1/* $NetBSD: ext2fs_lookup.c,v 1.88 2016/08/23 06:40:25 christos 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 1.88 2016/08/23 06:40:25 christos 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
77extern int dirchk;
78
79static void ext2fs_dirconv2ffs(struct m_ext2fs *fs,
80 struct ext2fs_direct *e2dir,
81 struct dirent *ffsdir);
82static 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 */
96static void
97ext2fs_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
126static int
127ext2fs_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 */
144int
145ext2fs_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 an 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 */
274int
275ext2fs_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 * Produce the auxiliary lookup results into i_crap. Increment
317 * its serial number so elsewhere we can tell if we're using
318 * stale results. This should not be done this way. XXX.
319 */
320 results = &dp->i_crap;
321 dp->i_crapcounter++;
322
323 /*
324 * Check accessiblity of directory.
325 */
326 if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
327 return error;
328
329 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
330 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
331 return EROFS;
332
333 /*
334 * We now have a segment name to search for, and a directory to search.
335 *
336 * Before tediously performing a linear scan of the directory,
337 * check the name cache to see if the directory/name pair
338 * we are looking for is known already.
339 */
340 if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
341 cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
342 return *vpp == NULLVP ? ENOENT : 0;
343 }
344
345 /*
346 * Suppress search for slots unless creating
347 * file and at end of pathname, in which case
348 * we watch for a place to put the new file in
349 * case it doesn't already exist.
350 */
351 slotstatus = FOUND;
352 slotfreespace = slotsize = slotneeded = 0;
353 if ((nameiop == CREATE || nameiop == RENAME) &&
354 (flags & ISLASTCN)) {
355 slotstatus = NONE;
356 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
357 }
358
359 /*
360 * If there is cached information on a previous search of
361 * this directory, pick up where we last left off.
362 * We cache only lookups as these are the most common
363 * and have the greatest payoff. Caching CREATE has little
364 * benefit as it usually must search the entire directory
365 * to determine that the entry does not exist. Caching the
366 * location of the last DELETE or RENAME has not reduced
367 * profiling time and hence has been removed in the interest
368 * of simplicity.
369 */
370 bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
371 if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
372 results->ulr_diroff >= ext2fs_size(dp)) {
373 entryoffsetinblock = 0;
374 results->ulr_offset = 0;
375 numdirpasses = 1;
376 } else {
377 results->ulr_offset = results->ulr_diroff;
378 if ((entryoffsetinblock = results->ulr_offset & bmask) &&
379 (error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
380 return error;
381 numdirpasses = 2;
382 namecache_count_2passes();
383 }
384 prevoff = results->ulr_offset;
385 endsearch = roundup(ext2fs_size(dp), dirblksiz);
386 enduseful = 0;
387 /*
388 * Try to lookup dir entry using htree directory index.
389 *
390 * If we got an error or we want to find '.' or '..' entry,
391 * we will fall back to linear search.
392 */
393 if (!ext2fs_is_dot_entry(cnp) && ext2fs_htree_has_idx(dp)) {
394 doff_t i_offset; /* cached i_offset value */
395 struct ext2fs_searchslot ss;
396 numdirpasses = 1;
397 entryoffsetinblock = 0;
398
399 int htree_lookup_ret = ext2fs_htree_lookup(dp, cnp->cn_nameptr,
400 cnp->cn_namelen, &bp, &entryoffsetinblock, &i_offset,
401 &prevoff, &enduseful, &ss);
402 switch (htree_lookup_ret) {
403 case 0:
404 ep = (void *)((char *)bp->b_data + (i_offset & bmask));
405 foundino = ep->e2d_ino;
406 goto found;
407 case ENOENT:
408 i_offset = roundup2(dp->i_size, dp->i_e2fs->e2fs_bsize);
409 goto notfound;
410 default:
411 /*
412 * Something failed; just fallback to do a linear
413 * search.
414 */
415 break;
416 }
417 }
418
419searchloop:
420 while (results->ulr_offset < endsearch) {
421 if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
422 preempt();
423 /*
424 * If necessary, get the next directory block.
425 */
426 if ((results->ulr_offset & bmask) == 0) {
427 if (bp != NULL)
428 brelse(bp, 0);
429 error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset,
430 NULL, &bp);
431 if (error != 0)
432 return error;
433 entryoffsetinblock = 0;
434 }
435 /*
436 * If still looking for a slot, and at a dirblksize
437 * boundary, have to start looking for free space again.
438 */
439 if (slotstatus == NONE &&
440 (entryoffsetinblock & (dirblksiz - 1)) == 0) {
441 slotoffset = -1;
442 slotfreespace = 0;
443 }
444 /*
445 * Get pointer to next entry.
446 * Full validation checks are slow, so we only check
447 * enough to insure forward progress through the
448 * directory. Complete checks can be run by patching
449 * "dirchk" to be true.
450 */
451 KASSERT(bp != NULL);
452 ep = (struct ext2fs_direct *)
453 ((char *)bp->b_data + entryoffsetinblock);
454 if (ep->e2d_reclen == 0 ||
455 (dirchk &&
456 ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
457 int i;
458
459 ufs_dirbad(dp, results->ulr_offset, "mangled entry");
460 i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
461 results->ulr_offset += i;
462 entryoffsetinblock += i;
463 continue;
464 }
465
466 /*
467 * If an appropriate sized slot has not yet been found,
468 * check to see if one is available. Also accumulate space
469 * in the current block so that we can determine if
470 * compaction is viable.
471 */
472 if (slotstatus != FOUND) {
473 int size = fs2h16(ep->e2d_reclen);
474
475 if (ep->e2d_ino != 0)
476 size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
477 if (size > 0) {
478 if (size >= slotneeded) {
479 slotstatus = FOUND;
480 slotoffset = results->ulr_offset;
481 slotsize = fs2h16(ep->e2d_reclen);
482 } else if (slotstatus == NONE) {
483 slotfreespace += size;
484 if (slotoffset == -1)
485 slotoffset = results->ulr_offset;
486 if (slotfreespace >= slotneeded) {
487 slotstatus = COMPACT;
488 slotsize = results->ulr_offset +
489 fs2h16(ep->e2d_reclen) -
490 slotoffset;
491 }
492 }
493 }
494 }
495
496 /*
497 * Check for a name match.
498 */
499 if (ep->e2d_ino) {
500 namlen = ep->e2d_namlen;
501 if (namlen == cnp->cn_namelen &&
502 !memcmp(cnp->cn_nameptr, ep->e2d_name,
503 (unsigned)namlen)) {
504 /*
505 * Save directory entry's inode number and
506 * reclen in ndp->ni_ufs area, and release
507 * directory buffer.
508 */
509 foundino = fs2h32(ep->e2d_ino);
510 results->ulr_reclen = fs2h16(ep->e2d_reclen);
511 goto found;
512 }
513 }
514 prevoff = results->ulr_offset;
515 results->ulr_offset += fs2h16(ep->e2d_reclen);
516 entryoffsetinblock += fs2h16(ep->e2d_reclen);
517 if (ep->e2d_ino)
518 enduseful = results->ulr_offset;
519 }
520notfound:
521 /*
522 * If we started in the middle of the directory and failed
523 * to find our target, we must check the beginning as well.
524 */
525 if (numdirpasses == 2) {
526 numdirpasses--;
527 results->ulr_offset = 0;
528 endsearch = results->ulr_diroff;
529 goto searchloop;
530 }
531 if (bp != NULL)
532 brelse(bp, 0);
533 /*
534 * If creating, and at end of pathname and current
535 * directory has not been removed, then can consider
536 * allowing file to be created.
537 */
538 if ((nameiop == CREATE || nameiop == RENAME) &&
539 (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
540 /*
541 * Access for write is interpreted as allowing
542 * creation of files in the directory.
543 */
544 error = VOP_ACCESS(vdp, VWRITE, cred);
545 if (error)
546 return error;
547 /*
548 * Return an indication of where the new directory
549 * entry should be put. If we didn't find a slot,
550 * then set results->ulr_count to 0 indicating
551 * that the new slot belongs at the end of the
552 * directory. If we found a slot, then the new entry
553 * can be put in the range from results->ulr_offset to
554 * results->ulr_offset + results->ulr_count.
555 */
556 if (slotstatus == NONE) {
557 results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
558 results->ulr_count = 0;
559 enduseful = results->ulr_offset;
560 } else {
561 results->ulr_offset = slotoffset;
562 results->ulr_count = slotsize;
563 if (enduseful < slotoffset + slotsize)
564 enduseful = slotoffset + slotsize;
565 }
566 results->ulr_endoff = roundup(enduseful, dirblksiz);
567#if 0
568 dp->i_flag |= IN_CHANGE | IN_UPDATE;
569#endif
570 /*
571 * We return with the directory locked, so that
572 * the parameters we set up above will still be
573 * valid if we actually decide to do a direnter().
574 * We return ni_vp == NULL to indicate that the entry
575 * does not currently exist; we leave a pointer to
576 * the (locked) directory inode in ndp->ni_dvp.
577 *
578 * NB - if the directory is unlocked, then this
579 * information cannot be used.
580 */
581 return EJUSTRETURN;
582 }
583 /*
584 * Insert name into cache (as non-existent) if appropriate.
585 */
586 if (nameiop != CREATE) {
587 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
588 cnp->cn_flags);
589 }
590 return ENOENT;
591
592found:
593 if (numdirpasses == 2)
594 namecache_count_pass2();
595 /*
596 * Check that directory length properly reflects presence
597 * of this entry.
598 */
599 if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
600 ufs_dirbad(dp, results->ulr_offset, "i_size too small");
601 error = ext2fs_setsize(dp,
602 results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
603 if (error) {
604 brelse(bp, 0);
605 return error;
606 }
607 dp->i_flag |= IN_CHANGE | IN_UPDATE;
608 uvm_vnp_setsize(vdp, ext2fs_size(dp));
609 }
610 brelse(bp, 0);
611
612 /*
613 * Found component in pathname.
614 * If the final component of path name, save information
615 * in the cache as to where the entry was found.
616 */
617 if ((flags & ISLASTCN) && nameiop == LOOKUP)
618 results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
619
620 /*
621 * If deleting, and at end of pathname, return
622 * parameters which can be used to remove file.
623 * Lock the inode, being careful with ".".
624 */
625 if (nameiop == DELETE && (flags & ISLASTCN)) {
626 /*
627 * Return pointer to current entry in results->ulr_offset,
628 * and distance past previous entry (if there
629 * is a previous entry in this block) in results->ulr_count.
630 * Save directory inode pointer in ndp->ni_dvp for dirremove().
631 */
632 if ((results->ulr_offset & (dirblksiz - 1)) == 0)
633 results->ulr_count = 0;
634 else
635 results->ulr_count = results->ulr_offset - prevoff;
636 if (dp->i_number == foundino) {
637 vref(vdp);
638 tdp = vdp;
639 } else {
640 error = vcache_get(vdp->v_mount,
641 &foundino, sizeof(foundino), &tdp);
642 if (error)
643 return error;
644 }
645 /*
646 * Write access to directory required to delete files.
647 */
648 if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
649 vrele(tdp);
650 return error;
651 }
652 /*
653 * If directory is "sticky", then user must own
654 * the directory, or the file in it, else she
655 * may not delete it (unless she's root). This
656 * implements append-only directories.
657 */
658 if (dp->i_e2fs_mode & ISVTX) {
659 error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
660 tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
661 VTOI(tdp)->i_uid));
662 if (error) {
663 vrele(tdp);
664 return EPERM;
665 }
666 }
667 *vpp = tdp;
668 return 0;
669 }
670
671 /*
672 * If rewriting (RENAME), return the inode and the
673 * information required to rewrite the present directory
674 * Must get inode of directory entry to verify it's a
675 * regular file, or empty directory.
676 */
677 if (nameiop == RENAME && (flags & ISLASTCN)) {
678 error = VOP_ACCESS(vdp, VWRITE, cred);
679 if (error)
680 return error;
681 /*
682 * Careful about locking second inode.
683 * This can only occur if the target is ".".
684 */
685 if (dp->i_number == foundino)
686 return EISDIR;
687 error = vcache_get(vdp->v_mount,
688 &foundino, sizeof(foundino), &tdp);
689 if (error)
690 return error;
691 *vpp = tdp;
692 return 0;
693 }
694
695 if (dp->i_number == foundino) {
696 vref(vdp); /* we want ourself, ie "." */
697 *vpp = vdp;
698 } else {
699 error = vcache_get(vdp->v_mount,
700 &foundino, sizeof(foundino), &tdp);
701 if (error)
702 return error;
703 *vpp = tdp;
704 }
705
706 /*
707 * Insert name into cache if appropriate.
708 */
709 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
710 return 0;
711}
712static void
713ext2fs_accumulatespace (struct ext2fs_searchslot *ssp, struct ext2fs_direct *ep,
714 doff_t *offp)
715{
716 int size = ep->e2d_reclen;
717
718 if (ep->e2d_ino != 0)
719 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
720
721 if (size <= 0)
722 return;
723
724 if (size >= ssp->slotneeded) {
725 ssp->slotstatus = FOUND;
726 ssp->slotoffset = *offp;
727 ssp->slotsize = ep->e2d_reclen;
728 return;
729 }
730
731 if (ssp->slotstatus != NONE)
732 return;
733
734 ssp->slotfreespace += size;
735 if (ssp->slotoffset == -1)
736 ssp->slotoffset = *offp;
737
738 if (ssp->slotfreespace >= ssp->slotneeded) {
739 ssp->slotstatus = COMPACT;
740 ssp->slotsize = *offp + ep->e2d_reclen - ssp->slotoffset;
741 }
742}
743
744int
745ext2fs_search_dirblock(struct inode *ip, void *data, int *foundp,
746 const char *name, int namelen, int *entryoffsetinblockp,
747 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
748 struct ext2fs_searchslot *ssp)
749{
750 struct vnode *vdp = ITOV(ip);
751 struct ext2fs_direct *ep, *top;
752 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
753 int offset = *entryoffsetinblockp;
754 int namlen;
755
756 ep = (void *)((char *)data + offset);
757 top = (void *)((char *)data + bsize - EXT2_DIR_REC_LEN(0));
758
759 while (ep < top) {
760 /*
761 * Full validation checks are slow, so we only check
762 * enough to insure forward progress through the
763 * directory. Complete checks can be run by setting
764 * "vfs.e2fs.dirchk" to be true.
765 */
766 if (ep->e2d_reclen == 0 ||
767 (dirchk && ext2fs_dirbadentry(vdp, ep, offset))) {
768 int i;
769 ufs_dirbad(ip, *offp, "mangled entry");
770 i = bsize - (offset & (bsize - 1));
771 *offp += i;
772 offset += i;
773 continue;
774 }
775
776 /*
777 * If an appropriate sized slot has not yet been found,
778 * check to see if one is available. Also accumulate space
779 * in the current block so that we can determine if
780 * compaction is viable.
781 */
782 if (ssp->slotstatus != FOUND)
783 ext2fs_accumulatespace(ssp, ep, offp);
784
785 /*
786 * Check for a name match.
787 */
788 if (ep->e2d_ino) {
789 namlen = ep->e2d_namlen;
790 if (namlen == namelen &&
791 !memcmp(name, ep->e2d_name, (unsigned)namlen)) {
792 /*
793 * Save directory entry's inode number and
794 * reclen in ndp->ni_ufs area, and release
795 * directory buffer.
796 */
797 *foundp = 1;
798 return 0;
799 }
800 }
801 *prevoffp = *offp;
802 *offp += ep->e2d_reclen;
803 offset += ep->e2d_reclen;
804 *entryoffsetinblockp = offset;
805 if (ep->e2d_ino)
806 *endusefulp = *offp;
807 /*
808 * Get pointer to the next entry.
809 */
810 ep = (void *)((char *)data + offset);
811 }
812
813 return 0;
814}
815
816/*
817 * Do consistency checking on a directory entry:
818 * record length must be multiple of 4
819 * entry must fit in rest of its dirblksize block
820 * record must be large enough to contain entry
821 * name is not longer than EXT2FS_MAXNAMLEN
822 * name must be as long as advertised, and null terminated
823 */
824/*
825 * changed so that it confirms to ext2fs_check_dir_entry
826 */
827static int
828ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
829 int entryoffsetinblock)
830{
831 struct ufsmount *ump = VFSTOUFS(dp->v_mount);
832 int dirblksiz = ump->um_dirblksiz;
833
834 const char *error_msg = NULL;
835 int reclen = fs2h16(de->e2d_reclen);
836 int namlen = de->e2d_namlen;
837
838 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
839 error_msg = "rec_len is smaller than minimal";
840 else if (reclen % 4 != 0)
841 error_msg = "rec_len % 4 != 0";
842 else if (namlen > EXT2FS_MAXNAMLEN)
843 error_msg = "namlen > EXT2FS_MAXNAMLEN";
844 else if (reclen < EXT2FS_DIRSIZ(namlen))
845 error_msg = "reclen is too small for name_len";
846 else if (entryoffsetinblock + reclen > dirblksiz)
847 error_msg = "directory entry across blocks";
848 else if (fs2h32(de->e2d_ino) >
849 VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
850 error_msg = "inode out of bounds";
851
852 if (error_msg != NULL) {
853 printf( "bad directory entry: %s\n"
854 "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
855 error_msg, entryoffsetinblock,
856 (unsigned long) fs2h32(de->e2d_ino),
857 reclen, namlen);
858 panic("ext2fs_dirbadentry");
859 }
860 return error_msg == NULL ? 0 : 1;
861}
862
863/*
864 * Write a directory entry after a call to namei, using the parameters
865 * that it left in nameidata. The argument ip is the inode which the new
866 * directory entry will refer to. Dvp is a pointer to the directory to
867 * be written, which was left locked by namei. Remaining parameters
868 * (ulr_offset, ulr_count) indicate how the space for the new
869 * entry is to be obtained.
870 */
871int
872ext2fs_direnter(struct inode *ip, struct vnode *dvp,
873 const struct ufs_lookup_results *ulr, struct componentname *cnp)
874{
875 struct inode *dp;
876 struct ext2fs_direct newdir;
877 struct iovec aiov;
878 struct uio auio;
879 int error;
880 struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
881 int dirblksiz = ump->um_dirblksiz;
882 size_t newentrysize;
883
884 dp = VTOI(dvp);
885
886 newdir.e2d_ino = h2fs32(ip->i_number);
887 newdir.e2d_namlen = cnp->cn_namelen;
888 if (EXT2F_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) {
889 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
890 } else {
891 newdir.e2d_type = 0;
892 }
893 memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
894 newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
895
896 if (ext2fs_htree_has_idx(dp)) {
897 error = ext2fs_htree_add_entry(dvp, &newdir, cnp, newentrysize);
898 if (error) {
899 dp->i_e2fs_flags &= ~EXT2_INDEX;
900 dp->i_flag |= IN_CHANGE | IN_UPDATE;
901 }
902 return error;
903 }
904
905 /*
906 * TODO check if Htree index is not created for the directory then
907 * create one if directory entries get overflew the first dir-block
908 */
909 if (ulr->ulr_count == 0) {
910 /*
911 * If ulr_count is 0, then namei could find no
912 * space in the directory. Here, ulr_offset will
913 * be on a directory block boundary and we will write the
914 * new entry into a fresh block.
915 */
916 if (ulr->ulr_offset & (dirblksiz - 1))
917 panic("ext2fs_direnter: newblk");
918 auio.uio_offset = ulr->ulr_offset;
919 newdir.e2d_reclen = h2fs16(dirblksiz);
920 auio.uio_resid = newentrysize;
921 aiov.iov_len = newentrysize;
922 aiov.iov_base = (void *)&newdir;
923 auio.uio_iov = &aiov;
924 auio.uio_iovcnt = 1;
925 auio.uio_rw = UIO_WRITE;
926 UIO_SETUP_SYSSPACE(&auio);
927 error = ext2fs_bufwr(dvp, &auio, IO_SYNC, cnp->cn_cred);
928 if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
929 /* XXX should grow with balloc() */
930 panic("ext2fs_direnter: frag size");
931 else if (!error) {
932 error = ext2fs_setsize(dp,
933 roundup(ext2fs_size(dp), dirblksiz));
934 if (error)
935 return error;
936 dp->i_flag |= IN_CHANGE;
937 uvm_vnp_setsize(dvp, ext2fs_size(dp));
938 }
939 return error;
940 }
941
942 error = ext2fs_add_entry(dvp, &newdir, ulr, newentrysize);
943
944 if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
945 error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
946 cnp->cn_cred);
947 return error;
948}
949
950/*
951 * Insert an entry into the directory block.
952 * Compact the contents.
953 */
954
955int
956ext2fs_add_entry(struct vnode* dvp, struct ext2fs_direct *entry,
957 const struct ufs_lookup_results *ulr, size_t newentrysize)
958{
959 struct ext2fs_direct *ep, *nep;
960 struct inode *dp;
961 struct buf *bp;
962 u_int dsize;
963 int error, loc, spacefree;
964 char *dirbuf;
965
966 dp = VTOI(dvp);
967
968 /*
969 * If ulr_count is non-zero, then namei found space
970 * for the new entry in the range ulr_offset to
971 * ulr_offset + ulr_count in the directory.
972 * To use this space, we may have to compact the entries located
973 * there, by copying them together towards the beginning of the
974 * block, leaving the free space in one usable chunk at the end.
975 */
976
977 /*
978 * Get the block containing the space for the new directory entry.
979 */
980 if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
981 return error;
982 /*
983 * Find space for the new entry. In the simple case, the entry at
984 * offset base will have the space. If it does not, then namei
985 * arranged that compacting the region ulr_offset to
986 * ulr_offset + ulr_count would yield the
987 * space.
988 */
989 ep = (struct ext2fs_direct *)dirbuf;
990 dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
991 spacefree = fs2h16(ep->e2d_reclen) - dsize;
992 for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count;) {
993 nep = (struct ext2fs_direct *)(dirbuf + loc);
994 if (ep->e2d_ino) {
995 /* trim the existing slot */
996 ep->e2d_reclen = h2fs16(dsize);
997 ep = (struct ext2fs_direct *)((char *)ep + dsize);
998 } else {
999 /* overwrite; nothing there; header is ours */
1000 spacefree += dsize;
1001 }
1002 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
1003 spacefree += fs2h16(nep->e2d_reclen) - dsize;
1004 loc += fs2h16(nep->e2d_reclen);
1005 memcpy((void *)ep, (void *)nep, dsize);
1006 }
1007 /*
1008 * Update the pointer fields in the previous entry (if any),
1009 * copy in the new entry, and write out the block.
1010 */
1011 if (ep->e2d_ino == 0) {
1012#ifdef DIAGNOSTIC
1013 if (spacefree + dsize < newentrysize)
1014 panic("ext2fs_direnter: compact1");
1015#endif
1016 entry->e2d_reclen = h2fs16(spacefree + dsize);
1017 } else {
1018#ifdef DIAGNOSTIC
1019 if (spacefree < newentrysize) {
1020 printf("ext2fs_direnter: compact2 %u %u",
1021 (u_int)spacefree, (u_int)newentrysize);
1022 panic("ext2fs_direnter: compact2");
1023 }
1024#endif
1025 entry->e2d_reclen = h2fs16(spacefree);
1026 ep->e2d_reclen = h2fs16(dsize);
1027 ep = (struct ext2fs_direct *)((char *)ep + dsize);
1028 }
1029 memcpy(ep, entry, (u_int)newentrysize);
1030 error = VOP_BWRITE(bp->b_vp, bp);
1031 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1032 return error;
1033}
1034
1035/*
1036 * Remove a directory entry after a call to namei, using
1037 * the auxiliary results it provided. The entry
1038 * ulr_offset contains the offset into the directory of the
1039 * entry to be eliminated. The ulr_count field contains the
1040 * size of the previous record in the directory. If this
1041 * is 0, the first entry is being deleted, so we need only
1042 * zero the inode number to mark the entry as free. If the
1043 * entry is not the first in the directory, we must reclaim
1044 * the space of the now empty record by adding the record size
1045 * to the size of the previous entry.
1046 */
1047int
1048ext2fs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
1049 struct componentname *cnp)
1050{
1051 struct inode *dp;
1052 struct ext2fs_direct *ep;
1053 struct buf *bp;
1054 int error;
1055
1056 dp = VTOI(dvp);
1057
1058 if (ulr->ulr_count == 0) {
1059 /*
1060 * First entry in block: set d_ino to zero.
1061 */
1062 error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset,
1063 (void *)&ep, &bp);
1064 if (error != 0)
1065 return error;
1066 ep->e2d_ino = 0;
1067 error = VOP_BWRITE(bp->b_vp, bp);
1068 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1069 return error;
1070 }
1071 /*
1072 * Collapse new free space into previous entry.
1073 */
1074 error = ext2fs_blkatoff(dvp, (off_t)(ulr->ulr_offset - ulr->ulr_count),
1075 (void *)&ep, &bp);
1076 if (error != 0)
1077 return error;
1078 ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + ulr->ulr_reclen);
1079 error = VOP_BWRITE(bp->b_vp, bp);
1080 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1081 return error;
1082}
1083
1084/*
1085 * Rewrite an existing directory entry to point at the inode
1086 * supplied. The parameters describing the directory entry are
1087 * set up by a call to namei.
1088 */
1089int
1090ext2fs_dirrewrite(struct inode *dp, const struct ufs_lookup_results *ulr,
1091 struct inode *ip, struct componentname *cnp)
1092{
1093 struct buf *bp;
1094 struct ext2fs_direct *ep;
1095 struct vnode *vdp = ITOV(dp);
1096 int error;
1097
1098 error = ext2fs_blkatoff(vdp, (off_t)ulr->ulr_offset, (void *)&ep, &bp);
1099 if (error != 0)
1100 return error;
1101 ep->e2d_ino = h2fs32(ip->i_number);
1102 if (EXT2F_HAS_INCOMPAT_FEATURE(dp->i_e2fs, EXT2F_INCOMPAT_FTYPE)) {
1103 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
1104 } else {
1105 ep->e2d_type = 0;
1106 }
1107 error = VOP_BWRITE(bp->b_vp, bp);
1108 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1109 return error;
1110}
1111
1112/*
1113 * Check if a directory is empty or not.
1114 * Inode supplied must be locked.
1115 *
1116 * Using a struct dirtemplate here is not precisely
1117 * what we want, but better than using a struct ext2fs_direct.
1118 *
1119 * NB: does not handle corrupted directories.
1120 */
1121int
1122ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
1123{
1124 off_t off;
1125 struct ext2fs_dirtemplate dbuf;
1126 struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
1127 int error, namlen;
1128 size_t count;
1129
1130#define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
1131
1132 for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
1133 error = ufs_bufio(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ,
1134 off, IO_NODELOCKED, cred, &count, NULL);
1135 /*
1136 * Since we read MINDIRSIZ, residual must
1137 * be 0 unless we're at end of file.
1138 */
1139 if (error || count != 0)
1140 return 0;
1141 /* avoid infinite loops */
1142 if (dp->e2d_reclen == 0)
1143 return 0;
1144 /* skip empty entries */
1145 if (dp->e2d_ino == 0)
1146 continue;
1147 /* accept only "." and ".." */
1148 namlen = dp->e2d_namlen;
1149 if (namlen > 2)
1150 return 0;
1151 if (dp->e2d_name[0] != '.')
1152 return 0;
1153 /*
1154 * At this point namlen must be 1 or 2.
1155 * 1 implies ".", 2 implies ".." if second
1156 * char is also "."
1157 */
1158 if (namlen == 1)
1159 continue;
1160 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1161 continue;
1162 return 0;
1163 }
1164 return 1;
1165}
1166