1/*
2 * modified for Lites 1.1
3 *
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
6 *
7 * $FreeBSD: head/sys/gnu/fs/ext2fs/ext2_lookup.c 59259 2000-04-15 17:14:22Z rwatson $
8 */
9/*
10 * Copyright (c) 1989, 1993
11 * The Regents of the University of California. All rights reserved.
12 * (c) UNIX System Laboratories, Inc.
13 * All or some portions of this file are derived from material licensed
14 * to the University of California by American Telephone and Telegraph
15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16 * the permission of UNIX System Laboratories, Inc.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
20 * are met:
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
26 * 3. All advertising materials mentioning features or use of this software
27 * must display the following acknowledgement:
28 * This product includes software developed by the University of
29 * California, Berkeley and its contributors.
30 * 4. Neither the name of the University nor the names of its contributors
31 * may be used to endorse or promote products derived from this software
32 * without specific prior written permission.
33 *
34 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
35 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
38 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44 * SUCH DAMAGE.
45 *
46 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
47 */
48
49#include <stddef.h>
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/namei.h>
54#include <sys/buf.h>
55#include <sys/mount.h>
56#include <sys/vnode.h>
57#include <sys/malloc.h>
58#include <sys/dirent.h>
59
60#include <ufs/ufs/extattr.h>
61#include <ufs/ufs/quota.h>
62#include <ufs/ufs/inode.h>
63#include <ufs/ufs/dir.h>
64#include <ufs/ufs/ufsmount.h>
65#include <ufs/ufs/ufs_extern.h>
66
67#include <gnu/ext2fs/ext2_extern.h>
68#include <gnu/ext2fs/ext2_fs.h>
69#include <gnu/ext2fs/ext2_fs_sb.h>
70
71/*
72 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
73 while it is the native blocksize in ext2fs - thus, a #define
74 is no longer appropriate
75*/
76#undef DIRBLKSIZ
77
78extern int dirchk;
79
80static u_char ext2_ft_to_dt[] = {
81 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
82 DT_REG, /* EXT2_FT_REG_FILE */
83 DT_DIR, /* EXT2_FT_DIR */
84 DT_CHR, /* EXT2_FT_CHRDEV */
85 DT_BLK, /* EXT2_FT_BLKDEV */
86 DT_FIFO, /* EXT2_FT_FIFO */
87 DT_SOCK, /* EXT2_FT_SOCK */
88 DT_LNK, /* EXT2_FT_SYMLINK */
89};
90#define FTTODT(ft) \
91 ((ft) > sizeof(ext2_ft_to_dt) / sizeof(ext2_ft_to_dt[0]) ? \
92 DT_UNKNOWN : ext2_ft_to_dt[(ft)])
93
94static u_char dt_to_ext2_ft[] = {
95 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
96 EXT2_FT_FIFO, /* DT_FIFO */
97 EXT2_FT_CHRDEV, /* DT_CHR */
98 EXT2_FT_UNKNOWN, /* unused */
99 EXT2_FT_DIR, /* DT_DIR */
100 EXT2_FT_UNKNOWN, /* unused */
101 EXT2_FT_BLKDEV, /* DT_BLK */
102 EXT2_FT_UNKNOWN, /* unused */
103 EXT2_FT_REG_FILE, /* DT_REG */
104 EXT2_FT_UNKNOWN, /* unused */
105 EXT2_FT_SYMLINK, /* DT_LNK */
106 EXT2_FT_UNKNOWN, /* unused */
107 EXT2_FT_SOCK, /* DT_SOCK */
108 EXT2_FT_UNKNOWN, /* unused */
109 EXT2_FT_UNKNOWN, /* DT_WHT */
110};
111#define DTTOFT(dt) \
112 ((dt) > sizeof(dt_to_ext2_ft) / sizeof(dt_to_ext2_ft[0]) ? \
113 EXT2_FT_UNKNOWN : dt_to_ext2_ft[(dt)])
114
115static int ext2_dirbadentry __P((struct vnode *dp,
116 struct ext2_dir_entry_2 *de,
117 int entryoffsetinblock));
118
119/*
120 * Vnode op for reading directories.
121 *
122 * The routine below assumes that the on-disk format of a directory
123 * is the same as that defined by <sys/dirent.h>. If the on-disk
124 * format changes, then it will be necessary to do a conversion
125 * from the on-disk format that read returns to the format defined
126 * by <sys/dirent.h>.
127 */
128/*
129 * this is exactly what we do here - the problem is that the conversion
130 * will blow up some entries by four bytes, so it can't be done in place.
131 * This is too bad. Right now the conversion is done entry by entry, the
132 * converted entry is sent via uiomove.
133 *
134 * XXX allocate a buffer, convert as many entries as possible, then send
135 * the whole buffer to uiomove
136 */
137int
138ext2_readdir(ap)
139 struct vop_readdir_args /* {
140 struct vnode *a_vp;
141 struct uio *a_uio;
142 struct ucred *a_cred;
143 } */ *ap;
144{
145 register struct uio *uio = ap->a_uio;
146 int count, error;
147
148 struct ext2_dir_entry_2 *edp, *dp;
149 int ncookies;
150 struct dirent dstdp;
151 struct uio auio;
152 struct iovec aiov;
153 caddr_t dirbuf;
154 int readcnt;
155 u_quad_t startoffset = uio->uio_offset;
156
157 count = uio->uio_resid; /* legyenek boldogok akik akarnak ... */
158 uio->uio_resid = count;
159 uio->uio_iov->iov_len = count;
160
161#if 0
162printf("ext2_readdir called uio->uio_offset %d uio->uio_resid %d count %d \n",
163 (int)uio->uio_offset, (int)uio->uio_resid, (int)count);
164#endif
165
166 auio = *uio;
167 auio.uio_iov = &aiov;
168 auio.uio_iovcnt = 1;
169 auio.uio_segflg = UIO_SYSSPACE;
170 aiov.iov_len = count;
171 MALLOC(dirbuf, caddr_t, count, M_TEMP, M_WAITOK);
172 aiov.iov_base = dirbuf;
173 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
174 if (error == 0) {
175 readcnt = count - auio.uio_resid;
176 edp = (struct ext2_dir_entry_2 *)&dirbuf[readcnt];
177 ncookies = 0;
178 bzero(&dstdp, offsetof(struct dirent, d_name));
179 for (dp = (struct ext2_dir_entry_2 *)dirbuf;
180 !error && uio->uio_resid > 0 && dp < edp; ) {
181 /*-
182 * "New" ext2fs directory entries differ in 3 ways
183 * from ufs on-disk ones:
184 * - the name is not necessarily NUL-terminated.
185 * - the file type field always exists and always
186 * follows the name length field.
187 * - the file type is encoded in a different way.
188 *
189 * "Old" ext2fs directory entries need no special
190 * conversions, since they binary compatible with
191 * "new" entries having a file type of 0 (i.e.,
192 * EXT2_FT_UNKNOWN). Splitting the old name length
193 * field didn't make a mess like it did in ufs,
194 * because ext2fs uses a machine-dependent disk
195 * layout.
196 */
197 dstdp.d_fileno = dp->inode;
198 dstdp.d_type = FTTODT(dp->file_type);
199 dstdp.d_namlen = dp->name_len;
200 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
201 bcopy(dp->name, dstdp.d_name, dstdp.d_namlen);
202 bzero(dstdp.d_name + dstdp.d_namlen,
203 dstdp.d_reclen - offsetof(struct dirent, d_name) -
204 dstdp.d_namlen);
205
206 if (dp->rec_len > 0) {
207 if(dstdp.d_reclen <= uio->uio_resid) {
208 /* advance dp */
209 dp = (struct ext2_dir_entry_2 *)
210 ((char *)dp + dp->rec_len);
211 error =
212 uiomove((caddr_t)&dstdp,
213 dstdp.d_reclen, uio);
214 if (!error)
215 ncookies++;
216 } else
217 break;
218 } else {
219 error = EIO;
220 break;
221 }
222 }
223 /* we need to correct uio_offset */
224 uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
225
226 if (!error && ap->a_ncookies != NULL) {
227 u_long *cookies;
228 u_long *cookiep;
229 off_t off;
230
231 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
232 panic("ext2fs_readdir: unexpected uio from NFS server");
233 MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
234 M_WAITOK);
235 off = startoffset;
236 for (dp = (struct ext2_dir_entry_2 *)dirbuf, cookiep = cookies;
237 dp < edp;
238 dp = (struct ext2_dir_entry_2 *)((caddr_t) dp + dp->rec_len)) {
239 off += dp->rec_len;
240 *cookiep++ = (u_long) off;
241 }
242 *ap->a_ncookies = ncookies;
243 *ap->a_cookies = cookies;
244 }
245 }
246 FREE(dirbuf, M_TEMP);
247 if (ap->a_eofflag)
248 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
249 return (error);
250}
251
252/*
253 * Convert a component of a pathname into a pointer to a locked inode.
254 * This is a very central and rather complicated routine.
255 * If the file system is not maintained in a strict tree hierarchy,
256 * this can result in a deadlock situation (see comments in code below).
257 *
258 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
259 * on whether the name is to be looked up, created, renamed, or deleted.
260 * When CREATE, RENAME, or DELETE is specified, information usable in
261 * creating, renaming, or deleting a directory entry may be calculated.
262 * If flag has LOCKPARENT or'ed into it and the target of the pathname
263 * exists, lookup returns both the target and its parent directory locked.
264 * When creating or renaming and LOCKPARENT is specified, the target may
265 * not be ".". When deleting and LOCKPARENT is specified, the target may
266 * be "."., but the caller must check to ensure it does an vrele and vput
267 * instead of two vputs.
268 *
269 * Overall outline of ufs_lookup:
270 *
271 * search for name in directory, to found or notfound
272 * notfound:
273 * if creating, return locked directory, leaving info on available slots
274 * else return error
275 * found:
276 * if at end of path and deleting, return information to allow delete
277 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
278 * inode and return info to allow rewrite
279 * if not at end, add name to cache; if at end and neither creating
280 * nor deleting, add name to cache
281 */
282int
283ext2_lookup(ap)
284 struct vop_cachedlookup_args /* {
285 struct vnode *a_dvp;
286 struct vnode **a_vpp;
287 struct componentname *a_cnp;
288 } */ *ap;
289{
290 register struct vnode *vdp; /* vnode for directory being searched */
291 register struct inode *dp; /* inode for directory being searched */
292 struct buf *bp; /* a buffer of directory entries */
293 register struct ext2_dir_entry_2 *ep; /* the current directory entry */
294 int entryoffsetinblock; /* offset of ep in bp's buffer */
295 enum {NONE, COMPACT, FOUND} slotstatus;
296 doff_t slotoffset; /* offset of area with free space */
297 int slotsize; /* size of area at slotoffset */
298 int slotfreespace; /* amount of space free in slot */
299 int slotneeded; /* size of the entry we're seeking */
300 int numdirpasses; /* strategy for directory search */
301 doff_t endsearch; /* offset to end directory search */
302 doff_t prevoff; /* prev entry dp->i_offset */
303 struct vnode *pdp; /* saved dp during symlink work */
304 struct vnode *tdp; /* returned by VFS_VGET */
305 doff_t enduseful; /* pointer past last used dir slot */
306 u_long bmask; /* block offset mask */
307 int lockparent; /* 1 => lockparent flag is set */
308 int wantparent; /* 1 => wantparent or lockparent flag */
309 int namlen, error;
310 struct vnode **vpp = ap->a_vpp;
311 struct componentname *cnp = ap->a_cnp;
312 struct ucred *cred = cnp->cn_cred;
313 int flags = cnp->cn_flags;
314 int nameiop = cnp->cn_nameiop;
315 struct proc *p = cnp->cn_proc;
316
317 int DIRBLKSIZ = VTOI(ap->a_dvp)->i_e2fs->s_blocksize;
318
319 bp = NULL;
320 slotoffset = -1;
321 *vpp = NULL;
322 vdp = ap->a_dvp;
323 dp = VTOI(vdp);
324 lockparent = flags & LOCKPARENT;
325 wantparent = flags & (LOCKPARENT|WANTPARENT);
326
327 /*
328 * We now have a segment name to search for, and a directory to search.
329 */
330
331 /*
332 * Suppress search for slots unless creating
333 * file and at end of pathname, in which case
334 * we watch for a place to put the new file in
335 * case it doesn't already exist.
336 */
337 slotstatus = FOUND;
338 slotfreespace = slotsize = slotneeded = 0;
339 if ((nameiop == CREATE || nameiop == RENAME) &&
340 (flags & ISLASTCN)) {
341 slotstatus = NONE;
342 slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
343 /* was
344 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
345 cnp->cn_namelen + 3) &~ 3; */
346 }
347
348 /*
349 * If there is cached information on a previous search of
350 * this directory, pick up where we last left off.
351 * We cache only lookups as these are the most common
352 * and have the greatest payoff. Caching CREATE has little
353 * benefit as it usually must search the entire directory
354 * to determine that the entry does not exist. Caching the
355 * location of the last DELETE or RENAME has not reduced
356 * profiling time and hence has been removed in the interest
357 * of simplicity.
358 */
359 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
360 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
361 dp->i_diroff > dp->i_size) {
362 entryoffsetinblock = 0;
363 dp->i_offset = 0;
364 numdirpasses = 1;
365 } else {
366 dp->i_offset = dp->i_diroff;
367 if ((entryoffsetinblock = dp->i_offset & bmask) &&
368 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
369 return (error);
370 numdirpasses = 2;
371 nchstats.ncs_2passes++;
372 }
373 prevoff = dp->i_offset;
374 endsearch = roundup(dp->i_size, DIRBLKSIZ);
375 enduseful = 0;
376
377searchloop:
378 while (dp->i_offset < endsearch) {
379 /*
380 * If necessary, get the next directory block.
381 */
382 if ((dp->i_offset & bmask) == 0) {
383 if (bp != NULL)
384 brelse(bp);
385 if ((error =
386 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)) != 0)
387 return (error);
388 entryoffsetinblock = 0;
389 }
390 /*
391 * If still looking for a slot, and at a DIRBLKSIZE
392 * boundary, have to start looking for free space again.
393 */
394 if (slotstatus == NONE &&
395 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
396 slotoffset = -1;
397 slotfreespace = 0;
398 }
399 /*
400 * Get pointer to next entry.
401 * Full validation checks are slow, so we only check
402 * enough to insure forward progress through the
403 * directory. Complete checks can be run by patching
404 * "dirchk" to be true.
405 */
406 ep = (struct ext2_dir_entry_2 *)
407 ((char *)bp->b_data + entryoffsetinblock);
408 if (ep->rec_len == 0 ||
409 (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
410 int i;
411 ufs_dirbad(dp, dp->i_offset, "mangled entry");
412 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
413 dp->i_offset += i;
414 entryoffsetinblock += i;
415 continue;
416 }
417
418 /*
419 * If an appropriate sized slot has not yet been found,
420 * check to see if one is available. Also accumulate space
421 * in the current block so that we can determine if
422 * compaction is viable.
423 */
424 if (slotstatus != FOUND) {
425 int size = ep->rec_len;
426
427 if (ep->inode != 0)
428 size -= EXT2_DIR_REC_LEN(ep->name_len);
429 if (size > 0) {
430 if (size >= slotneeded) {
431 slotstatus = FOUND;
432 slotoffset = dp->i_offset;
433 slotsize = ep->rec_len;
434 } else if (slotstatus == NONE) {
435 slotfreespace += size;
436 if (slotoffset == -1)
437 slotoffset = dp->i_offset;
438 if (slotfreespace >= slotneeded) {
439 slotstatus = COMPACT;
440 slotsize = dp->i_offset +
441 ep->rec_len - slotoffset;
442 }
443 }
444 }
445 }
446
447 /*
448 * Check for a name match.
449 */
450 if (ep->inode) {
451 namlen = ep->name_len;
452 if (namlen == cnp->cn_namelen &&
453 !bcmp(cnp->cn_nameptr, ep->name,
454 (unsigned)namlen)) {
455 /*
456 * Save directory entry's inode number and
457 * reclen in ndp->ni_ufs area, and release
458 * directory buffer.
459 */
460 dp->i_ino = ep->inode;
461 dp->i_reclen = ep->rec_len;
462 brelse(bp);
463 goto found;
464 }
465 }
466 prevoff = dp->i_offset;
467 dp->i_offset += ep->rec_len;
468 entryoffsetinblock += ep->rec_len;
469 if (ep->inode)
470 enduseful = dp->i_offset;
471 }
472/* notfound: */
473 /*
474 * If we started in the middle of the directory and failed
475 * to find our target, we must check the beginning as well.
476 */
477 if (numdirpasses == 2) {
478 numdirpasses--;
479 dp->i_offset = 0;
480 endsearch = dp->i_diroff;
481 goto searchloop;
482 }
483 if (bp != NULL)
484 brelse(bp);
485 /*
486 * If creating, and at end of pathname and current
487 * directory has not been removed, then can consider
488 * allowing file to be created.
489 */
490 if ((nameiop == CREATE || nameiop == RENAME) &&
491 (flags & ISLASTCN) && dp->i_nlink != 0) {
492 /*
493 * Access for write is interpreted as allowing
494 * creation of files in the directory.
495 */
496 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
497 return (error);
498 /*
499 * Return an indication of where the new directory
500 * entry should be put. If we didn't find a slot,
501 * then set dp->i_count to 0 indicating
502 * that the new slot belongs at the end of the
503 * directory. If we found a slot, then the new entry
504 * can be put in the range from dp->i_offset to
505 * dp->i_offset + dp->i_count.
506 */
507 if (slotstatus == NONE) {
508 dp->i_offset = roundup(dp->i_size, DIRBLKSIZ);
509 dp->i_count = 0;
510 enduseful = dp->i_offset;
511 } else {
512 dp->i_offset = slotoffset;
513 dp->i_count = slotsize;
514 if (enduseful < slotoffset + slotsize)
515 enduseful = slotoffset + slotsize;
516 }
517 dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
518 dp->i_flag |= IN_CHANGE | IN_UPDATE;
519 /*
520 * We return with the directory locked, so that
521 * the parameters we set up above will still be
522 * valid if we actually decide to do a direnter().
523 * We return ni_vp == NULL to indicate that the entry
524 * does not currently exist; we leave a pointer to
525 * the (locked) directory inode in ndp->ni_dvp.
526 * The pathname buffer is saved so that the name
527 * can be obtained later.
528 *
529 * NB - if the directory is unlocked, then this
530 * information cannot be used.
531 */
532 cnp->cn_flags |= SAVENAME;
533 if (!lockparent)
534 VOP_UNLOCK(vdp, 0, p);
535 return (EJUSTRETURN);
536 }
537 /*
538 * Insert name into cache (as non-existent) if appropriate.
539 */
540 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
541 cache_enter(vdp, *vpp, cnp);
542 return (ENOENT);
543
544found:
545 if (numdirpasses == 2)
546 nchstats.ncs_pass2++;
547 /*
548 * Check that directory length properly reflects presence
549 * of this entry.
550 */
551 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->name_len)
552 > dp->i_size) {
553 ufs_dirbad(dp, dp->i_offset, "i_size too small");
554 dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->name_len);
555 dp->i_flag |= IN_CHANGE | IN_UPDATE;
556 }
557
558 /*
559 * Found component in pathname.
560 * If the final component of path name, save information
561 * in the cache as to where the entry was found.
562 */
563 if ((flags & ISLASTCN) && nameiop == LOOKUP)
564 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
565
566 /*
567 * If deleting, and at end of pathname, return
568 * parameters which can be used to remove file.
569 * If the wantparent flag isn't set, we return only
570 * the directory (in ndp->ni_dvp), otherwise we go
571 * on and lock the inode, being careful with ".".
572 */
573 if (nameiop == DELETE && (flags & ISLASTCN)) {
574 /*
575 * Write access to directory required to delete files.
576 */
577 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
578 return (error);
579 /*
580 * Return pointer to current entry in dp->i_offset,
581 * and distance past previous entry (if there
582 * is a previous entry in this block) in dp->i_count.
583 * Save directory inode pointer in ndp->ni_dvp for dirremove().
584 */
585 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
586 dp->i_count = 0;
587 else
588 dp->i_count = dp->i_offset - prevoff;
589 if (dp->i_number == dp->i_ino) {
590 VREF(vdp);
591 *vpp = vdp;
592 return (0);
593 }
594 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
595 return (error);
596 /*
597 * If directory is "sticky", then user must own
598 * the directory, or the file in it, else she
599 * may not delete it (unless she's root). This
600 * implements append-only directories.
601 */
602 if ((dp->i_mode & ISVTX) &&
603 cred->cr_uid != 0 &&
604 cred->cr_uid != dp->i_uid &&
605 VTOI(tdp)->i_uid != cred->cr_uid) {
606 vput(tdp);
607 return (EPERM);
608 }
609 *vpp = tdp;
610 if (!lockparent)
611 VOP_UNLOCK(vdp, 0, p);
612 return (0);
613 }
614
615 /*
616 * If rewriting (RENAME), return the inode and the
617 * information required to rewrite the present directory
618 * Must get inode of directory entry to verify it's a
619 * regular file, or empty directory.
620 */
621 if (nameiop == RENAME && wantparent &&
622 (flags & ISLASTCN)) {
623 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
624 return (error);
625 /*
626 * Careful about locking second inode.
627 * This can only occur if the target is ".".
628 */
629 if (dp->i_number == dp->i_ino)
630 return (EISDIR);
631 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
632 return (error);
633 *vpp = tdp;
634 cnp->cn_flags |= SAVENAME;
635 if (!lockparent)
636 VOP_UNLOCK(vdp, 0, p);
637 return (0);
638 }
639
640 /*
641 * Step through the translation in the name. We do not `vput' the
642 * directory because we may need it again if a symbolic link
643 * is relative to the current directory. Instead we save it
644 * unlocked as "pdp". We must get the target inode before unlocking
645 * the directory to insure that the inode will not be removed
646 * before we get it. We prevent deadlock by always fetching
647 * inodes from the root, moving down the directory tree. Thus
648 * when following backward pointers ".." we must unlock the
649 * parent directory before getting the requested directory.
650 * There is a potential race condition here if both the current
651 * and parent directories are removed before the VFS_VGET for the
652 * inode associated with ".." returns. We hope that this occurs
653 * infrequently since we cannot avoid this race condition without
654 * implementing a sophisticated deadlock detection algorithm.
655 * Note also that this simple deadlock detection scheme will not
656 * work if the file system has any hard links other than ".."
657 * that point backwards in the directory structure.
658 */
659 pdp = vdp;
660 if (flags & ISDOTDOT) {
661 VOP_UNLOCK(pdp, 0, p); /* race to get the inode */
662 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
663 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p);
664 return (error);
665 }
666 if (lockparent && (flags & ISLASTCN) &&
667 (error = vn_lock(pdp, LK_EXCLUSIVE, p))) {
668 vput(tdp);
669 return (error);
670 }
671 *vpp = tdp;
672 } else if (dp->i_number == dp->i_ino) {
673 VREF(vdp); /* we want ourself, ie "." */
674 *vpp = vdp;
675 } else {
676 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
677 return (error);
678 if (!lockparent || !(flags & ISLASTCN))
679 VOP_UNLOCK(pdp, 0, p);
680 *vpp = tdp;
681 }
682
683 /*
684 * Insert name into cache if appropriate.
685 */
686 if (cnp->cn_flags & MAKEENTRY)
687 cache_enter(vdp, *vpp, cnp);
688 return (0);
689}
690
691/*
692 * Do consistency checking on a directory entry:
693 * record length must be multiple of 4
694 * entry must fit in rest of its DIRBLKSIZ block
695 * record must be large enough to contain entry
696 * name is not longer than MAXNAMLEN
697 * name must be as long as advertised, and null terminated
698 */
699/*
700 * changed so that it confirms to ext2_check_dir_entry
701 */
702static int
703ext2_dirbadentry(dp, de, entryoffsetinblock)
704 struct vnode *dp;
705 register struct ext2_dir_entry_2 *de;
706 int entryoffsetinblock;
707{
708 int DIRBLKSIZ = VTOI(dp)->i_e2fs->s_blocksize;
709
710 char * error_msg = NULL;
711
712 if (de->rec_len < EXT2_DIR_REC_LEN(1))
713 error_msg = "rec_len is smaller than minimal";
714 else if (de->rec_len % 4 != 0)
715 error_msg = "rec_len % 4 != 0";
716 else if (de->rec_len < EXT2_DIR_REC_LEN(de->name_len))
717 error_msg = "reclen is too small for name_len";
718 else if (entryoffsetinblock + de->rec_len > DIRBLKSIZ)
719 error_msg = "directory entry across blocks";
720 /* else LATER
721 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
722 error_msg = "inode out of bounds";
723 */
724
725 if (error_msg != NULL) {
726 printf("bad directory entry: %s\n", error_msg);
727 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
728 entryoffsetinblock, (unsigned long)de->inode,
729 de->rec_len, de->name_len);
730 }
731 return error_msg == NULL ? 0 : 1;
732}
733
734/*
735 * Write a directory entry after a call to namei, using the parameters
736 * that it left in nameidata. The argument ip is the inode which the new
737 * directory entry will refer to. Dvp is a pointer to the directory to
738 * be written, which was left locked by namei. Remaining parameters
739 * (dp->i_offset, dp->i_count) indicate how the space for the new
740 * entry is to be obtained.
741 */
742int
743ext2_direnter(ip, dvp, cnp)
744 struct inode *ip;
745 struct vnode *dvp;
746 register struct componentname *cnp;
747{
748 register struct ext2_dir_entry_2 *ep, *nep;
749 register struct inode *dp;
750 struct buf *bp;
751 struct ext2_dir_entry_2 newdir;
752 struct iovec aiov;
753 struct uio auio;
754 u_int dsize;
755 int error, loc, newentrysize, spacefree;
756 char *dirbuf;
757 int DIRBLKSIZ = ip->i_e2fs->s_blocksize;
758
759
760#if DIAGNOSTIC
761 if ((cnp->cn_flags & SAVENAME) == 0)
762 panic("direnter: missing name");
763#endif
764 dp = VTOI(dvp);
765 newdir.inode = ip->i_number;
766 newdir.name_len = cnp->cn_namelen;
767 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
768 EXT2_FEATURE_INCOMPAT_FILETYPE))
769 newdir.file_type = DTTOFT(IFTODT(ip->i_mode));
770 else
771 newdir.file_type = EXT2_FT_UNKNOWN;
772 bcopy(cnp->cn_nameptr, newdir.name, (unsigned)cnp->cn_namelen + 1);
773 newentrysize = EXT2_DIR_REC_LEN(newdir.name_len);
774 if (dp->i_count == 0) {
775 /*
776 * If dp->i_count is 0, then namei could find no
777 * space in the directory. Here, dp->i_offset will
778 * be on a directory block boundary and we will write the
779 * new entry into a fresh block.
780 */
781 if (dp->i_offset & (DIRBLKSIZ - 1))
782 panic("ext2_direnter: newblk");
783 auio.uio_offset = dp->i_offset;
784 newdir.rec_len = DIRBLKSIZ;
785 auio.uio_resid = newentrysize;
786 aiov.iov_len = newentrysize;
787 aiov.iov_base = (caddr_t)&newdir;
788 auio.uio_iov = &aiov;
789 auio.uio_iovcnt = 1;
790 auio.uio_rw = UIO_WRITE;
791 auio.uio_segflg = UIO_SYSSPACE;
792 auio.uio_procp = (struct proc *)0;
793 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
794 if (DIRBLKSIZ >
795 VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
796 /* XXX should grow with balloc() */
797 panic("ext2_direnter: frag size");
798 else if (!error) {
799 dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
800 dp->i_flag |= IN_CHANGE;
801 }
802 return (error);
803 }
804
805 /*
806 * If dp->i_count is non-zero, then namei found space
807 * for the new entry in the range dp->i_offset to
808 * dp->i_offset + dp->i_count in the directory.
809 * To use this space, we may have to compact the entries located
810 * there, by copying them together towards the beginning of the
811 * block, leaving the free space in one usable chunk at the end.
812 */
813
814 /*
815 * Increase size of directory if entry eats into new space.
816 * This should never push the size past a new multiple of
817 * DIRBLKSIZE.
818 *
819 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
820 */
821 if (dp->i_offset + dp->i_count > dp->i_size)
822 dp->i_size = dp->i_offset + dp->i_count;
823 /*
824 * Get the block containing the space for the new directory entry.
825 */
826 if ((error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0)
827 return (error);
828 /*
829 * Find space for the new entry. In the simple case, the entry at
830 * offset base will have the space. If it does not, then namei
831 * arranged that compacting the region dp->i_offset to
832 * dp->i_offset + dp->i_count would yield the
833 * space.
834 */
835 ep = (struct ext2_dir_entry_2 *)dirbuf;
836 dsize = EXT2_DIR_REC_LEN(ep->name_len);
837 spacefree = ep->rec_len - dsize;
838 for (loc = ep->rec_len; loc < dp->i_count; ) {
839 nep = (struct ext2_dir_entry_2 *)(dirbuf + loc);
840 if (ep->inode) {
841 /* trim the existing slot */
842 ep->rec_len = dsize;
843 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
844 } else {
845 /* overwrite; nothing there; header is ours */
846 spacefree += dsize;
847 }
848 dsize = EXT2_DIR_REC_LEN(nep->name_len);
849 spacefree += nep->rec_len - dsize;
850 loc += nep->rec_len;
851 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
852 }
853 /*
854 * Update the pointer fields in the previous entry (if any),
855 * copy in the new entry, and write out the block.
856 */
857 if (ep->inode == 0) {
858 if (spacefree + dsize < newentrysize)
859 panic("ext2_direnter: compact1");
860 newdir.rec_len = spacefree + dsize;
861 } else {
862 if (spacefree < newentrysize)
863 panic("ext2_direnter: compact2");
864 newdir.rec_len = spacefree;
865 ep->rec_len = dsize;
866 ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
867 }
868 bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
869 error = BUF_WRITE(bp);
870 dp->i_flag |= IN_CHANGE | IN_UPDATE;
871 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
872 error = UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
873 cnp->cn_cred, cnp->cn_proc);
874 return (error);
875}
876
877/*
878 * Remove a directory entry after a call to namei, using
879 * the parameters which it left in nameidata. The entry
880 * dp->i_offset contains the offset into the directory of the
881 * entry to be eliminated. The dp->i_count field contains the
882 * size of the previous record in the directory. If this
883 * is 0, the first entry is being deleted, so we need only
884 * zero the inode number to mark the entry as free. If the
885 * entry is not the first in the directory, we must reclaim
886 * the space of the now empty record by adding the record size
887 * to the size of the previous entry.
888 */
889int
890ext2_dirremove(dvp, cnp)
891 struct vnode *dvp;
892 struct componentname *cnp;
893{
894 register struct inode *dp;
895 struct ext2_dir_entry_2 *ep;
896 struct buf *bp;
897 int error;
898
899 dp = VTOI(dvp);
900 if (dp->i_count == 0) {
901 /*
902 * First entry in block: set d_ino to zero.
903 */
904 if ((error =
905 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
906 return (error);
907 ep->inode = 0;
908 error = BUF_WRITE(bp);
909 dp->i_flag |= IN_CHANGE | IN_UPDATE;
910 return (error);
911 }
912 /*
913 * Collapse new free space into previous entry.
914 */
915 if ((error = UFS_BLKATOFF(dvp, (off_t)(dp->i_offset - dp->i_count),
916 (char **)&ep, &bp)) != 0)
917 return (error);
918 ep->rec_len += dp->i_reclen;
919 error = BUF_WRITE(bp);
920 dp->i_flag |= IN_CHANGE | IN_UPDATE;
921 return (error);
922}
923
924/*
925 * Rewrite an existing directory entry to point at the inode
926 * supplied. The parameters describing the directory entry are
927 * set up by a call to namei.
928 */
929int
930ext2_dirrewrite(dp, ip, cnp)
931 struct inode *dp, *ip;
932 struct componentname *cnp;
933{
934 struct buf *bp;
935 struct ext2_dir_entry_2 *ep;
936 struct vnode *vdp = ITOV(dp);
937 int error;
938
939 if ((error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
940 return (error);
941 ep->inode = ip->i_number;
942 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
943 EXT2_FEATURE_INCOMPAT_FILETYPE))
944 ep->file_type = DTTOFT(IFTODT(ip->i_mode));
945 else
946 ep->file_type = EXT2_FT_UNKNOWN;
947 error = BUF_WRITE(bp);
948 dp->i_flag |= IN_CHANGE | IN_UPDATE;
949 return (error);
950}
951
952/*
953 * Check if a directory is empty or not.
954 * Inode supplied must be locked.
955 *
956 * Using a struct dirtemplate here is not precisely
957 * what we want, but better than using a struct direct.
958 *
959 * NB: does not handle corrupted directories.
960 */
961int
962ext2_dirempty(ip, parentino, cred)
963 register struct inode *ip;
964 ino_t parentino;
965 struct ucred *cred;
966{
967 register off_t off;
968 struct dirtemplate dbuf;
969 register struct ext2_dir_entry_2 *dp = (struct ext2_dir_entry_2 *)&dbuf;
970 int error, count, namlen;
971
972#define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
973
974 for (off = 0; off < ip->i_size; off += dp->rec_len) {
975 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
976 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
977 /*
978 * Since we read MINDIRSIZ, residual must
979 * be 0 unless we're at end of file.
980 */
981 if (error || count != 0)
982 return (0);
983 /* avoid infinite loops */
984 if (dp->rec_len == 0)
985 return (0);
986 /* skip empty entries */
987 if (dp->inode == 0)
988 continue;
989 /* accept only "." and ".." */
990 namlen = dp->name_len;
991 if (namlen > 2)
992 return (0);
993 if (dp->name[0] != '.')
994 return (0);
995 /*
996 * At this point namlen must be 1 or 2.
997 * 1 implies ".", 2 implies ".." if second
998 * char is also "."
999 */
1000 if (namlen == 1)
1001 continue;
1002 if (dp->name[1] == '.' && dp->inode == parentino)
1003 continue;
1004 return (0);
1005 }
1006 return (1);
1007}
1008
1009/*
1010 * Check if source directory is in the path of the target directory.
1011 * Target is supplied locked, source is unlocked.
1012 * The target is always vput before returning.
1013 */
1014int
1015ext2_checkpath(source, target, cred)
1016 struct inode *source, *target;
1017 struct ucred *cred;
1018{
1019 struct vnode *vp;
1020 int error, rootino, namlen;
1021 struct dirtemplate dirbuf;
1022
1023 vp = ITOV(target);
1024 if (target->i_number == source->i_number) {
1025 error = EEXIST;
1026 goto out;
1027 }
1028 rootino = ROOTINO;
1029 error = 0;
1030 if (target->i_number == rootino)
1031 goto out;
1032
1033 for (;;) {
1034 if (vp->v_type != VDIR) {
1035 error = ENOTDIR;
1036 break;
1037 }
1038 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1039 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1040 IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
1041 if (error != 0)
1042 break;
1043 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1044 if (namlen != 2 ||
1045 dirbuf.dotdot_name[0] != '.' ||
1046 dirbuf.dotdot_name[1] != '.') {
1047 error = ENOTDIR;
1048 break;
1049 }
1050 if (dirbuf.dotdot_ino == source->i_number) {
1051 error = EINVAL;
1052 break;
1053 }
1054 if (dirbuf.dotdot_ino == rootino)
1055 break;
1056 vput(vp);
1057 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp)) != 0) {
1058 vp = NULL;
1059 break;
1060 }
1061 }
1062
1063out:
1064 if (error == ENOTDIR)
1065 printf("checkpath: .. not a directory\n");
1066 if (vp != NULL)
1067 vput(vp);
1068 return (error);
1069}
1070