1 /* $OpenBSD: ufs_lookup.c,v 1.59 2022/01/11 03:13:59 jsg Exp $ */
2 /* $NetBSD: ufs_lookup.c,v 1.7 1996/02/09 22:36:06 christos Exp $ */
3
4 /*
5 * Copyright (c) 1989, 1993
6 * The Regents of the University of California. All rights reserved.
7 * (c) UNIX System Laboratories, Inc.
8 * All or some portions of this file are derived from material licensed
9 * to the University of California by American Telephone and Telegraph
10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11 * the permission of UNIX System Laboratories, Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * @(#)ufs_lookup.c 8.9 (Berkeley) 8/11/94
38 */
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/namei.h>
44 #include <sys/buf.h>
45 #include <sys/stat.h>
46 #include <sys/mount.h>
47 #include <sys/proc.h>
48 #include <sys/vnode.h>
49
50 #include <ufs/ufs/quota.h>
51 #include <ufs/ufs/inode.h>
52 #include <ufs/ufs/dir.h>
53 #ifdef UFS_DIRHASH
54 #include <ufs/ufs/dirhash.h>
55 #endif
56 #include <ufs/ufs/ufsmount.h>
57 #include <ufs/ufs/ufs_extern.h>
58
59 extern struct nchstats nchstats;
60
61 #ifdef DIAGNOSTIC
62 int dirchk = 1;
63 #else
64 int dirchk = 0;
65 #endif
66
67 #define OFSFMT(ip) ((ip)->i_ump->um_maxsymlinklen == 0)
68
69 /*
70 * Convert a component of a pathname into a pointer to a locked inode.
71 * This is a very central and rather complicated routine.
72 * If the file system is not maintained in a strict tree hierarchy,
73 * this can result in a deadlock situation (see comments in code below).
74 *
75 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
76 * on whether the name is to be looked up, created, renamed, or deleted.
77 * When CREATE, RENAME, or DELETE is specified, information usable in
78 * creating, renaming, or deleting a directory entry may be calculated.
79 * If flag has LOCKPARENT or'ed into it and the target of the pathname
80 * exists, lookup returns both the target and its parent directory locked.
81 * When creating or renaming and LOCKPARENT is specified, the target may
82 * not be ".". When deleting and LOCKPARENT is specified, the target may
83 * be "."., but the caller must check to ensure it does an vrele and vput
84 * instead of two vputs.
85 *
86 * Overall outline of ufs_lookup:
87 *
88 * check accessibility of directory
89 * look for name in cache, if found, then if at end of path
90 * and deleting or creating, drop it, else return name
91 * search for name in directory, to found or notfound
92 * notfound:
93 * if creating, return locked directory, leaving info on available slots
94 * else return error
95 * found:
96 * if at end of path and deleting, return information to allow delete
97 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
98 * inode and return info to allow rewrite
99 * if not at end, add name to cache; if at end and neither creating
100 * nor deleting, add name to cache
101 */
102 int
103 ufs_lookup(void *v)
104 {
105 struct vop_lookup_args *ap = v;
106 struct vnode *vdp; /* vnode for directory being searched */
107 struct inode *dp; /* inode for directory being searched */
108 struct buf *bp; /* a buffer of directory entries */
109 struct direct *ep; /* the current directory entry */
110 int entryoffsetinblock; /* offset of ep in bp's buffer */
111 enum {NONE, COMPACT, FOUND} slotstatus;
112 doff_t slotoffset; /* offset of area with free space */
113 int slotsize; /* size of area at slotoffset */
114 int slotfreespace; /* amount of space free in slot */
115 int slotneeded; /* size of the entry we're seeking */
116 int numdirpasses; /* strategy for directory search */
117 doff_t endsearch; /* offset to end directory search */
118 doff_t prevoff; /* prev entry dp->i_offset */
119 struct vnode *pdp; /* saved dp during symlink work */
120 struct vnode *tdp; /* returned by VFS_VGET */
121 doff_t enduseful; /* pointer past last used dir slot */
122 u_long bmask; /* block offset mask */
123 int lockparent; /* 1 => lockparent flag is set */
124 int wantparent; /* 1 => wantparent or lockparent flag */
125 int namlen, error;
126 struct vnode **vpp = ap->a_vpp;
127 struct componentname *cnp = ap->a_cnp;
128 struct ucred *cred = cnp->cn_cred;
129 int flags;
130 int nameiop = cnp->cn_nameiop;
131
132 cnp->cn_flags &= ~PDIRUNLOCK;
133 flags = cnp->cn_flags;
134
135 bp = NULL;
136 slotoffset = -1;
137 *vpp = NULL;
138 vdp = ap->a_dvp;
139 dp = VTOI(vdp);
140 lockparent = flags & LOCKPARENT;
141 wantparent = flags & (LOCKPARENT|WANTPARENT);
142
143 /*
144 * Check accessibility of directory.
145 */
146 if ((DIP(dp, mode) & IFMT) != IFDIR)
147 return (ENOTDIR);
148 if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
149 return (error);
150
151 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
152 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
153 return (EROFS);
154
155 /*
156 * We now have a segment name to search for, and a directory to search.
157 *
158 * Before tediously performing a linear scan of the directory,
159 * check the name cache to see if the directory/name pair
160 * we are looking for is known already.
161 */
162 if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
163 return (error);
164
165 /*
166 * Suppress search for slots unless creating
167 * file and at end of pathname, in which case
168 * we watch for a place to put the new file in
169 * case it doesn't already exist.
170 */
171 slotstatus = FOUND;
172 slotfreespace = slotsize = slotneeded = 0;
173 if ((nameiop == CREATE || nameiop == RENAME) &&
174 (flags & ISLASTCN)) {
175 slotstatus = NONE;
176 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
177 cnp->cn_namelen + 3) &~ 3;
178 }
179
180 /*
181 * If there is cached information on a previous search of
182 * this directory, pick up where we last left off.
183 * We cache only lookups as these are the most common
184 * and have the greatest payoff. Caching CREATE has little
185 * benefit as it usually must search the entire directory
186 * to determine that the entry does not exist. Caching the
187 * location of the last DELETE or RENAME has not reduced
188 * profiling time and hence has been removed in the interest
189 * of simplicity.
190 */
191 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
192
193 #ifdef UFS_DIRHASH
194 /*
195 * Use dirhash for fast operations on large directories. The logic
196 * to determine whether to hash the directory is contained within
197 * ufsdirhash_build(); a zero return means that it decided to hash
198 * this directory and it successfully built up the hash table.
199 */
200 if (ufsdirhash_build(dp) == 0) {
201 /* Look for a free slot if needed. */
202 enduseful = DIP(dp, size);
203 if (slotstatus != FOUND) {
204 slotoffset = ufsdirhash_findfree(dp, slotneeded,
205 &slotsize);
206 if (slotoffset >= 0) {
207 slotstatus = COMPACT;
208 enduseful = ufsdirhash_enduseful(dp);
209 if (enduseful < 0)
210 enduseful = DIP(dp, size);
211 }
212 }
213 /* Look up the component. */
214 numdirpasses = 1;
215 entryoffsetinblock = 0; /* silence compiler warning */
216 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
217 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
218 case 0:
219 ep = (struct direct *)((char *)bp->b_data +
220 (dp->i_offset & bmask));
221 goto foundentry;
222 case ENOENT:
223 #define roundup2(x, y) (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */
224 dp->i_offset = roundup2(DIP(dp, size), DIRBLKSIZ);
225 goto notfound;
226 default:
227 /* Something failed; just do a linear search. */
228 break;
229 }
230 }
231 #endif /* UFS_DIRHASH */
232
233 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
234 dp->i_diroff >= DIP(dp, size)) {
235 entryoffsetinblock = 0;
236 dp->i_offset = 0;
237 numdirpasses = 1;
238 } else {
239 dp->i_offset = dp->i_diroff;
240 if ((entryoffsetinblock = dp->i_offset & bmask) &&
241 (error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, NULL, &bp)))
242 return (error);
243 numdirpasses = 2;
244 nchstats.ncs_2passes++;
245 }
246 prevoff = dp->i_offset;
247 endsearch = roundup(DIP(dp, size), DIRBLKSIZ);
248 enduseful = 0;
249
250 searchloop:
251 while (dp->i_offset < endsearch) {
252 /*
253 * If necessary, get the next directory block.
254 */
255 if ((dp->i_offset & bmask) == 0) {
256 if (bp != NULL)
257 brelse(bp);
258 error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, NULL,
259 &bp);
260 if (error)
261 return (error);
262 entryoffsetinblock = 0;
263 }
264 /*
265 * If still looking for a slot, and at a DIRBLKSIZE
266 * boundary, have to start looking for free space again.
267 */
268 if (slotstatus == NONE &&
269 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
270 slotoffset = -1;
271 slotfreespace = 0;
272 }
273 /*
274 * Get pointer to next entry.
275 * Full validation checks are slow, so we only check
276 * enough to insure forward progress through the
277 * directory. Complete checks can be run by patching
278 * "dirchk" to be true.
279 */
280 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
281 if (ep->d_reclen == 0 ||
282 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
283 int i;
284
285 ufs_dirbad(dp, dp->i_offset, "mangled entry");
286 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
287 dp->i_offset += i;
288 entryoffsetinblock += i;
289 continue;
290 }
291
292 /*
293 * If an appropriate sized slot has not yet been found,
294 * check to see if one is available. Also accumulate space
295 * in the current block so that we can determine if
296 * compaction is viable.
297 */
298 if (slotstatus != FOUND) {
299 int size = ep->d_reclen;
300
301 if (ep->d_ino != 0)
302 size -= DIRSIZ(OFSFMT(dp), ep);
303 if (size > 0) {
304 if (size >= slotneeded) {
305 slotstatus = FOUND;
306 slotoffset = dp->i_offset;
307 slotsize = ep->d_reclen;
308 } else if (slotstatus == NONE) {
309 slotfreespace += size;
310 if (slotoffset == -1)
311 slotoffset = dp->i_offset;
312 if (slotfreespace >= slotneeded) {
313 slotstatus = COMPACT;
314 slotsize = dp->i_offset +
315 ep->d_reclen - slotoffset;
316 }
317 }
318 }
319 }
320
321 /*
322 * Check for a name match.
323 */
324 if (ep->d_ino) {
325 # if (BYTE_ORDER == LITTLE_ENDIAN)
326 if (OFSFMT(dp))
327 namlen = ep->d_type;
328 else
329 namlen = ep->d_namlen;
330 # else
331 namlen = ep->d_namlen;
332 # endif
333 if (namlen == cnp->cn_namelen &&
334 !memcmp(cnp->cn_nameptr, ep->d_name, namlen)) {
335 #ifdef UFS_DIRHASH
336 foundentry:
337 #endif
338 /*
339 * Save directory entry's inode number and
340 * reclen in ndp->ni_ufs area, and release
341 * directory buffer.
342 */
343 dp->i_ino = ep->d_ino;
344 dp->i_reclen = ep->d_reclen;
345 goto found;
346 }
347 }
348 prevoff = dp->i_offset;
349 dp->i_offset += ep->d_reclen;
350 entryoffsetinblock += ep->d_reclen;
351 if (ep->d_ino)
352 enduseful = dp->i_offset;
353 }
354 #ifdef UFS_DIRHASH
355 notfound:
356 #endif
357 /*
358 * If we started in the middle of the directory and failed
359 * to find our target, we must check the beginning as well.
360 */
361 if (numdirpasses == 2) {
362 numdirpasses--;
363 dp->i_offset = 0;
364 endsearch = dp->i_diroff;
365 goto searchloop;
366 }
367 if (bp != NULL)
368 brelse(bp);
369 /*
370 * If creating, and at end of pathname and current
371 * directory has not been removed, then can consider
372 * allowing file to be created.
373 */
374 if ((nameiop == CREATE || nameiop == RENAME) &&
375 (flags & ISLASTCN) && dp->i_effnlink != 0) {
376 /*
377 * Access for write is interpreted as allowing
378 * creation of files in the directory.
379 */
380 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
381 if (error)
382 return (error);
383 /*
384 * Return an indication of where the new directory
385 * entry should be put. If we didn't find a slot,
386 * then set dp->i_count to 0 indicating
387 * that the new slot belongs at the end of the
388 * directory. If we found a slot, then the new entry
389 * can be put in the range from dp->i_offset to
390 * dp->i_offset + dp->i_count.
391 */
392 if (slotstatus == NONE) {
393 dp->i_offset = roundup(DIP(dp, size), DIRBLKSIZ);
394 dp->i_count = 0;
395 enduseful = dp->i_offset;
396 } else if (nameiop == DELETE) {
397 dp->i_offset = slotoffset;
398 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
399 dp->i_count = 0;
400 else
401 dp->i_count = dp->i_offset - prevoff;
402 } else {
403 dp->i_offset = slotoffset;
404 dp->i_count = slotsize;
405 if (enduseful < slotoffset + slotsize)
406 enduseful = slotoffset + slotsize;
407 }
408 dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
409 /*
410 * We return with the directory locked, so that
411 * the parameters we set up above will still be
412 * valid if we actually decide to do a direnter().
413 * We return ni_vp == NULL to indicate that the entry
414 * does not currently exist; we leave a pointer to
415 * the (locked) directory inode in ndp->ni_dvp.
416 * The pathname buffer is saved so that the name
417 * can be obtained later.
418 *
419 * NB - if the directory is unlocked, then this
420 * information cannot be used.
421 */
422 cnp->cn_flags |= SAVENAME;
423 if (!lockparent) {
424 VOP_UNLOCK(vdp);
425 cnp->cn_flags |= PDIRUNLOCK;
426 }
427 return (EJUSTRETURN);
428 }
429 /*
430 * Insert name into cache (as non-existent) if appropriate.
431 */
432 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
433 cache_enter(vdp, *vpp, cnp);
434 return (ENOENT);
435
436 found:
437 if (numdirpasses == 2)
438 nchstats.ncs_pass2++;
439 /*
440 * Check that directory length properly reflects presence
441 * of this entry.
442 */
443 if (dp->i_offset + DIRSIZ(OFSFMT(dp), ep) > DIP(dp, size)) {
444 ufs_dirbad(dp, dp->i_offset, "i_ffs_size too small");
445 DIP_ASSIGN(dp, size, dp->i_offset + DIRSIZ(OFSFMT(dp), ep));
446 dp->i_flag |= IN_CHANGE | IN_UPDATE;
447 }
448 brelse(bp);
449
450 /*
451 * Found component in pathname.
452 * If the final component of path name, save information
453 * in the cache as to where the entry was found.
454 */
455 if ((flags & ISLASTCN) && nameiop == LOOKUP)
456 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
457
458 /*
459 * If deleting, and at end of pathname, return
460 * parameters which can be used to remove file.
461 * If the wantparent flag isn't set, we return only
462 * the directory (in ndp->ni_dvp), otherwise we go
463 * on and lock the inode, being careful with ".".
464 */
465 if (nameiop == DELETE && (flags & ISLASTCN)) {
466 /*
467 * Write access to directory required to delete files.
468 */
469 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
470 if (error)
471 return (error);
472 /*
473 * Return pointer to current entry in dp->i_offset,
474 * and distance past previous entry (if there
475 * is a previous entry in this block) in dp->i_count.
476 * Save directory inode pointer in ndp->ni_dvp for dirremove().
477 */
478 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
479 dp->i_count = 0;
480 else
481 dp->i_count = dp->i_offset - prevoff;
482 if (dp->i_number == dp->i_ino) {
483 vref(vdp);
484 *vpp = vdp;
485 return (0);
486 }
487 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
488 if (error)
489 return (error);
490 /*
491 * If directory is "sticky", then user must own
492 * the directory, or the file in it, else she
493 * may not delete it (unless she's root). This
494 * implements append-only directories.
495 */
496 if ((DIP(dp, mode) & ISVTX) &&
497 cred->cr_uid != 0 &&
498 cred->cr_uid != DIP(dp, uid) &&
499 !vnoperm(vdp) &&
500 DIP(VTOI(tdp), uid) != cred->cr_uid) {
501 vput(tdp);
502 return (EPERM);
503 }
504 *vpp = tdp;
505 if (!lockparent) {
506 VOP_UNLOCK(vdp);
507 cnp->cn_flags |= PDIRUNLOCK;
508 }
509 return (0);
510 }
511
512 /*
513 * If rewriting (RENAME), return the inode and the
514 * information required to rewrite the present directory
515 * Must get inode of directory entry to verify it's a
516 * regular file, or empty directory.
517 */
518 if (nameiop == RENAME && wantparent &&
519 (flags & ISLASTCN)) {
520 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
521 if (error)
522 return (error);
523 /*
524 * Careful about locking second inode.
525 * This can only occur if the target is ".".
526 */
527 if (dp->i_number == dp->i_ino)
528 return (EISDIR);
529 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
530 if (error)
531 return (error);
532 *vpp = tdp;
533 cnp->cn_flags |= SAVENAME;
534 if (!lockparent) {
535 VOP_UNLOCK(vdp);
536 cnp->cn_flags |= PDIRUNLOCK;
537 }
538 return (0);
539 }
540
541 /*
542 * Step through the translation in the name. We do not `vput' the
543 * directory because we may need it again if a symbolic link
544 * is relative to the current directory. Instead we save it
545 * unlocked as "pdp". We must get the target inode before unlocking
546 * the directory to insure that the inode will not be removed
547 * before we get it. We prevent deadlock by always fetching
548 * inodes from the root, moving down the directory tree. Thus
549 * when following backward pointers ".." we must unlock the
550 * parent directory before getting the requested directory.
551 * There is a potential race condition here if both the current
552 * and parent directories are removed before the VFS_VGET for the
553 * inode associated with ".." returns. We hope that this occurs
554 * infrequently since we cannot avoid this race condition without
555 * implementing a sophisticated deadlock detection algorithm.
556 * Note also that this simple deadlock detection scheme will not
557 * work if the file system has any hard links other than ".."
558 * that point backwards in the directory structure.
559 */
560 pdp = vdp;
561 if (flags & ISDOTDOT) {
562 VOP_UNLOCK(pdp); /* race to get the inode */
563 cnp->cn_flags |= PDIRUNLOCK;
564 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
565 if (error) {
566 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
567 cnp->cn_flags &= ~PDIRUNLOCK;
568 return (error);
569 }
570 if (lockparent && (flags & ISLASTCN)) {
571 if ((error = vn_lock(pdp, LK_EXCLUSIVE))) {
572 vput(tdp);
573 return (error);
574 }
575 cnp->cn_flags &= ~PDIRUNLOCK;
576 }
577 *vpp = tdp;
578 } else if (dp->i_number == dp->i_ino) {
579 vref(vdp); /* we want ourself, ie "." */
580 *vpp = vdp;
581 } else {
582 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
583 if (error)
584 return (error);
585 if (!lockparent || !(flags & ISLASTCN)) {
586 VOP_UNLOCK(pdp);
587 cnp->cn_flags |= PDIRUNLOCK;
588 }
589 *vpp = tdp;
590 }
591
592 /*
593 * Insert name into cache if appropriate.
594 */
595 if (cnp->cn_flags & MAKEENTRY)
596 cache_enter(vdp, *vpp, cnp);
597 return (0);
598 }
599
600 void
601 ufs_dirbad(struct inode *ip, doff_t offset, char *how)
602 {
603 struct mount *mp;
604
605 mp = ITOV(ip)->v_mount;
606 (void)printf("%s: bad dir ino %u at offset %d: %s\n",
607 mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
608 if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
609 panic("bad dir");
610 }
611
612 /*
613 * Do consistency checking on a directory entry:
614 * record length must be multiple of 4
615 * entry must fit in rest of its DIRBLKSIZ block
616 * record must be large enough to contain entry
617 * name is not longer than MAXNAMLEN
618 * name must be as long as advertised, and null terminated
619 */
620 int
621 ufs_dirbadentry(struct vnode *vdp, struct direct *ep, int entryoffsetinblock)
622 {
623 struct inode *dp;
624 int i;
625 int namlen;
626
627 dp = VTOI(vdp);
628
629 # if (BYTE_ORDER == LITTLE_ENDIAN)
630 if (OFSFMT(dp))
631 namlen = ep->d_type;
632 else
633 namlen = ep->d_namlen;
634 # else
635 namlen = ep->d_namlen;
636 # endif
637 if ((ep->d_reclen & 0x3) != 0 ||
638 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
639 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
640 /*return (1); */
641 printf("First bad\n");
642 goto bad;
643 }
644 if (ep->d_ino == 0)
645 return (0);
646 for (i = 0; i < namlen; i++)
647 if (ep->d_name[i] == '\0') {
648 /*return (1); */
649 printf("Second bad\n");
650 goto bad;
651 }
652 if (ep->d_name[i])
653 goto bad;
654 return (0);
655 bad:
656 return (1);
657 }
658
659 /*
660 * Construct a new directory entry after a call to namei, using the
661 * parameters that it left in the componentname argument cnp. The
662 * argument ip is the inode to which the new directory entry will refer.
663 */
664 void
665 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
666 struct direct *newdirp)
667 {
668 #ifdef DIAGNOSTIC
669 if ((cnp->cn_flags & SAVENAME) == 0)
670 panic("ufs_makedirentry: missing name");
671 #endif
672 newdirp->d_ino = ip->i_number;
673 newdirp->d_namlen = cnp->cn_namelen;
674 memset(newdirp->d_name + (cnp->cn_namelen & ~(DIR_ROUNDUP-1)),
675 0, DIR_ROUNDUP);
676 memcpy(newdirp->d_name, cnp->cn_nameptr, cnp->cn_namelen);
677 if (OFSFMT(ip)) {
678 newdirp->d_type = 0;
679 # if (BYTE_ORDER == LITTLE_ENDIAN)
680 { u_char tmp = newdirp->d_namlen;
681 newdirp->d_namlen = newdirp->d_type;
682 newdirp->d_type = tmp; }
683 # endif
684 } else
685 newdirp->d_type = IFTODT(DIP(ip, mode));
686 }
687
688 /*
689 * Write a directory entry after a call to namei, using the parameters
690 * that it left in nameidata. The argument dirp is the new directory
691 * entry contents. Dvp is a pointer to the directory to be written,
692 * which was left locked by namei. Remaining parameters (dp->i_offset,
693 * dp->i_count) indicate how the space for the new entry is to be obtained.
694 * Non-null bp indicates that a directory is being created (for the
695 * soft dependency code).
696 */
697 int
698 ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
699 struct componentname *cnp, struct buf *newdirbp)
700 {
701 struct ucred *cr;
702 struct proc *p;
703 int newentrysize;
704 struct inode *dp;
705 struct buf *bp;
706 u_int dsize;
707 struct direct *ep, *nep;
708 int error, ret, blkoff, loc, spacefree, flags;
709 char *dirbuf;
710
711 error = 0;
712 cr = cnp->cn_cred;
713 p = cnp->cn_proc;
714 dp = VTOI(dvp);
715 newentrysize = DIRSIZ(OFSFMT(dp), dirp);
716
717 if (dp->i_count == 0) {
718 /*
719 * If dp->i_count is 0, then namei could find no
720 * space in the directory. Here, dp->i_offset will
721 * be on a directory block boundary and we will write the
722 * new entry into a fresh block.
723 */
724 if (dp->i_offset & (DIRBLKSIZ - 1))
725 panic("ufs_direnter: newblk");
726 flags = B_CLRBUF;
727 if (!DOINGSOFTDEP(dvp))
728 flags |= B_SYNC;
729 if ((error = UFS_BUF_ALLOC(dp, (off_t)dp->i_offset, DIRBLKSIZ,
730 cr, flags, &bp)) != 0) {
731 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
732 bdwrite(newdirbp);
733 return (error);
734 }
735 DIP_ASSIGN(dp, size, dp->i_offset + DIRBLKSIZ);
736 dp->i_flag |= IN_CHANGE | IN_UPDATE;
737 uvm_vnp_setsize(dvp, DIP(dp, size));
738 dirp->d_reclen = DIRBLKSIZ;
739 blkoff = dp->i_offset &
740 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
741 memcpy(bp->b_data + blkoff, dirp, newentrysize);
742
743 #ifdef UFS_DIRHASH
744 if (dp->i_dirhash != NULL) {
745 ufsdirhash_newblk(dp, dp->i_offset);
746 ufsdirhash_add(dp, dirp, dp->i_offset);
747 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
748 dp->i_offset);
749 }
750 #endif
751
752 if (DOINGSOFTDEP(dvp)) {
753 /*
754 * Ensure that the entire newly allocated block is a
755 * valid directory so that future growth within the
756 * block does not have to ensure that the block is
757 * written before the inode.
758 */
759 blkoff += DIRBLKSIZ;
760 while (blkoff < bp->b_bcount) {
761 ((struct direct *)
762 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
763 blkoff += DIRBLKSIZ;
764 }
765 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
766 dirp->d_ino, newdirbp, 1) == 0) {
767 bdwrite(bp);
768 return (UFS_UPDATE(dp, 0));
769 }
770 /* We have just allocated a directory block in an
771 * indirect block. Rather than tracking when it gets
772 * claimed by the inode, we simply do a VOP_FSYNC
773 * now to ensure that it is there (in case the user
774 * does a future fsync). Note that we have to unlock
775 * the inode for the entry that we just entered, as
776 * the VOP_FSYNC may need to lock other inodes which
777 * can lead to deadlock if we also hold a lock on
778 * the newly entered node.
779 */
780 if ((error = VOP_BWRITE(bp)))
781 return (error);
782 if (tvp != NULL)
783 VOP_UNLOCK(tvp);
784 error = VOP_FSYNC(dvp, p->p_ucred, MNT_WAIT, p);
785 if (tvp != NULL)
786 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
787 return (error);
788 }
789 error = VOP_BWRITE(bp);
790 ret = UFS_UPDATE(dp, !DOINGSOFTDEP(dvp));
791 if (error == 0)
792 return (ret);
793 return (error);
794 }
795
796 /*
797 * If dp->i_count is non-zero, then namei found space for the new
798 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
799 * in the directory. To use this space, we may have to compact
800 * the entries located there, by copying them together towards the
801 * beginning of the block, leaving the free space in one usable
802 * chunk at the end.
803 */
804
805 /*
806 * Increase size of directory if entry eats into new space.
807 * This should never push the size past a new multiple of
808 * DIRBLKSIZE.
809 *
810 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
811 */
812 if (dp->i_offset + dp->i_count > DIP(dp, size))
813 DIP_ASSIGN(dp, size, dp->i_offset + dp->i_count);
814 /*
815 * Get the block containing the space for the new directory entry.
816 */
817 if ((error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, &dirbuf, &bp))
818 != 0) {
819 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
820 bdwrite(newdirbp);
821 return (error);
822 }
823 /*
824 * Find space for the new entry. In the simple case, the entry at
825 * offset base will have the space. If it does not, then namei
826 * arranged that compacting the region dp->i_offset to
827 * dp->i_offset + dp->i_count would yield the space.
828 */
829 ep = (struct direct *)dirbuf;
830 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dp), ep) : 0;
831 spacefree = ep->d_reclen - dsize;
832 for (loc = ep->d_reclen; loc < dp->i_count; ) {
833 nep = (struct direct *)(dirbuf + loc);
834
835 /* Trim the existing slot (NB: dsize may be zero). */
836 ep->d_reclen = dsize;
837 ep = (struct direct *)((char *)ep + dsize);
838
839 /* Read nep->d_reclen now as the memmove() may clobber it. */
840 loc += nep->d_reclen;
841 if (nep->d_ino == 0) {
842 /*
843 * A mid-block unused entry. Such entries are
844 * never created by the kernel, but fsck_ffs
845 * can create them (and it doesn't fix them).
846 *
847 * Add up the free space, and initialise the
848 * relocated entry since we don't memmove it.
849 */
850 spacefree += nep->d_reclen;
851 ep->d_ino = 0;
852 dsize = 0;
853 continue;
854 }
855 dsize = DIRSIZ(OFSFMT(dp), nep);
856 spacefree += nep->d_reclen - dsize;
857 #ifdef UFS_DIRHASH
858 if (dp->i_dirhash != NULL)
859 ufsdirhash_move(dp, nep,
860 dp->i_offset + ((char *)nep - dirbuf),
861 dp->i_offset + ((char *)ep - dirbuf));
862 #endif
863 if (DOINGSOFTDEP(dvp))
864 softdep_change_directoryentry_offset(dp, dirbuf,
865 (caddr_t)nep, (caddr_t)ep, dsize);
866 else
867 memmove(ep, nep, dsize);
868 }
869 /*
870 * Here, `ep' points to a directory entry containing `dsize' in-use
871 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
872 * then the entry is completely unused (dsize == 0). The value
873 * of ep->d_reclen is always indeterminate.
874 *
875 * Update the pointer fields in the previous entry (if any),
876 * copy in the new entry, and write out the block.
877 */
878 if (ep->d_ino == 0) {
879 if (spacefree + dsize < newentrysize)
880 panic("ufs_direnter: compact1");
881 dirp->d_reclen = spacefree + dsize;
882 } else {
883 if (spacefree < newentrysize)
884 panic("ufs_direnter: compact2");
885 dirp->d_reclen = spacefree;
886 ep->d_reclen = dsize;
887 ep = (struct direct *)((char *)ep + dsize);
888 }
889
890 #ifdef UFS_DIRHASH
891 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
892 dirp->d_reclen == spacefree))
893 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
894 #endif
895 memcpy(ep, dirp, newentrysize);
896 #ifdef UFS_DIRHASH
897 if (dp->i_dirhash != NULL)
898 ufsdirhash_checkblock(dp, dirbuf -
899 (dp->i_offset & (DIRBLKSIZ - 1)),
900 dp->i_offset & ~(DIRBLKSIZ - 1));
901 #endif
902
903 if (DOINGSOFTDEP(dvp)) {
904 (void)softdep_setup_directory_add(bp, dp,
905 dp->i_offset + (caddr_t)ep - dirbuf,
906 dirp->d_ino, newdirbp, 0);
907 bdwrite(bp);
908 } else {
909 error = VOP_BWRITE(bp);
910 }
911 dp->i_flag |= IN_CHANGE | IN_UPDATE;
912
913 /*
914 * If all went well, and the directory can be shortened, proceed
915 * with the truncation. Note that we have to unlock the inode for
916 * the entry that we just entered, as the truncation may need to
917 * lock other inodes which can lead to deadlock if we also hold a
918 * lock on the newly entered node.
919 */
920
921 if (error == 0 && dp->i_endoff && dp->i_endoff < DIP(dp, size)) {
922 if (tvp != NULL)
923 VOP_UNLOCK(tvp);
924 error = UFS_TRUNCATE(dp, (off_t)dp->i_endoff, IO_SYNC, cr);
925 #ifdef UFS_DIRHASH
926 if (error == 0 && dp->i_dirhash != NULL)
927 ufsdirhash_dirtrunc(dp, dp->i_endoff);
928 #endif
929 if (tvp != NULL)
930 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
931 }
932 return (error);
933 }
934
935 /*
936 * Remove a directory entry after a call to namei, using
937 * the parameters which it left in nameidata. The entry
938 * dp->i_offset contains the offset into the directory of the
939 * entry to be eliminated. The dp->i_count field contains the
940 * size of the previous record in the directory. If this
941 * is 0, the first entry is being deleted, so we need only
942 * zero the inode number to mark the entry as free. If the
943 * entry is not the first in the directory, we must reclaim
944 * the space of the now empty record by adding the record size
945 * to the size of the previous entry.
946 */
947 int
948 ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
949 {
950 struct inode *dp;
951 struct direct *ep;
952 struct buf *bp;
953 int error;
954
955 dp = VTOI(dvp);
956
957 if ((error = UFS_BUFATOFF(dp,
958 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
959 return (error);
960 #ifdef UFS_DIRHASH
961 /*
962 * Remove the dirhash entry. This is complicated by the fact
963 * that `ep' is the previous entry when dp->i_count != 0.
964 */
965 if (dp->i_dirhash != NULL)
966 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
967 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
968 #endif
969
970 if (dp->i_count == 0) {
971 /*
972 * First entry in block: set d_ino to zero.
973 */
974 ep->d_ino = 0;
975 } else {
976 /*
977 * Collapse new free space into previous entry.
978 */
979 ep->d_reclen += dp->i_reclen;
980 }
981 #ifdef UFS_DIRHASH
982 if (dp->i_dirhash != NULL)
983 ufsdirhash_checkblock(dp, (char *)ep -
984 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
985 dp->i_offset & ~(DIRBLKSIZ - 1));
986 #endif
987 if (DOINGSOFTDEP(dvp)) {
988 if (ip) {
989 ip->i_effnlink--;
990 softdep_change_linkcnt(ip, 0);
991 softdep_setup_remove(bp, dp, ip, isrmdir);
992 }
993 if (softdep_slowdown(dvp)) {
994 error = bwrite(bp);
995 } else {
996 bdwrite(bp);
997 error = 0;
998 }
999 } else {
1000 if (ip) {
1001 ip->i_effnlink--;
1002 DIP_ADD(ip, nlink, -1);
1003 ip->i_flag |= IN_CHANGE;
1004 }
1005 if (DOINGASYNC(dvp) && dp->i_count != 0) {
1006 bdwrite(bp);
1007 error = 0;
1008 } else
1009 error = bwrite(bp);
1010 }
1011 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1012 return (error);
1013 }
1014
1015 /*
1016 * Rewrite an existing directory entry to point at the inode
1017 * supplied. The parameters describing the directory entry are
1018 * set up by a call to namei.
1019 */
1020 int
1021 ufs_dirrewrite(struct inode *dp, struct inode *oip, ufsino_t newinum,
1022 int newtype, int isrmdir)
1023 {
1024 struct buf *bp;
1025 struct direct *ep;
1026 struct vnode *vdp = ITOV(dp);
1027 int error;
1028
1029 error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
1030 if (error)
1031 return (error);
1032 ep->d_ino = newinum;
1033 if (!OFSFMT(dp))
1034 ep->d_type = newtype;
1035 oip->i_effnlink--;
1036 if (DOINGSOFTDEP(vdp)) {
1037 softdep_change_linkcnt(oip, 0);
1038 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1039 bdwrite(bp);
1040 } else {
1041 DIP_ADD(oip, nlink, -1);
1042 oip->i_flag |= IN_CHANGE;
1043 if (DOINGASYNC(vdp)) {
1044 bdwrite(bp);
1045 error = 0;
1046 } else {
1047 error = VOP_BWRITE(bp);
1048 }
1049 }
1050 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1051 return (error);
1052 }
1053
1054 /*
1055 * Check if a directory is empty or not.
1056 * Inode supplied must be locked.
1057 *
1058 * Using a struct dirtemplate here is not precisely
1059 * what we want, but better than using a struct direct.
1060 *
1061 * NB: does not handle corrupted directories.
1062 */
1063 int
1064 ufs_dirempty(struct inode *ip, ufsino_t parentino, struct ucred *cred)
1065 {
1066 off_t off, m;
1067 struct dirtemplate dbuf;
1068 struct direct *dp = (struct direct *)&dbuf;
1069 int error, namlen;
1070 size_t count;
1071 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1072
1073 m = DIP(ip, size);
1074 for (off = 0; off < m; off += dp->d_reclen) {
1075 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1076 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, curproc);
1077 /*
1078 * Since we read MINDIRSIZ, residual must
1079 * be 0 unless we're at end of file.
1080 */
1081 if (error || count != 0)
1082 return (0);
1083 /* avoid infinite loops */
1084 if (dp->d_reclen == 0)
1085 return (0);
1086 /* skip empty entries */
1087 if (dp->d_ino == 0)
1088 continue;
1089 /* accept only "." and ".." */
1090 # if (BYTE_ORDER == LITTLE_ENDIAN)
1091 if (OFSFMT(ip))
1092 namlen = dp->d_type;
1093 else
1094 namlen = dp->d_namlen;
1095 # else
1096 namlen = dp->d_namlen;
1097 # endif
1098 if (namlen > 2)
1099 return (0);
1100 if (dp->d_name[0] != '.')
1101 return (0);
1102 /*
1103 * At this point namlen must be 1 or 2.
1104 * 1 implies ".", 2 implies ".." if second
1105 * char is also "."
1106 */
1107 if (namlen == 1 && dp->d_ino == ip->i_number)
1108 continue;
1109 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1110 continue;
1111 return (0);
1112 }
1113 return (1);
1114 }
1115
1116 /*
1117 * Check if source directory is in the path of the target directory.
1118 * Target is supplied locked, source is unlocked.
1119 * The target is always vput before returning.
1120 */
1121 int
1122 ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1123 {
1124 struct vnode *nextvp, *vp;
1125 int error, rootino, namlen;
1126 struct dirtemplate dirbuf;
1127
1128 vp = ITOV(target);
1129 if (target->i_number == source->i_number) {
1130 error = EEXIST;
1131 goto out;
1132 }
1133 rootino = ROOTINO;
1134 error = 0;
1135 if (target->i_number == rootino)
1136 goto out;
1137
1138 for (;;) {
1139 if (vp->v_type != VDIR) {
1140 error = ENOTDIR;
1141 break;
1142 }
1143 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1144 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1145 IO_NODELOCKED, cred, NULL, curproc);
1146 if (error != 0)
1147 break;
1148 # if (BYTE_ORDER == LITTLE_ENDIAN)
1149 if (OFSFMT(VTOI(vp)))
1150 namlen = dirbuf.dotdot_type;
1151 else
1152 namlen = dirbuf.dotdot_namlen;
1153 # else
1154 namlen = dirbuf.dotdot_namlen;
1155 # endif
1156 if (namlen != 2 ||
1157 dirbuf.dotdot_name[0] != '.' ||
1158 dirbuf.dotdot_name[1] != '.') {
1159 error = ENOTDIR;
1160 break;
1161 }
1162 if (dirbuf.dotdot_ino == source->i_number) {
1163 error = EINVAL;
1164 break;
1165 }
1166 if (dirbuf.dotdot_ino == rootino)
1167 break;
1168 VOP_UNLOCK(vp);
1169 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &nextvp);
1170 vrele(vp);
1171 if (error) {
1172 vp = NULL;
1173 break;
1174 }
1175 vp = nextvp;
1176 }
1177
1178 out:
1179 if (error == ENOTDIR)
1180 printf("checkpath: .. not a directory\n");
1181 if (vp != NULL)
1182 vput(vp);
1183 return (error);
1184 }
Cache object: 3556a8f326bdbd259f33c08222cb8e63
|