_count = slotsize;
504 if (enduseful < slotoffset + slotsize)
505 enduseful = slotoffset + slotsize;
506 }
507 dp->i_endoff = roundup(enduseful, dirblksize);
508 dp->i_flag |= IN_CHANGE | IN_UPDATE;
509 /*
510 * We return with the directory locked, so that
511 * the parameters we set up above will still be
512 * valid if we actually decide to do a direnter().
513 * We return ni_vp == NULL to indicate that the entry
514 * does not currently exist; we leave a pointer to
515 * the (locked) directory inode in ndp->ni_dvp.
516 * The pathname buffer is saved so that the name
517 * can be obtained later.
518 *
519 * NB - if the directory is unlocked, then this
520 * information cannot be used.
521 */
522 cnp->cn_flags |= SAVENAME;
523 if (!lockparent) {
524 VOP_UNLOCK(vdp, 0);
525 cnp->cn_flags |= PDIRUNLOCK;
526 }
527 return (EJUSTRETURN);
528 }
529 /*
530 * Insert name into cache (as non-existent) if appropriate.
531 */
532 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
533 cache_enter(vdp, *vpp, cnp);
534 return (ENOENT);
535
536 found:
537 /*
538 * Check that directory length properly reflects presence
539 * of this entry.
540 */
541 if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
542 > dp->i_e2fs_size) {
543 ufs_dirbad(dp, dp->i_offset, "i_size too small");
544 dp->i_e2fs_size = entryoffsetinblock +
545 EXT2FS_DIRSIZ(ep->e2d_namlen);
546 dp->i_flag |= IN_CHANGE | IN_UPDATE;
547 }
548
549 /*
550 * Found component in pathname.
551 * If the final component of path name, save information
552 * in the cache as to where the entry was found.
553 */
554 if ((flags & ISLASTCN) && nameiop == LOOKUP)
555 dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
556
557 /*
558 * If deleting, and at end of pathname, return
559 * parameters which can be used to remove file.
560 * If the wantparent flag isn't set, we return only
561 * the directory (in ndp->ni_dvp), otherwise we go
562 * on and lock the inode, being careful with ".".
563 */
564 if (nameiop == DELETE && (flags & ISLASTCN)) {
565 /*
566 * Write access to directory required to delete files.
567 */
568 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
569 return (error);
570 /*
571 * Return pointer to current entry in dp->i_offset,
572 * and distance past previous entry (if there
573 * is a previous entry in this block) in dp->i_count.
574 * Save directory inode pointer in ndp->ni_dvp for dirremove().
575 */
576 if ((dp->i_offset & (dirblksize - 1)) == 0)
577 dp->i_count = 0;
578 else
579 dp->i_count = dp->i_offset - prevoff;
580 if (dp->i_number == foundino) {
581 VREF(vdp);
582 *vpp = vdp;
583 return (0);
584 }
585 if ((error = VFS_VGET(vdp->v_mount, foundino, &tdp)) != 0)
586 return (error);
587 /*
588 * If directory is "sticky", then user must own
589 * the directory, or the file in it, else she
590 * may not delete it (unless she's root). This
591 * implements append-only directories.
592 */
593 if ((dp->i_e2fs_mode & ISVTX) &&
594 cred->cr_uid != 0 &&
595 cred->cr_uid != dp->i_e2fs_uid &&
596 VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
597 vput(tdp);
598 return (EPERM);
599 }
600 *vpp = tdp;
601 if (!lockparent) {
602 VOP_UNLOCK(vdp, 0);
603 cnp->cn_flags |= PDIRUNLOCK;
604 }
605 return (0);
606 }
607
608 /*
609 * If rewriting (RENAME), return the inode and the
610 * information required to rewrite the present directory
611 * Must get inode of directory entry to verify it's a
612 * regular file, or empty directory.
613 */
614 if (nameiop == RENAME && wantparent &&
615 (flags & ISLASTCN)) {
616 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
617 if (error)
618 return (error);
619 /*
620 * Careful about locking second inode.
621 * This can only occur if the target is ".".
622 */
623 if (dp->i_number == foundino)
624 return (EISDIR);
625 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
626 if (error)
627 return (error);
628 *vpp = tdp;
629 cnp->cn_flags |= SAVENAME;
630 if (!lockparent) {
631 VOP_UNLOCK(vdp, 0);
632 cnp->cn_flags |= PDIRUNLOCK;
633 }
634 return (0);
635 }
636
637 /*
638 * Step through the translation in the name. We do not `vput' the
639 * directory because we may need it again if a symbolic link
640 * is relative to the current directory. Instead we save it
641 * unlocked as "pdp". We must get the target inode before unlocking
642 * the directory to insure that the inode will not be removed
643 * before we get it. We prevent deadlock by always fetching
644 * inodes from the root, moving down the directory tree. Thus
645 * when following backward pointers ".." we must unlock the
646 * parent directory before getting the requested directory.
647 * There is a potential race condition here if both the current
648 * and parent directories are removed before the VFS_VGET for the
649 * inode associated with ".." returns. We hope that this occurs
650 * infrequently since we cannot avoid this race condition without
651 * implementing a sophisticated deadlock detection algorithm.
652 * Note also that this simple deadlock detection scheme will not
653 * work if the file system has any hard links other than ".."
654 * that point backwards in the directory structure.
655 */
656 pdp = vdp;
657 if (flags & ISDOTDOT) {
658 VOP_UNLOCK(pdp, 0); /* race to get the inode */
659 cnp->cn_flags |= PDIRUNLOCK;
660 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
661 if (error) {
662 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
663 cnp->cn_flags &= ~PDIRUNLOCK;
664 return (error);
665 }
666 if (lockparent && (flags & ISLASTCN)) {
667 if ((error = vn_lock(pdp, LK_EXCLUSIVE))) {
668 vput(tdp);
669 return (error);
670 }
671 cnp->cn_flags &= ~PDIRUNLOCK;
672 }
673 *vpp = tdp;
674 } else if (dp->i_number == foundino) {
675 VREF(vdp); /* we want ourself, ie "." */
676 *vpp = vdp;
677 } else {
678 if ((error = VFS_VGET(vdp->v_mount, foundino, &tdp)) != 0)
679 return (error);
680 if (!lockparent || !(flags & ISLASTCN)) {
681 VOP_UNLOCK(pdp, 0);
682 cnp->cn_flags |= PDIRUNLOCK;
683 }
684 *vpp = tdp;
685 }
686
687 /*
688 * Insert name into cache if appropriate.
689 */
690 if (cnp->cn_flags & MAKEENTRY)
691 cache_enter(vdp, *vpp, cnp);
692 return (0);
693 }
694
695 /*
696 * Do consistency checking on a directory entry:
697 * record length must be multiple of 4
698 * entry must fit in rest of its dirblksize block
699 * record must be large enough to contain entry
700 * name is not longer than MAXNAMLEN
701 * name must be as long as advertised, and null terminated
702 */
703 /*
704 * changed so that it confirms to ext2fs_check_dir_entry
705 */
706 static int
707 ext2fs_dirbadentry(dp, de, entryoffsetinblock)
708 struct vnode *dp;
709 struct ext2fs_direct *de;
710 int entryoffsetinblock;
711 {
712 int dirblksize = VTOI(dp)->i_e2fs->e2fs_bsize;
713
714 char * error_msg = NULL;
715 int reclen = fs2h16(de->e2d_reclen);
716 int namlen = de->e2d_namlen;
717
718 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
719 error_msg = "rec_len is smaller than minimal";
720 else if (reclen % 4 != 0)
721 error_msg = "rec_len % 4 != 0";
722 else if (reclen < EXT2FS_DIRSIZ(namlen))
723 error_msg = "reclen is too small for name_len";
724 else if (entryoffsetinblock + reclen > dirblksize)
725 error_msg = "directory entry across blocks";
726 else if (fs2h32(de->e2d_ino) >
727 VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
728 error_msg = "inode out of bounds";
729
730 if (error_msg != NULL) {
731 printf( "bad directory entry: %s\n"
732 "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
733 error_msg, entryoffsetinblock,
734 (unsigned long) fs2h32(de->e2d_ino),
735 reclen, namlen);
736 panic("ext2fs_dirbadentry");
737 }
738 return error_msg == NULL ? 0 : 1;
739 }
740
741 /*
742 * Write a directory entry after a call to namei, using the parameters
743 * that it left in nameidata. The argument ip is the inode which the new
744 * directory entry will refer to. Dvp is a pointer to the directory to
745 * be written, which was left locked by namei. Remaining parameters
746 * (dp->i_offset, dp->i_count) indicate how the space for the new
747 * entry is to be obtained.
748 */
749 int
750 ext2fs_direnter(ip, dvp, cnp)
751 struct inode *ip;
752 struct vnode *dvp;
753 struct componentname *cnp;
754 {
755 struct ext2fs_direct *ep, *nep;
756 struct inode *dp;
757 struct buf *bp;
758 struct ext2fs_direct newdir;
759 struct iovec aiov;
760 struct uio auio;
761 u_int dsize;
762 int error, loc, newentrysize, spacefree;
763 char *dirbuf;
764 int dirblksize = ip->i_e2fs->e2fs_bsize;
765
766
767 #ifdef DIAGNOSTIC
768 if ((cnp->cn_flags & SAVENAME) == 0)
769 panic("direnter: missing name");
770 #endif
771 dp = VTOI(dvp);
772 newdir.e2d_ino = h2fs32(ip->i_number);
773 newdir.e2d_namlen = cnp->cn_namelen;
774 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
775 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
776 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
777 } else {
778 newdir.e2d_type = 0;
779 };
780 memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
781 newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
782 if (dp->i_count == 0) {
783 /*
784 * If dp->i_count is 0, then namei could find no
785 * space in the directory. Here, dp->i_offset will
786 * be on a directory block boundary and we will write the
787 * new entry into a fresh block.
788 */
789 if (dp->i_offset & (dirblksize - 1))
790 panic("ext2fs_direnter: newblk");
791 auio.uio_offset = dp->i_offset;
792 newdir.e2d_reclen = h2fs16(dirblksize);
793 auio.uio_resid = newentrysize;
794 aiov.iov_len = newentrysize;
795 aiov.iov_base = (caddr_t)&newdir;
796 auio.uio_iov = &aiov;
797 auio.uio_iovcnt = 1;
798 auio.uio_rw = UIO_WRITE;
799 auio.uio_segflg = UIO_SYSSPACE;
800 auio.uio_procp = (struct proc *)0;
801 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
802 if (dirblksize >
803 VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
804 /* XXX should grow with balloc() */
805 panic("ext2fs_direnter: frag size");
806 else if (!error) {
807 dp->i_e2fs_size = roundup(dp->i_e2fs_size, dirblksize);
808 dp->i_flag |= IN_CHANGE;
809 }
810 return (error);
811 }
812
813 /*
814 * If dp->i_count is non-zero, then namei found space
815 * for the new entry in the range dp->i_offset to
816 * dp->i_offset + dp->i_count in the directory.
817 * To use this space, we may have to compact the entries located
818 * there, by copying them together towards the beginning of the
819 * block, leaving the free space in one usable chunk at the end.
820 */
821
822 /*
823 * Get the block containing the space for the new directory entry.
824 */
825 if ((error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
826 return (error);
827 /*
828 * Find space for the new entry. In the simple case, the entry at
829 * offset base will have the space. If it does not, then namei
830 * arranged that compacting the region dp->i_offset to
831 * dp->i_offset + dp->i_count would yield the
832 * space.
833 */
834 ep = (struct ext2fs_direct *)dirbuf;
835 dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
836 spacefree = fs2h16(ep->e2d_reclen) - dsize;
837 for (loc = fs2h16(ep->e2d_reclen); loc < dp->i_count; ) {
838 nep = (struct ext2fs_direct *)(dirbuf + loc);
839 if (ep->e2d_ino) {
840 /* trim the existing slot */
841 ep->e2d_reclen = h2fs16(dsize);
842 ep = (struct ext2fs_direct *)((char *)ep + dsize);
843 } else {
844 /* overwrite; nothing there; header is ours */
845 spacefree += dsize;
846 }
847 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
848 spacefree += fs2h16(nep->e2d_reclen) - dsize;
849 loc += fs2h16(nep->e2d_reclen);
850 memcpy((caddr_t)ep, (caddr_t)nep, dsize);
851 }
852 /*
853 * Update the pointer fields in the previous entry (if any),
854 * copy in the new entry, and write out the block.
855 */
856 if (ep->e2d_ino == 0) {
857 #ifdef DIAGNOSTIC
858 if (spacefree + dsize < newentrysize)
859 panic("ext2fs_direnter: compact1");
860 #endif
861 newdir.e2d_reclen = h2fs16(spacefree + dsize);
862 } else {
863 #ifdef DIAGNOSTIC
864 if (spacefree < newentrysize) {
865 printf("ext2fs_direnter: compact2 %u %u",
866 (u_int)spacefree, (u_int)newentrysize);
867 panic("ext2fs_direnter: compact2");
868 }
869 #endif
870 newdir.e2d_reclen = h2fs16(spacefree);
871 ep->e2d_reclen = h2fs16(dsize);
872 ep = (struct ext2fs_direct *)((char *)ep + dsize);
873 }
874 memcpy((caddr_t)ep, (caddr_t)&newdir, (u_int)newentrysize);
875 error = VOP_BWRITE(bp);
876 dp->i_flag |= IN_CHANGE | IN_UPDATE;
877 if (!error && dp->i_endoff && dp->i_endoff < dp->i_e2fs_size)
878 error = VOP_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
879 cnp->cn_cred, cnp->cn_proc);
880 return (error);
881 }
882
883 /*
884 * Remove a directory entry after a call to namei, using
885 * the parameters which it left in nameidata. The entry
886 * dp->i_offset contains the offset into the directory of the
887 * entry to be eliminated. The dp->i_count field contains the
888 * size of the previous record in the directory. If this
889 * is 0, the first entry is being deleted, so we need only
890 * zero the inode number to mark the entry as free. If the
891 * entry is not the first in the directory, we must reclaim
892 * the space of the now empty record by adding the record size
893 * to the size of the previous entry.
894 */
895 int
896 ext2fs_dirremove(dvp, cnp)
897 struct vnode *dvp;
898 struct componentname *cnp;
899 {
900 struct inode *dp;
901 struct ext2fs_direct *ep;
902 struct buf *bp;
903 int error;
904
905 dp = VTOI(dvp);
906 if (dp->i_count == 0) {
907 /*
908 * First entry in block: set d_ino to zero.
909 */
910 error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset,
911 (void *)&ep, &bp);
912 if (error != 0)
913 return (error);
914 ep->e2d_ino = 0;
915 error = VOP_BWRITE(bp);
916 dp->i_flag |= IN_CHANGE | IN_UPDATE;
917 return (error);
918 }
919 /*
920 * Collapse new free space into previous entry.
921 */
922 error = VOP_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
923 (void *)&ep, &bp);
924 if (error != 0)
925 return (error);
926 ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + dp->i_reclen);
927 error = VOP_BWRITE(bp);
928 dp->i_flag |= IN_CHANGE | IN_UPDATE;
929 return (error);
930 }
931
932 /*
933 * Rewrite an existing directory entry to point at the inode
934 * supplied. The parameters describing the directory entry are
935 * set up by a call to namei.
936 */
937 int
938 ext2fs_dirrewrite(dp, ip, cnp)
939 struct inode *dp, *ip;
940 struct componentname *cnp;
941 {
942 struct buf *bp;
943 struct ext2fs_direct *ep;
944 struct vnode *vdp = ITOV(dp);
945 int error;
946
947 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, (void *)&ep, &bp);
948 if (error != 0)
949 return (error);
950 ep->e2d_ino = h2fs32(ip->i_number);
951 if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
952 (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
953 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
954 } else {
955 ep->e2d_type = 0;
956 }
957 error = VOP_BWRITE(bp);
958 dp->i_flag |= IN_CHANGE | IN_UPDATE;
959 return (error);
960 }
961
962 /*
963 * Check if a directory is empty or not.
964 * Inode supplied must be locked.
965 *
966 * Using a struct dirtemplate here is not precisely
967 * what we want, but better than using a struct ext2fs_direct.
968 *
969 * NB: does not handle corrupted directories.
970 */
971 int
972 ext2fs_dirempty(ip, parentino, cred)
973 struct inode *ip;
974 ino_t parentino;
975 struct ucred *cred;
976 {
977 off_t off;
978 struct ext2fs_dirtemplate dbuf;
979 struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
980 int error, namlen;
981 size_t count;
982
983 #define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
984
985 for (off = 0; off < ip->i_e2fs_size; off += fs2h16(dp->e2d_reclen)) {
986 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
987 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
988 /*
989 * Since we read MINDIRSIZ, residual must
990 * be 0 unless we're at end of file.
991 */
992 if (error || count != 0)
993 return (0);
994 /* avoid infinite loops */
995 if (dp->e2d_reclen == 0)
996 return (0);
997 /* skip empty entries */
998 if (dp->e2d_ino == 0)
999 continue;
1000 /* accept only "." and ".." */
1001 namlen = dp->e2d_namlen;
1002 if (namlen > 2)
1003 return (0);
1004 if (dp->e2d_name[0] != '.')
1005 return (0);
1006 /*
1007 * At this point namlen must be 1 or 2.
1008 * 1 implies ".", 2 implies ".." if second
1009 * char is also "."
1010 */
1011 if (namlen == 1)
1012 continue;
1013 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1014 continue;
1015 return (0);
1016 }
1017 return (1);
1018 }
1019
1020 /*
1021 * Check if source directory is in the path of the target directory.
1022 * Target is supplied locked, source is unlocked.
1023 * The target is always vput before returning.
1024 */
1025 int
1026 ext2fs_checkpath(source, target, cred)
1027 struct inode *source, *target;
1028 struct ucred *cred;
1029 {
1030 struct vnode *vp;
1031 int error, rootino, namlen;
1032 struct ext2fs_dirtemplate dirbuf;
1033 u_int32_t ino;
1034
1035 vp = ITOV(target);
1036 if (target->i_number == source->i_number) {
1037 error = EEXIST;
1038 goto out;
1039 }
1040 rootino = ROOTINO;
1041 error = 0;
1042 if (target->i_number == rootino)
1043 goto out;
1044
1045 for (;;) {
1046 if (vp->v_type != VDIR) {
1047 error = ENOTDIR;
1048 break;
1049 }
1050 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1051 sizeof (struct ext2fs_dirtemplate), (off_t)0,
1052 UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1053 (struct proc *)0);
1054 if (error != 0)
1055 break;
1056 namlen = dirbuf.dotdot_namlen;
1057 if (namlen != 2 ||
1058 dirbuf.dotdot_name[0] != '.' ||
1059 dirbuf.dotdot_name[1] != '.') {
1060 error = ENOTDIR;
1061 break;
1062 }
1063 ino = fs2h32(dirbuf.dotdot_ino);
1064 if (ino == source->i_number) {
1065 error = EINVAL;
1066 break;
1067 }
1068 if (ino == rootino)
1069 break;
1070 vput(vp);
1071 error = VFS_VGET(vp->v_mount, ino, &vp);
1072 if (error != 0) {
1073 vp = NULL;
1074 break;
1075 }
1076 }
1077
1078 out:
1079 if (error == ENOTDIR) {
1080 printf("checkpath: .. not a directory\n");
1081 panic("checkpath");
1082 }
1083 if (vp != NULL)
1084 vput(vp);
1085 return (error);
1086 }
Cache object: d212240389f206d4cf4e95590dfb5fdf
|