1 /* $NetBSD: ext2fs_lookup.c,v 1.95 2024/09/08 09:36:52 rillig 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 1.95 2024/09/08 09:36:52 rillig 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/kmem.h>
61 #include <sys/malloc.h>
62 #include <sys/dirent.h>
63 #include <sys/kauth.h>
64 #include <sys/proc.h>
65
66 #include <ufs/ufs/inode.h>
67 #include <ufs/ufs/ufsmount.h>
68 #include <ufs/ufs/ufs_extern.h>
69
70 #include <ufs/ext2fs/ext2fs_extern.h>
71 #include <ufs/ext2fs/ext2fs_dir.h>
72 #include <ufs/ext2fs/ext2fs.h>
73 #include <ufs/ext2fs/ext2fs_htree.h>
74
75 #include <miscfs/genfs/genfs.h>
76
77 extern int dirchk;
78
79 static void ext2fs_dirconv2ffs(struct m_ext2fs *fs,
80 struct ext2fs_direct *e2dir,
81 struct dirent *ffsdir);
82 static int ext2fs_dirbadentry(struct vnode *dp,
83 struct ext2fs_direct *de,
84 int entryoffsetinblock);
85
86 /*
87 * the problem that is tackled below is the fact that FFS
88 * includes the terminating zero on disk while EXT2FS doesn't
89 * this implies that we need to introduce some padding.
90 * For instance, a filename "sbin" has normally a reclen 12
91 * in EXT2, but 16 in FFS.
92 * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
93 * If it wasn't for that, the complete ufs code for directories would
94 * have worked w/o changes (except for the difference in DIRBLKSIZ)
95 */
96 static void
ext2fs_dirconv2ffs(struct m_ext2fs * fs,struct ext2fs_direct * e2dir,struct dirent * ffsdir)97 ext2fs_dirconv2ffs(struct m_ext2fs *fs, struct ext2fs_direct *e2dir, struct dirent *ffsdir)
98 {
99 memset(ffsdir, 0, sizeof(struct dirent));
100 ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
101 ffsdir->d_namlen = e2dir->e2d_namlen;
102
103 if (EXT2F_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FTYPE))
104 ffsdir->d_type = ext2dt2dt(e2dir->e2d_type);
105 else
106 ffsdir->d_type = DT_UNKNOWN;
107
108 #ifdef DIAGNOSTIC
109 #if MAXNAMLEN < E2FS_MAXNAMLEN
110 /*
111 * we should handle this more gracefully !
112 */
113 if (e2dir->e2d_namlen > MAXNAMLEN)
114 panic("ext2fs: e2dir->e2d_namlen");
115 #endif
116 #endif
117 strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
118
119 /* Godmar thinks: since e2dir->e2d_reclen can be big and means
120 nothing anyway, we compute our own reclen according to what
121 we think is right
122 */
123 ffsdir->d_reclen = _DIRENT_SIZE(ffsdir);
124 }
125
126 static int
ext2fs_is_dot_entry(struct componentname * cnp)127 ext2fs_is_dot_entry(struct componentname *cnp)
128 {
129 return cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
130 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0');
131 }
132
133 /*
134 * Vnode op for reading directories.
135 *
136 * Convert the on-disk entries to <sys/dirent.h> entries.
137 * the problem is that the conversion will blow up some entries by four bytes,
138 * so it can't be done in place. This is too bad. Right now the conversion is
139 * done entry by entry, the converted entry is sent via uiomove.
140 *
141 * XXX allocate a buffer, convert as many entries as possible, then send
142 * the whole buffer to uiomove
143 */
144 int
ext2fs_readdir(void * v)145 ext2fs_readdir(void *v)
146 {
147 struct vop_readdir_args /* {
148 struct vnode *a_vp;
149 struct uio *a_uio;
150 kauth_cred_t a_cred;
151 int **a_eofflag;
152 off_t **a_cookies;
153 int ncookies;
154 } */ *ap = v;
155 struct uio *uio = ap->a_uio;
156 int error;
157 size_t e2fs_count, readcnt;
158 struct vnode *vp = ap->a_vp;
159 struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
160
161 struct ext2fs_direct *dp;
162 struct dirent *dstd;
163 struct uio auio;
164 struct iovec aiov;
165 void *dirbuf;
166 off_t off = uio->uio_offset;
167 off_t *cookies = NULL;
168 int nc = 0, ncookies = 0;
169 int e2d_reclen;
170
171 if (vp->v_type != VDIR)
172 return ENOTDIR;
173
174 e2fs_count = uio->uio_resid;
175 /* Make sure we don't return partial entries. */
176 e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
177 if (e2fs_count <= 0)
178 return EINVAL;
179
180 auio = *uio;
181 auio.uio_iov = &aiov;
182 auio.uio_iovcnt = 1;
183 aiov.iov_len = e2fs_count;
184 auio.uio_resid = e2fs_count;
185 UIO_SETUP_SYSSPACE(&auio);
186 dirbuf = kmem_alloc(e2fs_count, KM_SLEEP);
187 dstd = kmem_zalloc(sizeof(struct dirent), KM_SLEEP);
188 if (ap->a_ncookies) {
189 nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
190 ncookies = nc;
191 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
192 *ap->a_cookies = cookies;
193 }
194 aiov.iov_base = dirbuf;
195
196 error = UFS_BUFRD(ap->a_vp, &auio, 0, ap->a_cred);
197 if (error == 0) {
198 readcnt = e2fs_count - auio.uio_resid;
199 for (dp = (struct ext2fs_direct *)dirbuf;
200 (char *)dp < (char *)dirbuf + readcnt;) {
201 e2d_reclen = fs2h16(dp->e2d_reclen);
202 if (e2d_reclen == 0) {
203 error = EIO;
204 break;
205 }
206 ext2fs_dirconv2ffs(fs, dp, dstd);
207 if(dstd->d_reclen > uio->uio_resid) {
208 break;
209 }
210 error = uiomove(dstd, dstd->d_reclen, uio);
211 if (error != 0) {
212 break;
213 }
214 off = off + e2d_reclen;
215 if (cookies != NULL) {
216 *cookies++ = off;
217 if (--ncookies <= 0){
218 break; /* out of cookies */
219 }
220 }
221 /* advance dp */
222 dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
223 }
224 /* we need to correct uio_offset */
225 uio->uio_offset = off;
226 }
227 kmem_free(dirbuf, e2fs_count);
228 kmem_free(dstd, sizeof(*dstd));
229 *ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
230 if (ap->a_ncookies) {
231 if (error) {
232 free(*ap->a_cookies, M_TEMP);
233 *ap->a_ncookies = 0;
234 *ap->a_cookies = NULL;
235 } else
236 *ap->a_ncookies = nc - ncookies;
237 }
238 return error;
239 }
240
241 /*
242 * Convert a component of a pathname into a pointer to a locked inode.
243 * This is a very central and rather complicated routine.
244 * If the file system is not maintained in a strict tree hierarchy,
245 * this can result in a deadlock situation (see comments in code below).
246 *
247 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
248 * on whether the name is to be looked up, created, renamed, or deleted.
249 * When CREATE, RENAME, or DELETE is specified, information usable in
250 * creating, renaming, or deleting a directory entry may be calculated.
251 * If flag has LOCKPARENT or'ed into it and the target of the pathname
252 * exists, lookup returns both the target and its parent directory locked.
253 * When creating or renaming and LOCKPARENT is specified, the target may
254 * not be ".". When deleting and LOCKPARENT is specified, the target may
255 * be ".", but the caller must check to ensure it does a vrele and vput
256 * instead of two vputs.
257 *
258 * Overall outline of ext2fs_lookup:
259 *
260 * check accessibility of directory
261 * look for name in cache, if found, then if at end of path
262 * and deleting or creating, drop it, else return name
263 * search for name in directory, to found or notfound
264 * notfound:
265 * if creating, return locked directory, leaving info on available slots
266 * else return error
267 * found:
268 * if at end of path and deleting, return information to allow delete
269 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
270 * inode and return info to allow rewrite
271 * if not at end, add name to cache; if at end and neither creating
272 * nor deleting, add name to cache
273 */
274 int
ext2fs_lookup(void * v)275 ext2fs_lookup(void *v)
276 {
277 struct vop_lookup_v2_args /* {
278 struct vnode *a_dvp;
279 struct vnode **a_vpp;
280 struct componentname *a_cnp;
281 } */ *ap = v;
282 struct vnode *vdp = ap->a_dvp; /* vnode for directory being searched */
283 struct inode *dp = VTOI(vdp); /* inode for directory being searched */
284 struct buf *bp; /* a buffer of directory entries */
285 struct ext2fs_direct *ep; /* the current directory entry */
286 int entryoffsetinblock; /* offset of ep in bp's buffer */
287 enum ext2fs_slotstatus slotstatus;
288 doff_t slotoffset; /* offset of area with free space */
289 int slotsize; /* size of area at slotoffset */
290 int slotfreespace; /* amount of space free in slot */
291 int slotneeded; /* size of the entry we're seeking */
292 int numdirpasses; /* strategy for directory search */
293 doff_t endsearch; /* offset to end directory search */
294 doff_t prevoff; /* prev entry dp->i_offset */
295 struct vnode *tdp; /* returned by vcache_get */
296 doff_t enduseful; /* pointer past last used dir slot */
297 u_long bmask; /* block offset mask */
298 int namlen, error;
299 struct vnode **vpp = ap->a_vpp;
300 struct componentname *cnp = ap->a_cnp;
301 kauth_cred_t cred = cnp->cn_cred;
302 int flags;
303 int nameiop = cnp->cn_nameiop;
304 struct ufsmount *ump = dp->i_ump;
305 int dirblksiz = ump->um_dirblksiz;
306 ino_t foundino;
307 struct ufs_lookup_results *results;
308
309 flags = cnp->cn_flags;
310
311 bp = NULL;
312 slotoffset = -1;
313 *vpp = NULL;
314
315 /*
316 * Check accessibility of directory.
317 */
318 if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
319 return error;
320
321 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
322 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
323 return EROFS;
324
325 /*
326 * We now have a segment name to search for, and a directory to search.
327 *
328 * Before tediously performing a linear scan of the directory,
329 * check the name cache to see if the directory/name pair
330 * we are looking for is known already.
331 */
332 if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
333 cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
334 return *vpp == NULLVP ? ENOENT : 0;
335 }
336
337 /* May need to restart the lookup with an exclusive lock. */
338 if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE)
339 return ENOLCK;
340
341 /*
342 * Produce the auxiliary lookup results into i_crap. Increment
343 * its serial number so elsewhere we can tell if we're using
344 * stale results. This should not be done this way. XXX.
345 */
346 results = &dp->i_crap;
347 dp->i_crapcounter++;
348
349 /*
350 * Suppress search for slots unless creating
351 * file and at end of pathname, in which case
352 * we watch for a place to put the new file in
353 * case it doesn't already exist.
354 */
355 slotstatus = FOUND;
356 slotfreespace = slotsize = slotneeded = 0;
357 if ((nameiop == CREATE || nameiop == RENAME) &&
358 (flags & ISLASTCN)) {
359 slotstatus = NONE;
360 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
361 }
362
363 /*
364 * If there is cached information on a previous search of
365 * this directory, pick up where we last left off.
366 * We cache only lookups as these are the most common
367 * and have the greatest payoff. Caching CREATE has little
368 * benefit as it usually must search the entire directory
369 * to determine that the entry does not exist. Caching the
370 * location of the last DELETE or RENAME has not reduced
371 * profiling time and hence has been removed in the interest
372 * of simplicity.
373 */
374 bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
375 if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
376 results->ulr_diroff >= ext2fs_size(dp)) {
377 entryoffsetinblock = 0;
378 results->ulr_offset = 0;
379 numdirpasses = 1;
380 } else {
381 results->ulr_offset = results->ulr_diroff;
382 if ((entryoffsetinblock = results->ulr_offset & bmask) &&
383 (error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
384 return error;
385 numdirpasses = 2;
386 namecache_count_2passes();
387 }
388 prevoff = results->ulr_offset;
389 endsearch = roundup(ext2fs_size(dp), dirblksiz);
390 enduseful = 0;
391 /*
392 * Try to lookup dir entry using htree directory index.
393 *
394 * If we got an error or we want to find '.' or '..' entry,
395 * we will fall back to linear search.
396 */
397 if (!ext2fs_is_dot_entry(cnp) && ext2fs_htree_has_idx(dp)) {
398 doff_t i_offset; /* cached i_offset value */
399 struct ext2fs_searchslot ss;
400 numdirpasses = 1;
401 entryoffsetinblock = 0;
402
403 int htree_lookup_ret = ext2fs_htree_lookup(dp, cnp->cn_nameptr,
404 cnp->cn_namelen, &bp, &entryoffsetinblock, &i_offset,
405 &prevoff, &enduseful, &ss);
406 switch (htree_lookup_ret) {
407 case 0:
408 ep = (void *)((char *)bp->b_data + (i_offset & bmask));
409 foundino = ep->e2d_ino;
410 goto found;
411 case ENOENT:
412 i_offset = roundup2(dp->i_size, dp->i_e2fs->e2fs_bsize);
413 goto notfound;
414 default:
415 /*
416 * Something failed; just fallback to do a linear
417 * search.
418 */
419 break;
420 }
421 }
422
423 searchloop:
424 while (results->ulr_offset < endsearch) {
425 preempt_point();
426
427 /*
428 * If necessary, get the next directory block.
429 */
430 if ((results->ulr_offset & bmask) == 0) {
431 if (bp != NULL)
432 brelse(bp, 0);
433 error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset,
434 NULL, &bp);
435 if (error != 0)
436 return error;
437 entryoffsetinblock = 0;
438 }
439 /*
440 * If still looking for a slot, and at a dirblksize
441 * boundary, have to start looking for free space again.
442 */
443 if (slotstatus == NONE &&
444 (entryoffsetinblock & (dirblksiz - 1)) == 0) {
445 slotoffset = -1;
446 slotfreespace = 0;
447 }
448 /*
449 * Get pointer to next entry.
450 * Full validation checks are slow, so we only check
451 * enough to insure forward progress through the
452 * directory. Complete checks can be run by patching
453 * "dirchk" to be true.
454 */
455 KASSERT(bp != NULL);
456 ep = (struct ext2fs_direct *)
457 ((char *)bp->b_data + entryoffsetinblock);
458 if (ep->e2d_reclen == 0 ||
459 (dirchk &&
460 ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
461 int i;
462
463 ufs_dirbad(dp, results->ulr_offset, "mangled entry");
464 i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
465 results->ulr_offset += i;
466 entryoffsetinblock += i;
467 continue;
468 }
469
470 /*
471 * If an appropriate sized slot has not yet been found,
472 * check to see if one is available. Also accumulate space
473 * in the current block so that we can determine if
474 * compaction is viable.
475 */
476 if (slotstatus != FOUND) {
477 int size = fs2h16(ep->e2d_reclen);
478
479 if (ep->e2d_ino != 0)
480 size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
481 if (size > 0) {
482 if (size >= slotneeded) {
483 slotstatus = FOUND;
484 slotoffset = results->ulr_offset;
485 slotsize = fs2h16(ep->e2d_reclen);
486 } else if (slotstatus == NONE) {
487 slotfreespace += size;
488 if (slotoffset == -1)
489 slotoffset = results->ulr_offset;
490 if (slotfreespace >= slotneeded) {
491 slotstatus = COMPACT;
492 slotsize = results->ulr_offset +
493 fs2h16(ep->e2d_reclen) -
494 slotoffset;
495 }
496 }
497 }
498 }
499
500 /*
501 * Check for a name match.
502 */
503 if (ep->e2d_ino) {
504 namlen = ep->e2d_namlen;
505 if (namlen == cnp->cn_namelen &&
506 !memcmp(cnp->cn_nameptr, ep->e2d_name,
507 (unsigned)namlen)) {
508 /*
509 * Save directory entry's inode number and
510 * reclen in ndp->ni_ufs area, and release
511 * directory buffer.
512 */
513 foundino = fs2h32(ep->e2d_ino);
514 results->ulr_reclen = fs2h16(ep->e2d_reclen);
515 goto found;
516 }
517 }
518 prevoff = results->ulr_offset;
519 results->ulr_offset += fs2h16(ep->e2d_reclen);
520 entryoffsetinblock += fs2h16(ep->e2d_reclen);
521 if (ep->e2d_ino)
522 enduseful = results->ulr_offset;
523 }
524 notfound:
525 /*
526 * If we started in the middle of the directory and failed
527 * to find our target, we must check the beginning as well.
528 */
529 if (numdirpasses == 2) {
530 numdirpasses--;
531 results->ulr_offset = 0;
532 endsearch = results->ulr_diroff;
533 goto searchloop;
534 }
535 if (bp != NULL)
536 brelse(bp, 0);
537 /*
538 * If creating, and at end of pathname and current
539 * directory has not been removed, then can consider
540 * allowing file to be created.
541 */
542 if ((nameiop == CREATE || nameiop == RENAME) &&
543 (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
544 /*
545 * Access for write is interpreted as allowing
546 * creation of files in the directory.
547 */
548 error = VOP_ACCESS(vdp, VWRITE, cred);
549 if (error)
550 return error;
551 /*
552 * Return an indication of where the new directory
553 * entry should be put. If we didn't find a slot,
554 * then set results->ulr_count to 0 indicating
555 * that the new slot belongs at the end of the
556 * directory. If we found a slot, then the new entry
557 * can be put in the range from results->ulr_offset to
558 * results->ulr_offset + results->ulr_count.
559 */
560 if (slotstatus == NONE) {
561 results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
562 results->ulr_count = 0;
563 enduseful = results->ulr_offset;
564 } else {
565 results->ulr_offset = slotoffset;
566 results->ulr_count = slotsize;
567 if (enduseful < slotoffset + slotsize)
568 enduseful = slotoffset + slotsize;
569 }
570 results->ulr_endoff = roundup(enduseful, dirblksiz);
571 #if 0
572 dp->i_flag |= IN_CHANGE | IN_UPDATE;
573 #endif
574 /*
575 * We return with the directory locked, so that
576 * the parameters we set up above will still be
577 * valid if we actually decide to do a direnter().
578 * We return ni_vp == NULL to indicate that the entry
579 * does not currently exist; we leave a pointer to
580 * the (locked) directory inode in ndp->ni_dvp.
581 *
582 * NB - if the directory is unlocked, then this
583 * information cannot be used.
584 */
585 return EJUSTRETURN;
586 }
587 /*
588 * Insert name into cache (as non-existent) if appropriate.
589 */
590 if (nameiop != CREATE) {
591 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
592 cnp->cn_flags);
593 }
594 return ENOENT;
595
596 found:
597 if (numdirpasses == 2)
598 namecache_count_pass2();
599 /*
600 * Check that directory length properly reflects presence
601 * of this entry.
602 */
603 if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
604 ufs_dirbad(dp, results->ulr_offset, "i_size too small");
605 error = ext2fs_setsize(dp,
606 results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
607 if (error) {
608 brelse(bp, 0);
609 return error;
610 }
611 dp->i_flag |= IN_CHANGE | IN_UPDATE;
612 uvm_vnp_setsize(vdp, ext2fs_size(dp));
613 }
614 brelse(bp, 0);
615
616 /*
617 * Found component in pathname.
618 * If the final component of path name, save information
619 * in the cache as to where the entry was found.
620 */
621 if ((flags & ISLASTCN) && nameiop == LOOKUP)
622 results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
623
624 /*
625 * If deleting, and at end of pathname, return
626 * parameters which can be used to remove file.
627 * Lock the inode, being careful with ".".
628 */
629 if (nameiop == DELETE && (flags & ISLASTCN)) {
630 /*
631 * Return pointer to current entry in results->ulr_offset,
632 * and distance past previous entry (if there
633 * is a previous entry in this block) in results->ulr_count.
634 * Save directory inode pointer in ndp->ni_dvp for dirremove().
635 */
636 if ((results->ulr_offset & (dirblksiz - 1)) == 0)
637 results->ulr_count = 0;
638 else
639 results->ulr_count = results->ulr_offset - prevoff;
640 if (dp->i_number == foundino) {
641 vref(vdp);
642 tdp = vdp;
643 } else {
644 error = vcache_get(vdp->v_mount,
645 &foundino, sizeof(foundino), &tdp);
646 if (error)
647 return error;
648 }
649 /*
650 * Write access to directory required to delete files.
651 */
652 if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
653 vrele(tdp);
654 return error;
655 }
656 /*
657 * If directory is "sticky", then user must own
658 * the directory, or the file in it, else she
659 * may not delete it (unless she's root). This
660 * implements append-only directories.
661 */
662 if (dp->i_e2fs_mode & ISVTX) {
663 error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
664 tdp, vdp, genfs_can_sticky(vdp, cred, dp->i_uid,
665 VTOI(tdp)->i_uid));
666 if (error) {
667 vrele(tdp);
668 return EPERM;
669 }
670 }
671 *vpp = tdp;
672 return 0;
673 }
674
675 /*
676 * If rewriting (RENAME), return the inode and the
677 * information required to rewrite the present directory
678 * Must get inode of directory entry to verify it's a
679 * regular file, or empty directory.
680 */
681 if (nameiop == RENAME && (flags & ISLASTCN)) {
682 error = VOP_ACCESS(vdp, VWRITE, cred);
683 if (error)
684 return error;
685 /*
686 * Careful about locking second inode.
687 * This can only occur if the target is ".".
688 */
689 if (dp->i_number == foundino)
690 return EISDIR;
691 error = vcache_get(vdp->v_mount,
692 &foundino, sizeof(foundino), &tdp);
693 if (error)
694 return error;
695 *vpp = tdp;
696 return 0;
697 }
698
699 if (dp->i_number == foundino) {
700 vref(vdp); /* we want ourself, ie "." */
701 *vpp = vdp;
702 } else {
703 error = vcache_get(vdp->v_mount,
704 &foundino, sizeof(foundino), &tdp);
705 if (error)
706 return error;
707 *vpp = tdp;
708 }
709
710 /*
711 * Insert name into cache if appropriate.
712 */
713 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
714 return 0;
715 }
716 static void
ext2fs_accumulatespace(struct ext2fs_searchslot * ssp,struct ext2fs_direct * ep,doff_t * offp)717 ext2fs_accumulatespace (struct ext2fs_searchslot *ssp, struct ext2fs_direct *ep,
718 doff_t *offp)
719 {
720 int size = ep->e2d_reclen;
721
722 if (ep->e2d_ino != 0)
723 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
724
725 if (size <= 0)
726 return;
727
728 if (size >= ssp->slotneeded) {
729 ssp->slotstatus = FOUND;
730 ssp->slotoffset = *offp;
731 ssp->slotsize = ep->e2d_reclen;
732 return;
733 }
734
735 if (ssp->slotstatus != NONE)
736 return;
737
738 ssp->slotfreespace += size;
739 if (ssp->slotoffset == -1)
740 ssp->slotoffset = *offp;
741
742 if (ssp->slotfreespace >= ssp->slotneeded) {
743 ssp->slotstatus = COMPACT;
744 ssp->slotsize = *offp + ep->e2d_reclen - ssp->slotoffset;
745 }
746 }
747
748 int
ext2fs_search_dirblock(struct inode * ip,void * data,int * foundp,const char * name,int namelen,int * entryoffsetinblockp,doff_t * offp,doff_t * prevoffp,doff_t * endusefulp,struct ext2fs_searchslot * ssp)749 ext2fs_search_dirblock(struct inode *ip, void *data, int *foundp,
750 const char *name, int namelen, int *entryoffsetinblockp,
751 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
752 struct ext2fs_searchslot *ssp)
753 {
754 struct vnode *vdp = ITOV(ip);
755 struct ext2fs_direct *ep, *top;
756 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
757 int offset = *entryoffsetinblockp;
758 int namlen;
759
760 ep = (void *)((char *)data + offset);
761 top = (void *)((char *)data + bsize - EXT2_DIR_REC_LEN(0));
762
763 while (ep < top) {
764 /*
765 * Full validation checks are slow, so we only check
766 * enough to insure forward progress through the
767 * directory. Complete checks can be run by setting
768 * "vfs.e2fs.dirchk" to be true.
769 */
770 if (ep->e2d_reclen == 0 ||
771 (dirchk && ext2fs_dirbadentry(vdp, ep, offset))) {
772 int i;
773 ufs_dirbad(ip, *offp, "mangled entry");
774 i = bsize - (offset & (bsize - 1));
775 *offp += i;
776 offset += i;
777 continue;
778 }
779
780 /*
781 * If an appropriate sized slot has not yet been found,
782 * check to see if one is available. Also accumulate space
783 * in the current block so that we can determine if
784 * compaction is viable.
785 */
786 if (ssp->slotstatus != FOUND)
787 ext2fs_accumulatespace(ssp, ep, offp);
788
789 /*
790 * Check for a name match.
791 */
792 if (ep->e2d_ino) {
793 namlen = ep->e2d_namlen;
794 if (namlen == namelen &&
795 !memcmp(name, ep->e2d_name, (unsigned)namlen)) {
796 /*
797 * Save directory entry's inode number and
798 * reclen in ndp->ni_ufs area, and release
799 * directory buffer.
800 */
801 *foundp = 1;
802 return 0;
803 }
804 }
805 *prevoffp = *offp;
806 *offp += ep->e2d_reclen;
807 offset += ep->e2d_reclen;
808 *entryoffsetinblockp = offset;
809 if (ep->e2d_ino)
810 *endusefulp = *offp;
811 /*
812 * Get pointer to the next entry.
813 */
814 ep = (void *)((char *)data + offset);
815 }
816
817 return 0;
818 }
819
820 /*
821 * Do consistency checking on a directory entry:
822 * record length must be multiple of 4
823 * entry must fit in rest of its dirblksize block
824 * record must be large enough to contain entry
825 * name is not longer than EXT2FS_MAXNAMLEN
826 * name must be as long as advertised, and null terminated
827 */
828 /*
829 * changed so that it confirms to ext2fs_check_dir_entry
830 */
831 static int
ext2fs_dirbadentry(struct vnode * dp,struct ext2fs_direct * de,int entryoffsetinblock)832 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
833 int entryoffsetinblock)
834 {
835 struct ufsmount *ump = VFSTOUFS(dp->v_mount);
836 int dirblksiz = ump->um_dirblksiz;
837
838 const char *error_msg = NULL;
839 int reclen = fs2h16(de->e2d_reclen);
840 int namlen = de->e2d_namlen;
841
842 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
843 error_msg = "rec_len is smaller than minimal";
844 else if (reclen % 4 != 0)
845 error_msg = "rec_len % 4 != 0";
846 else if (namlen > EXT2FS_MAXNAMLEN)
847 error_msg = "namlen > EXT2FS_MAXNAMLEN";
848 else if (reclen < EXT2FS_DIRSIZ(namlen))
849 error_msg = "reclen is too small for name_len";
850 else if (entryoffsetinblock + reclen > dirblksiz)
851 error_msg = "directory entry across blocks";
852 else if (fs2h32(de->e2d_ino) >
853 VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
854 error_msg = "inode out of bounds";
855
856 if (error_msg != NULL) {
857 printf( "bad directory entry: %s\n"
858 "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
859 error_msg, entryoffsetinblock,
860 (unsigned long) fs2h32(de->e2d_ino),
861 reclen, namlen);
862 panic("ext2fs_dirbadentry");
863 }
864 return error_msg == NULL ? 0 : 1;
865 }
866
867 /*
868 * Write a directory entry after a call to namei, using the parameters
869 * that it left in nameidata. The argument ip is the inode which the new
870 * directory entry will refer to. Dvp is a pointer to the directory to
871 * be written, which was left locked by namei. Remaining parameters
872 * (ulr_offset, ulr_count) indicate how the space for the new
873 * entry is to be obtained.
874 */
875 int
ext2fs_direnter(struct inode * ip,struct vnode * dvp,const struct ufs_lookup_results * ulr,struct componentname * cnp)876 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
877 const struct ufs_lookup_results *ulr, struct componentname *cnp)
878 {
879 struct inode *dp;
880 struct ext2fs_direct newdir;
881 struct iovec aiov;
882 struct uio auio;
883 int error;
884 struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
885 int dirblksiz = ump->um_dirblksiz;
886 size_t newentrysize;
887
888 dp = VTOI(dvp);
889
890 newdir.e2d_ino = h2fs32(ip->i_number);
891 newdir.e2d_namlen = cnp->cn_namelen;
892 if (EXT2F_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) {
893 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
894 } else {
895 newdir.e2d_type = 0;
896 }
897 memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
898 newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
899
900 if (ext2fs_htree_has_idx(dp)) {
901 error = ext2fs_htree_add_entry(dvp, &newdir, cnp, newentrysize);
902 if (error) {
903 dp->i_e2fs_flags &= ~EXT2_INDEX;
904 dp->i_flag |= IN_CHANGE | IN_UPDATE;
905 }
906 return error;
907 }
908
909 /*
910 * TODO check if Htree index is not created for the directory then
911 * create one if directory entries get overflew the first dir-block
912 */
913 if (ulr->ulr_count == 0) {
914 /*
915 * If ulr_count is 0, then namei could find no
916 * space in the directory. Here, ulr_offset will
917 * be on a directory block boundary and we will write the
918 * new entry into a fresh block.
919 */
920 if (ulr->ulr_offset & (dirblksiz - 1))
921 panic("ext2fs_direnter: newblk");
922 auio.uio_offset = ulr->ulr_offset;
923 newdir.e2d_reclen = h2fs16(dirblksiz);
924 auio.uio_resid = newentrysize;
925 aiov.iov_len = newentrysize;
926 aiov.iov_base = (void *)&newdir;
927 auio.uio_iov = &aiov;
928 auio.uio_iovcnt = 1;
929 auio.uio_rw = UIO_WRITE;
930 UIO_SETUP_SYSSPACE(&auio);
931 error = ext2fs_bufwr(dvp, &auio, IO_SYNC, cnp->cn_cred);
932 if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
933 /* XXX should grow with balloc() */
934 panic("ext2fs_direnter: frag size");
935 else if (!error) {
936 error = ext2fs_setsize(dp,
937 roundup(ext2fs_size(dp), dirblksiz));
938 if (error)
939 return error;
940 dp->i_flag |= IN_CHANGE;
941 uvm_vnp_setsize(dvp, ext2fs_size(dp));
942 }
943 return error;
944 }
945
946 error = ext2fs_add_entry(dvp, &newdir, ulr, newentrysize);
947
948 if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
949 error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
950 cnp->cn_cred);
951 return error;
952 }
953
954 /*
955 * Insert an entry into the directory block.
956 * Compact the contents.
957 */
958
959 int
ext2fs_add_entry(struct vnode * dvp,struct ext2fs_direct * entry,const struct ufs_lookup_results * ulr,size_t newentrysize)960 ext2fs_add_entry(struct vnode* dvp, struct ext2fs_direct *entry,
961 const struct ufs_lookup_results *ulr, size_t newentrysize)
962 {
963 struct ext2fs_direct *ep, *nep;
964 struct inode *dp;
965 struct buf *bp;
966 u_int dsize;
967 int error, loc, spacefree;
968 char *dirbuf;
969
970 dp = VTOI(dvp);
971
972 /*
973 * If ulr_count is non-zero, then namei found space
974 * for the new entry in the range ulr_offset to
975 * ulr_offset + ulr_count in the directory.
976 * To use this space, we may have to compact the entries located
977 * there, by copying them together towards the beginning of the
978 * block, leaving the free space in one usable chunk at the end.
979 */
980
981 /*
982 * Get the block containing the space for the new directory entry.
983 */
984 if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
985 return error;
986 /*
987 * Find space for the new entry. In the simple case, the entry at
988 * offset base will have the space. If it does not, then namei
989 * arranged that compacting the region ulr_offset to
990 * ulr_offset + ulr_count would yield the
991 * space.
992 */
993 ep = (struct ext2fs_direct *)dirbuf;
994 dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
995 spacefree = fs2h16(ep->e2d_reclen) - dsize;
996 for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count;) {
997 nep = (struct ext2fs_direct *)(dirbuf + loc);
998 if (ep->e2d_ino) {
999 /* trim the existing slot */
1000 ep->e2d_reclen = h2fs16(dsize);
1001 ep = (struct ext2fs_direct *)((char *)ep + dsize);
1002 } else {
1003 /* overwrite; nothing there; header is ours */
1004 spacefree += dsize;
1005 }
1006 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
1007 spacefree += fs2h16(nep->e2d_reclen) - dsize;
1008 loc += fs2h16(nep->e2d_reclen);
1009 memcpy((void *)ep, (void *)nep, dsize);
1010 }
1011 /*
1012 * Update the pointer fields in the previous entry (if any),
1013 * copy in the new entry, and write out the block.
1014 */
1015 if (ep->e2d_ino == 0) {
1016 #ifdef DIAGNOSTIC
1017 if (spacefree + dsize < newentrysize)
1018 panic("ext2fs_direnter: compact1");
1019 #endif
1020 entry->e2d_reclen = h2fs16(spacefree + dsize);
1021 } else {
1022 #ifdef DIAGNOSTIC
1023 if (spacefree < newentrysize) {
1024 printf("ext2fs_direnter: compact2 %u %u",
1025 (u_int)spacefree, (u_int)newentrysize);
1026 panic("ext2fs_direnter: compact2");
1027 }
1028 #endif
1029 entry->e2d_reclen = h2fs16(spacefree);
1030 ep->e2d_reclen = h2fs16(dsize);
1031 ep = (struct ext2fs_direct *)((char *)ep + dsize);
1032 }
1033 memcpy(ep, entry, (u_int)newentrysize);
1034 error = VOP_BWRITE(bp->b_vp, bp);
1035 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1036 return error;
1037 }
1038
1039 /*
1040 * Remove a directory entry after a call to namei, using
1041 * the auxiliary results it provided. The entry
1042 * ulr_offset contains the offset into the directory of the
1043 * entry to be eliminated. The ulr_count field contains the
1044 * size of the previous record in the directory. If this
1045 * is 0, the first entry is being deleted, so we need only
1046 * zero the inode number to mark the entry as free. If the
1047 * entry is not the first in the directory, we must reclaim
1048 * the space of the now empty record by adding the record size
1049 * to the size of the previous entry.
1050 */
1051 int
ext2fs_dirremove(struct vnode * dvp,const struct ufs_lookup_results * ulr,struct componentname * cnp)1052 ext2fs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
1053 struct componentname *cnp)
1054 {
1055 struct inode *dp;
1056 struct ext2fs_direct *ep;
1057 struct buf *bp;
1058 int error;
1059
1060 dp = VTOI(dvp);
1061
1062 if (ulr->ulr_count == 0) {
1063 /*
1064 * First entry in block: set d_ino to zero.
1065 */
1066 error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset,
1067 (void *)&ep, &bp);
1068 if (error != 0)
1069 return error;
1070 ep->e2d_ino = 0;
1071 error = VOP_BWRITE(bp->b_vp, bp);
1072 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1073 return error;
1074 }
1075 /*
1076 * Collapse new free space into previous entry.
1077 */
1078 error = ext2fs_blkatoff(dvp, (off_t)(ulr->ulr_offset - ulr->ulr_count),
1079 (void *)&ep, &bp);
1080 if (error != 0)
1081 return error;
1082 ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + ulr->ulr_reclen);
1083 error = VOP_BWRITE(bp->b_vp, bp);
1084 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1085 return error;
1086 }
1087
1088 /*
1089 * Rewrite an existing directory entry to point at the inode
1090 * supplied. The parameters describing the directory entry are
1091 * set up by a call to namei.
1092 */
1093 int
ext2fs_dirrewrite(struct inode * dp,const struct ufs_lookup_results * ulr,struct inode * ip,struct componentname * cnp)1094 ext2fs_dirrewrite(struct inode *dp, const struct ufs_lookup_results *ulr,
1095 struct inode *ip, struct componentname *cnp)
1096 {
1097 struct buf *bp;
1098 struct ext2fs_direct *ep;
1099 struct vnode *vdp = ITOV(dp);
1100 int error;
1101
1102 error = ext2fs_blkatoff(vdp, (off_t)ulr->ulr_offset, (void *)&ep, &bp);
1103 if (error != 0)
1104 return error;
1105 ep->e2d_ino = h2fs32(ip->i_number);
1106 if (EXT2F_HAS_INCOMPAT_FEATURE(dp->i_e2fs, EXT2F_INCOMPAT_FTYPE)) {
1107 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
1108 } else {
1109 ep->e2d_type = 0;
1110 }
1111 error = VOP_BWRITE(bp->b_vp, bp);
1112 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1113 return error;
1114 }
1115
1116 /*
1117 * Check if a directory is empty or not.
1118 * Inode supplied must be locked.
1119 *
1120 * Using a struct dirtemplate here is not precisely
1121 * what we want, but better than using a struct ext2fs_direct.
1122 *
1123 * NB: does not handle corrupted directories.
1124 */
1125 int
ext2fs_dirempty(struct inode * ip,ino_t parentino,kauth_cred_t cred)1126 ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
1127 {
1128 off_t off;
1129 struct ext2fs_direct dbuf;
1130 struct ext2fs_direct *dp = &dbuf;
1131 int error, namlen;
1132 size_t count;
1133
1134 #define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
1135
1136 for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
1137 error = ufs_bufio(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ,
1138 off, IO_NODELOCKED, cred, &count, NULL);
1139 /*
1140 * Since we read MINDIRSIZ, residual must
1141 * be 0 unless we're at end of file.
1142 */
1143 if (error || count != 0)
1144 return 0;
1145 /* avoid infinite loops */
1146 if (dp->e2d_reclen == 0)
1147 return 0;
1148 /* skip empty entries */
1149 if (dp->e2d_ino == 0)
1150 continue;
1151 /* accept only "." and ".." */
1152 namlen = dp->e2d_namlen;
1153 if (namlen > 2)
1154 return 0;
1155 if (dp->e2d_name[0] != '.')
1156 return 0;
1157 /*
1158 * At this point namlen must be 1 or 2.
1159 * 1 implies ".", 2 implies ".." if second
1160 * char is also "."
1161 */
1162 if (namlen == 1)
1163 continue;
1164 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1165 continue;
1166 return 0;
1167 }
1168 return 1;
1169 }
1170