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 /*-
8 * Copyright (c) 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 * (c) UNIX System Laboratories, Inc.
11 * All or some portions of this file are derived from material licensed
12 * to the University of California by American Telephone and Telegraph
13 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14 * the permission of UNIX System Laboratories, Inc.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 *
40 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
41 * $FreeBSD: releng/10.3/sys/fs/ext2fs/ext2_lookup.c 281841 2015-04-22 00:38:11Z pfg $
42 */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/namei.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/endian.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/malloc.h>
53 #include <sys/dirent.h>
54 #include <sys/sysctl.h>
55
56 #include <ufs/ufs/dir.h>
57
58 #include <fs/ext2fs/inode.h>
59 #include <fs/ext2fs/ext2_mount.h>
60 #include <fs/ext2fs/ext2fs.h>
61 #include <fs/ext2fs/ext2_dinode.h>
62 #include <fs/ext2fs/ext2_dir.h>
63 #include <fs/ext2fs/ext2_extern.h>
64
65 #ifdef INVARIANTS
66 static int dirchk = 1;
67 #else
68 static int dirchk = 0;
69 #endif
70
71 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
72 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
73
74 /*
75 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
76 while it is the native blocksize in ext2fs - thus, a #define
77 is no longer appropriate
78 */
79 #undef DIRBLKSIZ
80
81 static u_char ext2_ft_to_dt[] = {
82 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
83 DT_REG, /* EXT2_FT_REG_FILE */
84 DT_DIR, /* EXT2_FT_DIR */
85 DT_CHR, /* EXT2_FT_CHRDEV */
86 DT_BLK, /* EXT2_FT_BLKDEV */
87 DT_FIFO, /* EXT2_FT_FIFO */
88 DT_SOCK, /* EXT2_FT_SOCK */
89 DT_LNK, /* EXT2_FT_SYMLINK */
90 };
91 #define FTTODT(ft) \
92 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
93
94 static 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) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
113
114 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
115 int entryoffsetinblock);
116 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
117 struct componentname *cnp, ino_t *dd_ino);
118
119 /*
120 * Vnode op for reading directories.
121 */
122 int
123 ext2_readdir(struct vop_readdir_args *ap)
124 {
125 struct vnode *vp = ap->a_vp;
126 struct uio *uio = ap->a_uio;
127 struct buf *bp;
128 struct inode *ip;
129 struct ext2fs_direct_2 *dp, *edp;
130 u_long *cookies;
131 struct dirent dstdp;
132 off_t offset, startoffset;
133 size_t readcnt, skipcnt;
134 ssize_t startresid;
135 int ncookies;
136 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
137 int error;
138
139 if (uio->uio_offset < 0)
140 return (EINVAL);
141 ip = VTOI(vp);
142 if (ap->a_ncookies != NULL) {
143 ncookies = uio->uio_resid;
144 if (uio->uio_offset >= ip->i_size)
145 ncookies = 0;
146 else if (ip->i_size - uio->uio_offset < ncookies)
147 ncookies = ip->i_size - uio->uio_offset;
148 ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
149 e2d_namlen) + 4) + 1;
150 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
151 *ap->a_ncookies = ncookies;
152 *ap->a_cookies = cookies;
153 } else {
154 ncookies = 0;
155 cookies = NULL;
156 }
157 offset = startoffset = uio->uio_offset;
158 startresid = uio->uio_resid;
159 error = 0;
160 while (error == 0 && uio->uio_resid > 0 &&
161 uio->uio_offset < ip->i_size) {
162 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
163 if (error)
164 break;
165 if (bp->b_offset + bp->b_bcount > ip->i_size)
166 readcnt = ip->i_size - bp->b_offset;
167 else
168 readcnt = bp->b_bcount;
169 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
170 ~(size_t)(DIRBLKSIZ - 1);
171 offset = bp->b_offset + skipcnt;
172 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
173 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
174 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
175 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
176 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen >
177 (caddr_t)edp) {
178 error = EIO;
179 break;
180 }
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 are binary compatible
191 * with "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-independent disk
195 * layout.
196 */
197 dstdp.d_namlen = dp->e2d_namlen;
198 dstdp.d_type = FTTODT(dp->e2d_type);
199 if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
200 dstdp.d_namlen > dp->e2d_reclen) {
201 error = EIO;
202 break;
203 }
204 if (offset < startoffset || dp->e2d_ino == 0)
205 goto nextentry;
206 dstdp.d_fileno = dp->e2d_ino;
207 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
208 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
209 dstdp.d_name[dstdp.d_namlen] = '\0';
210 if (dstdp.d_reclen > uio->uio_resid) {
211 if (uio->uio_resid == startresid)
212 error = EINVAL;
213 else
214 error = EJUSTRETURN;
215 break;
216 }
217 /* Advance dp. */
218 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
219 if (error)
220 break;
221 if (cookies != NULL) {
222 KASSERT(ncookies > 0,
223 ("ext2_readdir: cookies buffer too small"));
224 *cookies = offset + dp->e2d_reclen;
225 cookies++;
226 ncookies--;
227 }
228 nextentry:
229 offset += dp->e2d_reclen;
230 dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
231 dp->e2d_reclen);
232 }
233 bqrelse(bp);
234 uio->uio_offset = offset;
235 }
236 /* We need to correct uio_offset. */
237 uio->uio_offset = offset;
238 if (error == EJUSTRETURN)
239 error = 0;
240 if (ap->a_ncookies != NULL) {
241 if (error == 0) {
242 ap->a_ncookies -= ncookies;
243 } else {
244 free(*ap->a_cookies, M_TEMP);
245 *ap->a_ncookies = 0;
246 *ap->a_cookies = NULL;
247 }
248 }
249 if (error == 0 && ap->a_eofflag)
250 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
251 return (error);
252 }
253
254 /*
255 * Convert a component of a pathname into a pointer to a locked inode.
256 * This is a very central and rather complicated routine.
257 * If the file system is not maintained in a strict tree hierarchy,
258 * this can result in a deadlock situation (see comments in code below).
259 *
260 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
261 * on whether the name is to be looked up, created, renamed, or deleted.
262 * When CREATE, RENAME, or DELETE is specified, information usable in
263 * creating, renaming, or deleting a directory entry may be calculated.
264 * If flag has LOCKPARENT or'ed into it and the target of the pathname
265 * exists, lookup returns both the target and its parent directory locked.
266 * When creating or renaming and LOCKPARENT is specified, the target may
267 * not be ".". When deleting and LOCKPARENT is specified, the target may
268 * be "."., but the caller must check to ensure it does an vrele and vput
269 * instead of two vputs.
270 *
271 * Overall outline of ext2_lookup:
272 *
273 * search for name in directory, to found or notfound
274 * notfound:
275 * if creating, return locked directory, leaving info on available slots
276 * else return error
277 * found:
278 * if at end of path and deleting, return information to allow delete
279 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
280 * inode and return info to allow rewrite
281 * if not at end, add name to cache; if at end and neither creating
282 * nor deleting, add name to cache
283 */
284 int
285 ext2_lookup(struct vop_cachedlookup_args *ap)
286 {
287
288 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
289 }
290
291 static int
292 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
293 ino_t *dd_ino)
294 {
295 struct inode *dp; /* inode for directory being searched */
296 struct buf *bp; /* a buffer of directory entries */
297 struct ext2fs_direct_2 *ep; /* the current directory entry */
298 int entryoffsetinblock; /* offset of ep in bp's buffer */
299 enum {NONE, COMPACT, FOUND} slotstatus;
300 doff_t slotoffset; /* offset of area with free space */
301 doff_t i_diroff; /* cached i_diroff value */
302 doff_t i_offset; /* cached i_offset value */
303 int slotsize; /* size of area at slotoffset */
304 int slotfreespace; /* amount of space free in slot */
305 int slotneeded; /* size of the entry we're seeking */
306 int numdirpasses; /* strategy for directory search */
307 doff_t endsearch; /* offset to end directory search */
308 doff_t prevoff; /* prev entry dp->i_offset */
309 struct vnode *pdp; /* saved dp during symlink work */
310 struct vnode *tdp; /* returned by VFS_VGET */
311 doff_t enduseful; /* pointer past last used dir slot */
312 u_long bmask; /* block offset mask */
313 int namlen, error;
314 struct ucred *cred = cnp->cn_cred;
315 int flags = cnp->cn_flags;
316 int nameiop = cnp->cn_nameiop;
317 ino_t ino, ino1;
318 int ltype;
319
320 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
321
322 if (vpp != NULL)
323 *vpp = NULL;
324
325 dp = VTOI(vdp);
326 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
327 restart:
328 bp = NULL;
329 slotoffset = -1;
330
331 /*
332 * We now have a segment name to search for, and a directory to search.
333 */
334
335 /*
336 * Suppress search for slots unless creating
337 * file and at end of pathname, in which case
338 * we watch for a place to put the new file in
339 * case it doesn't already exist.
340 */
341 i_diroff = dp->i_diroff;
342 slotstatus = FOUND;
343 slotfreespace = slotsize = slotneeded = 0;
344 if ((nameiop == CREATE || nameiop == RENAME) &&
345 (flags & ISLASTCN)) {
346 slotstatus = NONE;
347 slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
348 /* was
349 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
350 cnp->cn_namelen + 3) &~ 3; */
351 }
352
353 /*
354 * If there is cached information on a previous search of
355 * this directory, pick up where we last left off.
356 * We cache only lookups as these are the most common
357 * and have the greatest payoff. Caching CREATE has little
358 * benefit as it usually must search the entire directory
359 * to determine that the entry does not exist. Caching the
360 * location of the last DELETE or RENAME has not reduced
361 * profiling time and hence has been removed in the interest
362 * of simplicity.
363 */
364 if (nameiop != LOOKUP || i_diroff == 0 ||
365 i_diroff > dp->i_size) {
366 entryoffsetinblock = 0;
367 i_offset = 0;
368 numdirpasses = 1;
369 } else {
370 i_offset = i_diroff;
371 if ((entryoffsetinblock = i_offset & bmask) &&
372 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
373 &bp)))
374 return (error);
375 numdirpasses = 2;
376 nchstats.ncs_2passes++;
377 }
378 prevoff = i_offset;
379 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
380 enduseful = 0;
381
382 searchloop:
383 while (i_offset < endsearch) {
384 /*
385 * If necessary, get the next directory block.
386 */
387 if ((i_offset & bmask) == 0) {
388 if (bp != NULL)
389 brelse(bp);
390 if ((error =
391 ext2_blkatoff(vdp, (off_t)i_offset, NULL,
392 &bp)) != 0)
393 return (error);
394 entryoffsetinblock = 0;
395 }
396 /*
397 * If still looking for a slot, and at a DIRBLKSIZE
398 * boundary, have to start looking for free space again.
399 */
400 if (slotstatus == NONE &&
401 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
402 slotoffset = -1;
403 slotfreespace = 0;
404 }
405 /*
406 * Get pointer to next entry.
407 * Full validation checks are slow, so we only check
408 * enough to insure forward progress through the
409 * directory. Complete checks can be run by setting
410 * "vfs.e2fs.dirchk" to be true.
411 */
412 ep = (struct ext2fs_direct_2 *)
413 ((char *)bp->b_data + entryoffsetinblock);
414 if (ep->e2d_reclen == 0 ||
415 (dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
416 int i;
417 ext2_dirbad(dp, i_offset, "mangled entry");
418 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
419 i_offset += i;
420 entryoffsetinblock += i;
421 continue;
422 }
423
424 /*
425 * If an appropriate sized slot has not yet been found,
426 * check to see if one is available. Also accumulate space
427 * in the current block so that we can determine if
428 * compaction is viable.
429 */
430 if (slotstatus != FOUND) {
431 int size = ep->e2d_reclen;
432
433 if (ep->e2d_ino != 0)
434 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
435 if (size > 0) {
436 if (size >= slotneeded) {
437 slotstatus = FOUND;
438 slotoffset = i_offset;
439 slotsize = ep->e2d_reclen;
440 } else if (slotstatus == NONE) {
441 slotfreespace += size;
442 if (slotoffset == -1)
443 slotoffset = i_offset;
444 if (slotfreespace >= slotneeded) {
445 slotstatus = COMPACT;
446 slotsize = i_offset +
447 ep->e2d_reclen - slotoffset;
448 }
449 }
450 }
451 }
452
453 /*
454 * Check for a name match.
455 */
456 if (ep->e2d_ino) {
457 namlen = ep->e2d_namlen;
458 if (namlen == cnp->cn_namelen &&
459 !bcmp(cnp->cn_nameptr, ep->e2d_name,
460 (unsigned)namlen)) {
461 /*
462 * Save directory entry's inode number and
463 * reclen in ndp->ni_ufs area, and release
464 * directory buffer.
465 */
466 ino = ep->e2d_ino;
467 goto found;
468 }
469 }
470 prevoff = i_offset;
471 i_offset += ep->e2d_reclen;
472 entryoffsetinblock += ep->e2d_reclen;
473 if (ep->e2d_ino)
474 enduseful = i_offset;
475 }
476 /* notfound: */
477 /*
478 * If we started in the middle of the directory and failed
479 * to find our target, we must check the beginning as well.
480 */
481 if (numdirpasses == 2) {
482 numdirpasses--;
483 i_offset = 0;
484 endsearch = i_diroff;
485 goto searchloop;
486 }
487 if (bp != NULL)
488 brelse(bp);
489 /*
490 * If creating, and at end of pathname and current
491 * directory has not been removed, then can consider
492 * allowing file to be created.
493 */
494 if ((nameiop == CREATE || nameiop == RENAME) &&
495 (flags & ISLASTCN) && dp->i_nlink != 0) {
496 /*
497 * Access for write is interpreted as allowing
498 * creation of files in the directory.
499 */
500 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
501 return (error);
502 /*
503 * Return an indication of where the new directory
504 * entry should be put. If we didn't find a slot,
505 * then set dp->i_count to 0 indicating
506 * that the new slot belongs at the end of the
507 * directory. If we found a slot, then the new entry
508 * can be put in the range from dp->i_offset to
509 * dp->i_offset + dp->i_count.
510 */
511 if (slotstatus == NONE) {
512 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
513 dp->i_count = 0;
514 enduseful = dp->i_offset;
515 } else {
516 dp->i_offset = slotoffset;
517 dp->i_count = slotsize;
518 if (enduseful < slotoffset + slotsize)
519 enduseful = slotoffset + slotsize;
520 }
521 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
522 /*
523 * We return with the directory locked, so that
524 * the parameters we set up above will still be
525 * valid if we actually decide to do a direnter().
526 * We return ni_vp == NULL to indicate that the entry
527 * does not currently exist; we leave a pointer to
528 * the (locked) directory inode in ndp->ni_dvp.
529 * The pathname buffer is saved so that the name
530 * can be obtained later.
531 *
532 * NB - if the directory is unlocked, then this
533 * information cannot be used.
534 */
535 cnp->cn_flags |= SAVENAME;
536 return (EJUSTRETURN);
537 }
538 /*
539 * Insert name into cache (as non-existent) if appropriate.
540 */
541 if ((cnp->cn_flags & MAKEENTRY) != 0)
542 cache_enter(vdp, NULL, cnp);
543 return (ENOENT);
544
545 found:
546 if (dd_ino != NULL)
547 *dd_ino = ino;
548 if (numdirpasses == 2)
549 nchstats.ncs_pass2++;
550 /*
551 * Check that directory length properly reflects presence
552 * of this entry.
553 */
554 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
555 > dp->i_size) {
556 ext2_dirbad(dp, i_offset, "i_size too small");
557 dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->e2d_namlen);
558 dp->i_flag |= IN_CHANGE | IN_UPDATE;
559 }
560 brelse(bp);
561
562 /*
563 * Found component in pathname.
564 * If the final component of path name, save information
565 * in the cache as to where the entry was found.
566 */
567 if ((flags & ISLASTCN) && nameiop == LOOKUP)
568 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
569 /*
570 * If deleting, and at end of pathname, return
571 * parameters which can be used to remove file.
572 */
573 if (nameiop == DELETE && (flags & ISLASTCN)) {
574 if (flags & LOCKPARENT)
575 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
576 /*
577 * Write access to directory required to delete files.
578 */
579 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
580 return (error);
581 /*
582 * Return pointer to current entry in dp->i_offset,
583 * and distance past previous entry (if there
584 * is a previous entry in this block) in dp->i_count.
585 * Save directory inode pointer in ndp->ni_dvp for dirremove().
586 *
587 * Technically we shouldn't be setting these in the
588 * WANTPARENT case (first lookup in rename()), but any
589 * lookups that will result in directory changes will
590 * overwrite these.
591 */
592 dp->i_offset = i_offset;
593 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
594 dp->i_count = 0;
595 else
596 dp->i_count = dp->i_offset - prevoff;
597 if (dd_ino != NULL)
598 return (0);
599 if (dp->i_number == ino) {
600 VREF(vdp);
601 *vpp = vdp;
602 return (0);
603 }
604 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
605 &tdp)) != 0)
606 return (error);
607 /*
608 * If directory is "sticky", then user must own
609 * the directory, or the file in it, else she
610 * may not delete it (unless she's root). This
611 * implements append-only directories.
612 */
613 if ((dp->i_mode & ISVTX) &&
614 cred->cr_uid != 0 &&
615 cred->cr_uid != dp->i_uid &&
616 VTOI(tdp)->i_uid != cred->cr_uid) {
617 vput(tdp);
618 return (EPERM);
619 }
620 *vpp = tdp;
621 return (0);
622 }
623
624 /*
625 * If rewriting (RENAME), return the inode and the
626 * information required to rewrite the present directory
627 * Must get inode of directory entry to verify it's a
628 * regular file, or empty directory.
629 */
630 if (nameiop == RENAME && (flags & ISLASTCN)) {
631 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
632 return (error);
633 /*
634 * Careful about locking second inode.
635 * This can only occur if the target is ".".
636 */
637 dp->i_offset = i_offset;
638 if (dp->i_number == ino)
639 return (EISDIR);
640 if (dd_ino != NULL)
641 return (0);
642 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
643 &tdp)) != 0)
644 return (error);
645 *vpp = tdp;
646 cnp->cn_flags |= SAVENAME;
647 return (0);
648 }
649 if (dd_ino != NULL)
650 return (0);
651
652 /*
653 * Step through the translation in the name. We do not `vput' the
654 * directory because we may need it again if a symbolic link
655 * is relative to the current directory. Instead we save it
656 * unlocked as "pdp". We must get the target inode before unlocking
657 * the directory to insure that the inode will not be removed
658 * before we get it. We prevent deadlock by always fetching
659 * inodes from the root, moving down the directory tree. Thus
660 * when following backward pointers ".." we must unlock the
661 * parent directory before getting the requested directory.
662 * There is a potential race condition here if both the current
663 * and parent directories are removed before the VFS_VGET for the
664 * inode associated with ".." returns. We hope that this occurs
665 * infrequently since we cannot avoid this race condition without
666 * implementing a sophisticated deadlock detection algorithm.
667 * Note also that this simple deadlock detection scheme will not
668 * work if the file system has any hard links other than ".."
669 * that point backwards in the directory structure.
670 */
671 pdp = vdp;
672 if (flags & ISDOTDOT) {
673 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
674 if (pdp->v_iflag & VI_DOOMED) {
675 if (error == 0)
676 vput(tdp);
677 error = ENOENT;
678 }
679 if (error)
680 return (error);
681 /*
682 * Recheck that ".." entry in the vdp directory points
683 * to the inode we looked up before vdp lock was
684 * dropped.
685 */
686 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
687 if (error) {
688 vput(tdp);
689 return (error);
690 }
691 if (ino1 != ino) {
692 vput(tdp);
693 goto restart;
694 }
695 *vpp = tdp;
696 } else if (dp->i_number == ino) {
697 VREF(vdp); /* we want ourself, ie "." */
698 /*
699 * When we lookup "." we still can be asked to lock it
700 * differently.
701 */
702 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
703 if (ltype != VOP_ISLOCKED(vdp)) {
704 if (ltype == LK_EXCLUSIVE)
705 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
706 else /* if (ltype == LK_SHARED) */
707 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
708 }
709 *vpp = vdp;
710 } else {
711 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
712 &tdp)) != 0)
713 return (error);
714 *vpp = tdp;
715 }
716
717 /*
718 * Insert name into cache if appropriate.
719 */
720 if (cnp->cn_flags & MAKEENTRY)
721 cache_enter(vdp, *vpp, cnp);
722 return (0);
723 }
724
725 void
726 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
727 {
728 struct mount *mp;
729
730 mp = ITOV(ip)->v_mount;
731 if ((mp->mnt_flag & MNT_RDONLY) == 0)
732 panic("ext2_dirbad: %s: bad dir ino %lu at offset %ld: %s\n",
733 mp->mnt_stat.f_mntonname, (u_long)ip->i_number,(long)offset, how);
734 else
735 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
736 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
737
738 }
739
740 /*
741 * Do consistency checking on a directory entry:
742 * record length must be multiple of 4
743 * entry must fit in rest of its DIRBLKSIZ block
744 * record must be large enough to contain entry
745 * name is not longer than MAXNAMLEN
746 * name must be as long as advertised, and null terminated
747 */
748 /*
749 * changed so that it confirms to ext2_check_dir_entry
750 */
751 static int
752 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
753 int entryoffsetinblock)
754 {
755 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
756
757 char * error_msg = NULL;
758
759 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
760 error_msg = "rec_len is smaller than minimal";
761 else if (de->e2d_reclen % 4 != 0)
762 error_msg = "rec_len % 4 != 0";
763 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
764 error_msg = "reclen is too small for name_len";
765 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
766 error_msg = "directory entry across blocks";
767 /* else LATER
768 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
769 error_msg = "inode out of bounds";
770 */
771
772 if (error_msg != NULL) {
773 printf("bad directory entry: %s\n", error_msg);
774 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
775 entryoffsetinblock, (unsigned long)de->e2d_ino,
776 de->e2d_reclen, de->e2d_namlen);
777 }
778 return error_msg == NULL ? 0 : 1;
779 }
780
781 /*
782 * Write a directory entry after a call to namei, using the parameters
783 * that it left in nameidata. The argument ip is the inode which the new
784 * directory entry will refer to. Dvp is a pointer to the directory to
785 * be written, which was left locked by namei. Remaining parameters
786 * (dp->i_offset, dp->i_count) indicate how the space for the new
787 * entry is to be obtained.
788 */
789 int
790 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
791 {
792 struct ext2fs_direct_2 *ep, *nep;
793 struct inode *dp;
794 struct buf *bp;
795 struct ext2fs_direct_2 newdir;
796 struct iovec aiov;
797 struct uio auio;
798 u_int dsize;
799 int error, loc, newentrysize, spacefree;
800 char *dirbuf;
801 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
802
803
804 #ifdef INVARIANTS
805 if ((cnp->cn_flags & SAVENAME) == 0)
806 panic("ext2_direnter: missing name");
807 #endif
808 dp = VTOI(dvp);
809 newdir.e2d_ino = ip->i_number;
810 newdir.e2d_namlen = cnp->cn_namelen;
811 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
812 EXT2F_INCOMPAT_FTYPE))
813 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
814 else
815 newdir.e2d_type = EXT2_FT_UNKNOWN;
816 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
817 newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen);
818 if (dp->i_count == 0) {
819 /*
820 * If dp->i_count is 0, then namei could find no
821 * space in the directory. Here, dp->i_offset will
822 * be on a directory block boundary and we will write the
823 * new entry into a fresh block.
824 */
825 if (dp->i_offset & (DIRBLKSIZ - 1))
826 panic("ext2_direnter: newblk");
827 auio.uio_offset = dp->i_offset;
828 newdir.e2d_reclen = DIRBLKSIZ;
829 auio.uio_resid = newentrysize;
830 aiov.iov_len = newentrysize;
831 aiov.iov_base = (caddr_t)&newdir;
832 auio.uio_iov = &aiov;
833 auio.uio_iovcnt = 1;
834 auio.uio_rw = UIO_WRITE;
835 auio.uio_segflg = UIO_SYSSPACE;
836 auio.uio_td = (struct thread *)0;
837 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
838 if (DIRBLKSIZ >
839 VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
840 /* XXX should grow with balloc() */
841 panic("ext2_direnter: frag size");
842 else if (!error) {
843 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ);
844 dp->i_flag |= IN_CHANGE;
845 }
846 return (error);
847 }
848
849 /*
850 * If dp->i_count is non-zero, then namei found space
851 * for the new entry in the range dp->i_offset to
852 * dp->i_offset + dp->i_count in the directory.
853 * To use this space, we may have to compact the entries located
854 * there, by copying them together towards the beginning of the
855 * block, leaving the free space in one usable chunk at the end.
856 */
857
858 /*
859 * Increase size of directory if entry eats into new space.
860 * This should never push the size past a new multiple of
861 * DIRBLKSIZE.
862 *
863 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
864 */
865 if (dp->i_offset + dp->i_count > dp->i_size)
866 dp->i_size = dp->i_offset + dp->i_count;
867 /*
868 * Get the block containing the space for the new directory entry.
869 */
870 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
871 &bp)) != 0)
872 return (error);
873 /*
874 * Find space for the new entry. In the simple case, the entry at
875 * offset base will have the space. If it does not, then namei
876 * arranged that compacting the region dp->i_offset to
877 * dp->i_offset + dp->i_count would yield the
878 * space.
879 */
880 ep = (struct ext2fs_direct_2 *)dirbuf;
881 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
882 spacefree = ep->e2d_reclen - dsize;
883 for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
884 nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
885 if (ep->e2d_ino) {
886 /* trim the existing slot */
887 ep->e2d_reclen = dsize;
888 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
889 } else {
890 /* overwrite; nothing there; header is ours */
891 spacefree += dsize;
892 }
893 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
894 spacefree += nep->e2d_reclen - dsize;
895 loc += nep->e2d_reclen;
896 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
897 }
898 /*
899 * Update the pointer fields in the previous entry (if any),
900 * copy in the new entry, and write out the block.
901 */
902 if (ep->e2d_ino == 0) {
903 if (spacefree + dsize < newentrysize)
904 panic("ext2_direnter: compact1");
905 newdir.e2d_reclen = spacefree + dsize;
906 } else {
907 if (spacefree < newentrysize)
908 panic("ext2_direnter: compact2");
909 newdir.e2d_reclen = spacefree;
910 ep->e2d_reclen = dsize;
911 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
912 }
913 bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
914 if (DOINGASYNC(dvp)) {
915 bdwrite(bp);
916 error = 0;
917 } else {
918 error = bwrite(bp);
919 }
920 dp->i_flag |= IN_CHANGE | IN_UPDATE;
921 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
922 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
923 cnp->cn_cred, cnp->cn_thread);
924 return (error);
925 }
926
927 /*
928 * Remove a directory entry after a call to namei, using
929 * the parameters which it left in nameidata. The entry
930 * dp->i_offset contains the offset into the directory of the
931 * entry to be eliminated. The dp->i_count field contains the
932 * size of the previous record in the directory. If this
933 * is 0, the first entry is being deleted, so we need only
934 * zero the inode number to mark the entry as free. If the
935 * entry is not the first in the directory, we must reclaim
936 * the space of the now empty record by adding the record size
937 * to the size of the previous entry.
938 */
939 int
940 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
941 {
942 struct inode *dp;
943 struct ext2fs_direct_2 *ep, *rep;
944 struct buf *bp;
945 int error;
946
947 dp = VTOI(dvp);
948 if (dp->i_count == 0) {
949 /*
950 * First entry in block: set d_ino to zero.
951 */
952 if ((error =
953 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
954 &bp)) != 0)
955 return (error);
956 ep->e2d_ino = 0;
957 error = bwrite(bp);
958 dp->i_flag |= IN_CHANGE | IN_UPDATE;
959 return (error);
960 }
961 /*
962 * Collapse new free space into previous entry.
963 */
964 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
965 (char **)&ep, &bp)) != 0)
966 return (error);
967
968 /* Set 'rep' to the entry being removed. */
969 if (dp->i_count == 0)
970 rep = ep;
971 else
972 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
973 ep->e2d_reclen += rep->e2d_reclen;
974 if (DOINGASYNC(dvp) && dp->i_count != 0)
975 bdwrite(bp);
976 else
977 error = bwrite(bp);
978 dp->i_flag |= IN_CHANGE | IN_UPDATE;
979 return (error);
980 }
981
982 /*
983 * Rewrite an existing directory entry to point at the inode
984 * supplied. The parameters describing the directory entry are
985 * set up by a call to namei.
986 */
987 int
988 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
989 {
990 struct buf *bp;
991 struct ext2fs_direct_2 *ep;
992 struct vnode *vdp = ITOV(dp);
993 int error;
994
995 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
996 &bp)) != 0)
997 return (error);
998 ep->e2d_ino = ip->i_number;
999 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1000 EXT2F_INCOMPAT_FTYPE))
1001 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1002 else
1003 ep->e2d_type = EXT2_FT_UNKNOWN;
1004 error = bwrite(bp);
1005 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1006 return (error);
1007 }
1008
1009 /*
1010 * Check if a directory is empty or not.
1011 * Inode supplied must be locked.
1012 *
1013 * Using a struct dirtemplate here is not precisely
1014 * what we want, but better than using a struct direct.
1015 *
1016 * NB: does not handle corrupted directories.
1017 */
1018 int
1019 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1020 {
1021 off_t off;
1022 struct dirtemplate dbuf;
1023 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1024 int error, namlen;
1025 ssize_t count;
1026 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1027
1028 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1029 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1030 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1031 NOCRED, &count, (struct thread *)0);
1032 /*
1033 * Since we read MINDIRSIZ, residual must
1034 * be 0 unless we're at end of file.
1035 */
1036 if (error || count != 0)
1037 return (0);
1038 /* avoid infinite loops */
1039 if (dp->e2d_reclen == 0)
1040 return (0);
1041 /* skip empty entries */
1042 if (dp->e2d_ino == 0)
1043 continue;
1044 /* accept only "." and ".." */
1045 namlen = dp->e2d_namlen;
1046 if (namlen > 2)
1047 return (0);
1048 if (dp->e2d_name[0] != '.')
1049 return (0);
1050 /*
1051 * At this point namlen must be 1 or 2.
1052 * 1 implies ".", 2 implies ".." if second
1053 * char is also "."
1054 */
1055 if (namlen == 1)
1056 continue;
1057 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1058 continue;
1059 return (0);
1060 }
1061 return (1);
1062 }
1063
1064 /*
1065 * Check if source directory is in the path of the target directory.
1066 * Target is supplied locked, source is unlocked.
1067 * The target is always vput before returning.
1068 */
1069 int
1070 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1071 {
1072 struct vnode *vp;
1073 int error, namlen;
1074 struct dirtemplate dirbuf;
1075
1076 vp = ITOV(target);
1077 if (target->i_number == source->i_number) {
1078 error = EEXIST;
1079 goto out;
1080 }
1081 if (target->i_number == EXT2_ROOTINO) {
1082 error = 0;
1083 goto out;
1084 }
1085
1086 for (;;) {
1087 if (vp->v_type != VDIR) {
1088 error = ENOTDIR;
1089 break;
1090 }
1091 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1092 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1093 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1094 NULL);
1095 if (error != 0)
1096 break;
1097 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1098 if (namlen != 2 ||
1099 dirbuf.dotdot_name[0] != '.' ||
1100 dirbuf.dotdot_name[1] != '.') {
1101 error = ENOTDIR;
1102 break;
1103 }
1104 if (dirbuf.dotdot_ino == source->i_number) {
1105 error = EINVAL;
1106 break;
1107 }
1108 if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1109 break;
1110 vput(vp);
1111 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1112 LK_EXCLUSIVE, &vp)) != 0) {
1113 vp = NULL;
1114 break;
1115 }
1116 }
1117
1118 out:
1119 if (error == ENOTDIR)
1120 printf("checkpath: .. not a directory\n");
1121 if (vp != NULL)
1122 vput(vp);
1123 return (error);
1124 }
Cache object: b9c4af0b954ca22262ae90e3fefa30a2
|