1 /* $NetBSD: ufs_lookup.c,v 1.55 2004/03/06 06:54:12 yamt Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)ufs_lookup.c 8.9 (Berkeley) 8/11/94
37 */
38
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: ufs_lookup.c,v 1.55 2004/03/06 06:54:12 yamt Exp $");
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/namei.h>
45 #include <sys/buf.h>
46 #include <sys/file.h>
47 #include <sys/stat.h>
48 #include <sys/mount.h>
49 #include <sys/vnode.h>
50 #include <sys/kernel.h>
51
52 #include <ufs/ufs/inode.h>
53 #include <ufs/ufs/dir.h>
54 #include <ufs/ufs/ufsmount.h>
55 #include <ufs/ufs/ufs_extern.h>
56 #include <ufs/ufs/ufs_bswap.h>
57
58 #ifdef DIAGNOSTIC
59 int dirchk = 1;
60 #else
61 int dirchk = 0;
62 #endif
63
64 #define FSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
65
66 /*
67 * Convert a component of a pathname into a pointer to a locked inode.
68 * This is a very central and rather complicated routine.
69 * If the file system is not maintained in a strict tree hierarchy,
70 * this can result in a deadlock situation (see comments in code below).
71 *
72 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
73 * on whether the name is to be looked up, created, renamed, or deleted.
74 * When CREATE, RENAME, or DELETE is specified, information usable in
75 * creating, renaming, or deleting a directory entry may be calculated.
76 * If flag has LOCKPARENT or'ed into it and the target of the pathname
77 * exists, lookup returns both the target and its parent directory locked.
78 * When creating or renaming and LOCKPARENT is specified, the target may
79 * not be ".". When deleting and LOCKPARENT is specified, the target may
80 * be "."., but the caller must check to ensure it does an vrele and vput
81 * instead of two vputs.
82 *
83 * Overall outline of ufs_lookup:
84 *
85 * check accessibility of directory
86 * look for name in cache, if found, then if at end of path
87 * and deleting or creating, drop it, else return name
88 * search for name in directory, to found or notfound
89 * notfound:
90 * if creating, return locked directory, leaving info on available slots
91 * else return error
92 * found:
93 * if at end of path and deleting, return information to allow delete
94 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
95 * inode and return info to allow rewrite
96 * if not at end, add name to cache; if at end and neither creating
97 * nor deleting, add name to cache
98 */
99 int
100 ufs_lookup(v)
101 void *v;
102 {
103 struct vop_lookup_args /* {
104 struct vnode *a_dvp;
105 struct vnode **a_vpp;
106 struct componentname *a_cnp;
107 } */ *ap = v;
108 struct vnode *vdp; /* vnode for directory being searched */
109 struct inode *dp; /* inode for directory being searched */
110 struct buf *bp; /* a buffer of directory entries */
111 struct direct *ep; /* the current directory entry */
112 int entryoffsetinblock; /* offset of ep in bp's buffer */
113 enum {NONE, COMPACT, FOUND} slotstatus;
114 doff_t slotoffset; /* offset of area with free space */
115 int slotsize; /* size of area at slotoffset */
116 int slotfreespace; /* amount of space free in slot */
117 int slotneeded; /* size of the entry we're seeking */
118 int numdirpasses; /* strategy for directory search */
119 doff_t endsearch; /* offset to end directory search */
120 doff_t prevoff; /* prev entry dp->i_offset */
121 struct vnode *pdp; /* saved dp during symlink work */
122 struct vnode *tdp; /* returned by VFS_VGET */
123 doff_t enduseful; /* pointer past last used dir slot */
124 u_long bmask; /* block offset mask */
125 int lockparent; /* 1 => lockparent flag is set */
126 int wantparent; /* 1 => wantparent or lockparent flag */
127 int namlen, error;
128 struct vnode **vpp = ap->a_vpp;
129 struct componentname *cnp = ap->a_cnp;
130 struct ucred *cred = cnp->cn_cred;
131 int flags;
132 int nameiop = cnp->cn_nameiop;
133 const int needswap = UFS_MPNEEDSWAP(ap->a_dvp->v_mount);
134 int dirblksiz = DIRBLKSIZ;
135 ino_t foundino;
136 if (UFS_MPISAPPLEUFS(ap->a_dvp->v_mount)) {
137 dirblksiz = APPLEUFS_DIRBLKSIZ;
138 }
139
140 cnp->cn_flags &= ~PDIRUNLOCK;
141 flags = cnp->cn_flags;
142
143 bp = NULL;
144 slotoffset = -1;
145 *vpp = NULL;
146 vdp = ap->a_dvp;
147 dp = VTOI(vdp);
148 lockparent = flags & LOCKPARENT;
149 wantparent = flags & (LOCKPARENT|WANTPARENT);
150
151
152 /*
153 * Check accessiblity of directory.
154 */
155 if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 0)
156 return (error);
157
158 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
159 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
160 return (EROFS);
161
162 /*
163 * We now have a segment name to search for, and a directory to search.
164 *
165 * Before tediously performing a linear scan of the directory,
166 * check the name cache to see if the directory/name pair
167 * we are looking for is known already.
168 */
169 if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
170 return (error);
171
172 /*
173 * Suppress search for slots unless creating
174 * file and at end of pathname, in which case
175 * we watch for a place to put the new file in
176 * case it doesn't already exist.
177 */
178 slotstatus = FOUND;
179 slotfreespace = slotsize = slotneeded = 0;
180 if ((nameiop == CREATE || nameiop == RENAME) &&
181 (flags & ISLASTCN)) {
182 slotstatus = NONE;
183 slotneeded = (sizeof(struct direct) - MAXNAMLEN +
184 cnp->cn_namelen + 3) &~ 3;
185 }
186
187 /*
188 * If there is cached information on a previous search of
189 * this directory, pick up where we last left off.
190 * We cache only lookups as these are the most common
191 * and have the greatest payoff. Caching CREATE has little
192 * benefit as it usually must search the entire directory
193 * to determine that the entry does not exist. Caching the
194 * location of the last DELETE or RENAME has not reduced
195 * profiling time and hence has been removed in the interest
196 * of simplicity.
197 */
198 bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
199 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
200 dp->i_diroff >= dp->i_size) {
201 entryoffsetinblock = 0;
202 dp->i_offset = 0;
203 numdirpasses = 1;
204 } else {
205 dp->i_offset = dp->i_diroff;
206 if ((entryoffsetinblock = dp->i_offset & bmask) &&
207 (error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
208 return (error);
209 numdirpasses = 2;
210 nchstats.ncs_2passes++;
211 }
212 prevoff = dp->i_offset;
213 endsearch = roundup(dp->i_size, dirblksiz);
214 enduseful = 0;
215
216 searchloop:
217 while (dp->i_offset < endsearch) {
218 if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
219 preempt(1);
220 /*
221 * If necessary, get the next directory block.
222 */
223 if ((dp->i_offset & bmask) == 0) {
224 if (bp != NULL)
225 brelse(bp);
226 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, NULL,
227 &bp);
228 if (error)
229 return (error);
230 entryoffsetinblock = 0;
231 }
232 /*
233 * If still looking for a slot, and at a DIRBLKSIZE
234 * boundary, have to start looking for free space again.
235 */
236 if (slotstatus == NONE &&
237 (entryoffsetinblock & (dirblksiz - 1)) == 0) {
238 slotoffset = -1;
239 slotfreespace = 0;
240 }
241 /*
242 * Get pointer to next entry.
243 * Full validation checks are slow, so we only check
244 * enough to insure forward progress through the
245 * directory. Complete checks can be run by patching
246 * "dirchk" to be true.
247 */
248 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
249 if (ep->d_reclen == 0 ||
250 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
251 int i;
252
253 ufs_dirbad(dp, dp->i_offset, "mangled entry");
254 i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
255 dp->i_offset += i;
256 entryoffsetinblock += i;
257 continue;
258 }
259
260 /*
261 * If an appropriate sized slot has not yet been found,
262 * check to see if one is available. Also accumulate space
263 * in the current block so that we can determine if
264 * compaction is viable.
265 */
266 if (slotstatus != FOUND) {
267 int size = ufs_rw16(ep->d_reclen, needswap);
268
269 if (ep->d_ino != 0)
270 size -= DIRSIZ(FSFMT(vdp), ep, needswap);
271 if (size > 0) {
272 if (size >= slotneeded) {
273 slotstatus = FOUND;
274 slotoffset = dp->i_offset;
275 slotsize = ufs_rw16(ep->d_reclen,
276 needswap);
277 } else if (slotstatus == NONE) {
278 slotfreespace += size;
279 if (slotoffset == -1)
280 slotoffset = dp->i_offset;
281 if (slotfreespace >= slotneeded) {
282 slotstatus = COMPACT;
283 slotsize = dp->i_offset +
284 ufs_rw16(ep->d_reclen,
285 needswap)
286 - slotoffset;
287 }
288 }
289 }
290 }
291
292 /*
293 * Check for a name match.
294 */
295 if (ep->d_ino) {
296 #if (BYTE_ORDER == LITTLE_ENDIAN)
297 if (vdp->v_mount->mnt_maxsymlinklen > 0 ||
298 needswap != 0)
299 namlen = ep->d_namlen;
300 else
301 namlen = ep->d_type;
302 #else
303 if (vdp->v_mount->mnt_maxsymlinklen <= 0
304 && needswap != 0)
305 namlen = ep->d_type;
306 else
307 namlen = ep->d_namlen;
308 #endif
309 if (namlen == cnp->cn_namelen &&
310 !memcmp(cnp->cn_nameptr, ep->d_name,
311 (unsigned)namlen)) {
312 /*
313 * Save directory entry's inode number and
314 * reclen in ndp->ni_ufs area, and release
315 * directory buffer.
316 */
317 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
318 ep->d_type == DT_WHT) {
319 slotstatus = FOUND;
320 slotoffset = dp->i_offset;
321 slotsize = ufs_rw16(ep->d_reclen,
322 needswap);
323 dp->i_reclen = slotsize;
324 /*
325 * This is used to set dp->i_endoff,
326 * which may be used by ufs_direnter2()
327 * as a length to truncate the
328 * directory to. Therefore, it must
329 * point past the end of the last
330 * non-empty directory entry. We don't
331 * know where that is in this case, so
332 * we effectively disable shrinking by
333 * using the existing size of the
334 * directory.
335 *
336 * Note that we wouldn't expect to
337 * shrink the directory while rewriting
338 * an existing entry anyway.
339 */
340 enduseful = endsearch;
341 ap->a_cnp->cn_flags |= ISWHITEOUT;
342 numdirpasses--;
343 goto notfound;
344 }
345 foundino = ufs_rw32(ep->d_ino, needswap);
346 dp->i_reclen = ufs_rw16(ep->d_reclen, needswap);
347 goto found;
348 }
349 }
350 prevoff = dp->i_offset;
351 dp->i_offset += ufs_rw16(ep->d_reclen, needswap);
352 entryoffsetinblock += ufs_rw16(ep->d_reclen, needswap);
353 if (ep->d_ino)
354 enduseful = dp->i_offset;
355 }
356 notfound:
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 (nameiop == DELETE &&
376 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
377 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
378 (flags & ISLASTCN) && dp->i_ffs_effnlink != 0) {
379 /*
380 * Access for write is interpreted as allowing
381 * creation of files in the directory.
382 */
383 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
384 if (error)
385 return (error);
386 /*
387 * Return an indication of where the new directory
388 * entry should be put. If we didn't find a slot,
389 * then set dp->i_count to 0 indicating
390 * that the new slot belongs at the end of the
391 * directory. If we found a slot, then the new entry
392 * can be put in the range from dp->i_offset to
393 * dp->i_offset + dp->i_count.
394 */
395 if (slotstatus == NONE) {
396 dp->i_offset = roundup(dp->i_size, dirblksiz);
397 dp->i_count = 0;
398 enduseful = dp->i_offset;
399 } else if (nameiop == DELETE) {
400 dp->i_offset = slotoffset;
401 if ((dp->i_offset & (dirblksiz - 1)) == 0)
402 dp->i_count = 0;
403 else
404 dp->i_count = dp->i_offset - prevoff;
405 } else {
406 dp->i_offset = slotoffset;
407 dp->i_count = slotsize;
408 if (enduseful < slotoffset + slotsize)
409 enduseful = slotoffset + slotsize;
410 }
411 dp->i_endoff = roundup(enduseful, dirblksiz);
412 #if 0 /* commented out by dbj. none of the on disk fields changed */
413 dp->i_flag |= IN_CHANGE | IN_UPDATE;
414 #endif
415 /*
416 * We return with the directory locked, so that
417 * the parameters we set up above will still be
418 * valid if we actually decide to do a direnter().
419 * We return ni_vp == NULL to indicate that the entry
420 * does not currently exist; we leave a pointer to
421 * the (locked) directory inode in ndp->ni_dvp.
422 * The pathname buffer is saved so that the name
423 * can be obtained later.
424 *
425 * NB - if the directory is unlocked, then this
426 * information cannot be used.
427 */
428 cnp->cn_flags |= SAVENAME;
429 if (!lockparent) {
430 VOP_UNLOCK(vdp, 0);
431 cnp->cn_flags |= PDIRUNLOCK;
432 }
433 return (EJUSTRETURN);
434 }
435 /*
436 * Insert name into cache (as non-existent) if appropriate.
437 */
438 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
439 cache_enter(vdp, *vpp, cnp);
440 return (ENOENT);
441
442 found:
443 if (numdirpasses == 2)
444 nchstats.ncs_pass2++;
445 /*
446 * Check that directory length properly reflects presence
447 * of this entry.
448 */
449 if (dp->i_offset + DIRSIZ(FSFMT(vdp), ep, needswap) > dp->i_size) {
450 ufs_dirbad(dp, dp->i_offset, "i_size too small");
451 dp->i_size = dp->i_offset + DIRSIZ(FSFMT(vdp), ep, needswap);
452 DIP_ASSIGN(dp, size, dp->i_size);
453 dp->i_flag |= IN_CHANGE | IN_UPDATE;
454 }
455 brelse(bp);
456
457 /*
458 * Found component in pathname.
459 * If the final component of path name, save information
460 * in the cache as to where the entry was found.
461 */
462 if ((flags & ISLASTCN) && nameiop == LOOKUP)
463 dp->i_diroff = dp->i_offset &~ (dirblksiz - 1);
464
465 /*
466 * If deleting, and at end of pathname, return
467 * parameters which can be used to remove file.
468 * If the wantparent flag isn't set, we return only
469 * the directory (in ndp->ni_dvp), otherwise we go
470 * on and lock the inode, being careful with ".".
471 */
472 if (nameiop == DELETE && (flags & ISLASTCN)) {
473 /*
474 * Write access to directory required to delete files.
475 */
476 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
477 if (error)
478 return (error);
479 /*
480 * Return pointer to current entry in dp->i_offset,
481 * and distance past previous entry (if there
482 * is a previous entry in this block) in dp->i_count.
483 * Save directory inode pointer in ndp->ni_dvp for dirremove().
484 */
485 if ((dp->i_offset & (dirblksiz - 1)) == 0)
486 dp->i_count = 0;
487 else
488 dp->i_count = dp->i_offset - prevoff;
489 if (dp->i_number == foundino) {
490 VREF(vdp);
491 *vpp = vdp;
492 return (0);
493 }
494 if (flags & ISDOTDOT)
495 VOP_UNLOCK(vdp, 0); /* race to get the inode */
496 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
497 if (flags & ISDOTDOT)
498 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
499 if (error)
500 return (error);
501 /*
502 * If directory is "sticky", then user must own
503 * the directory, or the file in it, else she
504 * may not delete it (unless she's root). This
505 * implements append-only directories.
506 */
507 if ((dp->i_mode & ISVTX) &&
508 cred->cr_uid != 0 &&
509 cred->cr_uid != dp->i_uid &&
510 VTOI(tdp)->i_uid != cred->cr_uid) {
511 vput(tdp);
512 return (EPERM);
513 }
514 *vpp = tdp;
515 if (!lockparent) {
516 VOP_UNLOCK(vdp, 0);
517 cnp->cn_flags |= PDIRUNLOCK;
518 }
519 return (0);
520 }
521
522 /*
523 * If rewriting (RENAME), return the inode and the
524 * information required to rewrite the present directory
525 * Must get inode of directory entry to verify it's a
526 * regular file, or empty directory.
527 */
528 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
529 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
530 if (error)
531 return (error);
532 /*
533 * Careful about locking second inode.
534 * This can only occur if the target is ".".
535 */
536 if (dp->i_number == foundino)
537 return (EISDIR);
538 if (flags & ISDOTDOT)
539 VOP_UNLOCK(vdp, 0); /* race to get the inode */
540 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
541 if (flags & ISDOTDOT)
542 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
543 if (error)
544 return (error);
545 *vpp = tdp;
546 cnp->cn_flags |= SAVENAME;
547 if (!lockparent) {
548 VOP_UNLOCK(vdp, 0);
549 cnp->cn_flags |= PDIRUNLOCK;
550 }
551 return (0);
552 }
553
554 /*
555 * Step through the translation in the name. We do not `vput' the
556 * directory because we may need it again if a symbolic link
557 * is relative to the current directory. Instead we save it
558 * unlocked as "pdp". We must get the target inode before unlocking
559 * the directory to insure that the inode will not be removed
560 * before we get it. We prevent deadlock by always fetching
561 * inodes from the root, moving down the directory tree. Thus
562 * when following backward pointers ".." we must unlock the
563 * parent directory before getting the requested directory.
564 * There is a potential race condition here if both the current
565 * and parent directories are removed before the VFS_VGET for the
566 * inode associated with ".." returns. We hope that this occurs
567 * infrequently since we cannot avoid this race condition without
568 * implementing a sophisticated deadlock detection algorithm.
569 * Note also that this simple deadlock detection scheme will not
570 * work if the file system has any hard links other than ".."
571 * that point backwards in the directory structure.
572 */
573 pdp = vdp;
574 if (flags & ISDOTDOT) {
575 VOP_UNLOCK(pdp, 0); /* race to get the inode */
576 cnp->cn_flags |= PDIRUNLOCK;
577 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
578 if (error) {
579 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
580 cnp->cn_flags &= ~PDIRUNLOCK;
581 return (error);
582 }
583 if (lockparent && (flags & ISLASTCN)) {
584 if ((error = vn_lock(pdp, LK_EXCLUSIVE))) {
585 vput(tdp);
586 return (error);
587 }
588 cnp->cn_flags &= ~PDIRUNLOCK;
589 }
590 *vpp = tdp;
591 } else if (dp->i_number == foundino) {
592 VREF(vdp); /* we want ourself, ie "." */
593 *vpp = vdp;
594 } else {
595 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
596 if (error)
597 return (error);
598 if (!lockparent || !(flags & ISLASTCN)) {
599 VOP_UNLOCK(pdp, 0);
600 cnp->cn_flags |= PDIRUNLOCK;
601 }
602 *vpp = tdp;
603 }
604
605 /*
606 * Insert name into cache if appropriate.
607 */
608 if (cnp->cn_flags & MAKEENTRY)
609 cache_enter(vdp, *vpp, cnp);
610 return (0);
611 }
612
613 void
614 ufs_dirbad(ip, offset, how)
615 struct inode *ip;
616 doff_t offset;
617 char *how;
618 {
619 struct mount *mp;
620
621 mp = ITOV(ip)->v_mount;
622 printf("%s: bad dir ino %d at offset %d: %s\n",
623 mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
624 if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
625 panic("bad dir");
626 }
627
628 /*
629 * Do consistency checking on a directory entry:
630 * record length must be multiple of 4
631 * entry must fit in rest of its DIRBLKSIZ block
632 * record must be large enough to contain entry
633 * name is not longer than MAXNAMLEN
634 * name must be as long as advertised, and null terminated
635 */
636 int
637 ufs_dirbadentry(dp, ep, entryoffsetinblock)
638 struct vnode *dp;
639 struct direct *ep;
640 int entryoffsetinblock;
641 {
642 int i;
643 int namlen;
644 const int needswap = UFS_MPNEEDSWAP(dp->v_mount);
645 int dirblksiz = DIRBLKSIZ;
646 if (UFS_MPISAPPLEUFS(dp->v_mount)) {
647 dirblksiz = APPLEUFS_DIRBLKSIZ;
648 }
649
650 #if (BYTE_ORDER == LITTLE_ENDIAN)
651 if (dp->v_mount->mnt_maxsymlinklen > 0 || needswap != 0)
652 namlen = ep->d_namlen;
653 else
654 namlen = ep->d_type;
655 #else
656 if (dp->v_mount->mnt_maxsymlinklen <= 0 && needswap != 0)
657 namlen = ep->d_type;
658 else
659 namlen = ep->d_namlen;
660 #endif
661 if ((ufs_rw16(ep->d_reclen, needswap) & 0x3) != 0 ||
662 ufs_rw16(ep->d_reclen, needswap) >
663 dirblksiz - (entryoffsetinblock & (dirblksiz - 1)) ||
664 ufs_rw16(ep->d_reclen, needswap) <
665 DIRSIZ(FSFMT(dp), ep, needswap) ||
666 namlen > MAXNAMLEN) {
667 /*return (1); */
668 printf("First bad, reclen=%x, DIRSIZ=%lu, namlen=%d, flags=%x "
669 "entryoffsetinblock=%d, dirblksiz = %d\n",
670 ufs_rw16(ep->d_reclen, needswap),
671 (u_long)DIRSIZ(FSFMT(dp), ep, needswap),
672 namlen, dp->v_mount->mnt_flag, entryoffsetinblock,dirblksiz);
673 goto bad;
674 }
675 if (ep->d_ino == 0)
676 return (0);
677 for (i = 0; i < namlen; i++)
678 if (ep->d_name[i] == '\0') {
679 /*return (1); */
680 printf("Second bad\n");
681 goto bad;
682 }
683 if (ep->d_name[i])
684 goto bad;
685 return (0);
686 bad:
687 return (1);
688 }
689
690 /*
691 * Construct a new directory entry after a call to namei, using the
692 * parameters that it left in the componentname argument cnp. The
693 * argument ip is the inode to which the new directory entry will refer.
694 */
695 void
696 ufs_makedirentry(ip, cnp, newdirp)
697 struct inode *ip;
698 struct componentname *cnp;
699 struct direct *newdirp;
700 {
701 #ifdef DIAGNOSTIC
702 if ((cnp->cn_flags & SAVENAME) == 0)
703 panic("makedirentry: missing name");
704 #endif
705 newdirp->d_ino = ip->i_number;
706 newdirp->d_namlen = cnp->cn_namelen;
707 memcpy(newdirp->d_name, cnp->cn_nameptr, (size_t)cnp->cn_namelen);
708 newdirp->d_name[cnp->cn_namelen] = '\0';
709 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
710 newdirp->d_type = IFTODT(ip->i_mode);
711 else {
712 newdirp->d_type = 0;
713 }
714 }
715
716 /*
717 * Write a directory entry after a call to namei, using the parameters
718 * that it left in nameidata. The argument dirp is the new directory
719 * entry contents. Dvp is a pointer to the directory to be written,
720 * which was left locked by namei. Remaining parameters (dp->i_offset,
721 * dp->i_count) indicate how the space for the new entry is to be obtained.
722 * Non-null bp indicates that a directory is being created (for the
723 * soft dependency code).
724 */
725 int
726 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
727 struct vnode *dvp;
728 struct vnode *tvp;
729 struct direct *dirp;
730 struct componentname *cnp;
731 struct buf *newdirbp;
732 {
733 struct ucred *cr;
734 struct proc *p;
735 int newentrysize;
736 struct inode *dp;
737 struct buf *bp;
738 u_int dsize;
739 struct direct *ep, *nep;
740 int error, ret, blkoff, loc, spacefree, flags;
741 char *dirbuf;
742 struct timespec ts;
743 const int needswap = UFS_MPNEEDSWAP(dvp->v_mount);
744 int dirblksiz = DIRBLKSIZ;
745 if (UFS_MPISAPPLEUFS(dvp->v_mount)) {
746 dirblksiz = APPLEUFS_DIRBLKSIZ;
747 }
748
749 error = 0;
750 cr = cnp->cn_cred;
751 p = cnp->cn_proc;
752
753 dp = VTOI(dvp);
754 newentrysize = DIRSIZ(0, dirp, 0);
755
756 if (dp->i_count == 0) {
757 /*
758 * If dp->i_count is 0, then namei could find no
759 * space in the directory. Here, dp->i_offset will
760 * be on a directory block boundary and we will write the
761 * new entry into a fresh block.
762 */
763 if (dp->i_offset & (dirblksiz - 1))
764 panic("ufs_direnter: newblk");
765 flags = B_CLRBUF;
766 if (!DOINGSOFTDEP(dvp))
767 flags |= B_SYNC;
768 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, dirblksiz,
769 cr, flags, &bp)) != 0) {
770 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
771 bdwrite(newdirbp);
772 return (error);
773 }
774 dp->i_size = dp->i_offset + dirblksiz;
775 DIP_ASSIGN(dp, size, dp->i_size);
776 dp->i_flag |= IN_CHANGE | IN_UPDATE;
777 uvm_vnp_setsize(dvp, dp->i_size);
778 dirp->d_reclen = ufs_rw16(dirblksiz, needswap);
779 dirp->d_ino = ufs_rw32(dirp->d_ino, needswap);
780 if (dvp->v_mount->mnt_maxsymlinklen <= 0) {
781 #if (BYTE_ORDER == LITTLE_ENDIAN)
782 if (needswap == 0) {
783 #else
784 if (needswap != 0) {
785 #endif
786 u_char tmp = dirp->d_namlen;
787 dirp->d_namlen = dirp->d_type;
788 dirp->d_type = tmp;
789 }
790 }
791 blkoff = dp->i_offset &
792 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
793 memcpy((caddr_t)bp->b_data + blkoff, (caddr_t)dirp,
794 newentrysize);
795 if (DOINGSOFTDEP(dvp)) {
796 /*
797 * Ensure that the entire newly allocated block is a
798 * valid directory so that future growth within the
799 * block does not have to ensure that the block is
800 * written before the inode.
801 */
802 blkoff += dirblksiz;
803 while (blkoff < bp->b_bcount) {
804 ((struct direct *)
805 (bp->b_data + blkoff))->d_reclen = dirblksiz;
806 blkoff += dirblksiz;
807 }
808 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
809 ufs_rw32(dirp->d_ino, needswap), newdirbp, 1) == 0) {
810 bdwrite(bp);
811 TIMEVAL_TO_TIMESPEC(&time, &ts);
812 return VOP_UPDATE(dvp, &ts, &ts, UPDATE_DIROP);
813 }
814 /* We have just allocated a directory block in an
815 * indirect block. Rather than tracking when it gets
816 * claimed by the inode, we simply do a VOP_FSYNC
817 * now to ensure that it is there (in case the user
818 * does a future fsync). Note that we have to unlock
819 * the inode for the entry that we just entered, as
820 * the VOP_FSYNC may need to lock other inodes which
821 * can lead to deadlock if we also hold a lock on
822 * the newly entered node.
823 */
824 error = VOP_BWRITE(bp);
825 if (error != 0)
826 return (error);
827 if (tvp != NULL)
828 VOP_UNLOCK(tvp, 0);
829 error = VOP_FSYNC(dvp, p->p_ucred, FSYNC_WAIT, 0, 0, p);
830 if (tvp != 0)
831 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
832 return (error);
833 } else {
834 error = VOP_BWRITE(bp);
835 }
836 TIMEVAL_TO_TIMESPEC(&time, &ts);
837 ret = VOP_UPDATE(dvp, &ts, &ts, UPDATE_DIROP);
838 if (error == 0)
839 return (ret);
840 return (error);
841 }
842
843 /*
844 * If dp->i_count is non-zero, then namei found space for the new
845 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
846 * in the directory. To use this space, we may have to compact
847 * the entries located there, by copying them together towards the
848 * beginning of the block, leaving the free space in one usable
849 * chunk at the end.
850 */
851
852 /*
853 * Increase size of directory if entry eats into new space.
854 * This should never push the size past a new multiple of
855 * DIRBLKSIZE.
856 *
857 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
858 */
859 if (dp->i_offset + dp->i_count > dp->i_size) {
860 dp->i_size = dp->i_offset + dp->i_count;
861 DIP_ASSIGN(dp, size, dp->i_size);
862 dp->i_flag |= IN_CHANGE | IN_UPDATE;
863 }
864 /*
865 * Get the block containing the space for the new directory entry.
866 */
867 error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
868 if (error) {
869 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
870 bdwrite(newdirbp);
871 return (error);
872 }
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 space.
878 */
879 ep = (struct direct *)dirbuf;
880 dsize = DIRSIZ(FSFMT(dvp), ep, needswap);
881 spacefree = ufs_rw16(ep->d_reclen, needswap) - dsize;
882 for (loc = ufs_rw16(ep->d_reclen, needswap); loc < dp->i_count; ) {
883 nep = (struct direct *)(dirbuf + loc);
884 if (ep->d_ino) {
885 /* trim the existing slot */
886 ep->d_reclen = ufs_rw16(dsize, needswap);
887 ep = (struct direct *)((char *)ep + dsize);
888 } else {
889 /* overwrite; nothing there; header is ours */
890 spacefree += dsize;
891 }
892 dsize = DIRSIZ(FSFMT(dvp), nep, needswap);
893 spacefree += ufs_rw16(nep->d_reclen, needswap) - dsize;
894 loc += ufs_rw16(nep->d_reclen, needswap);
895 if (DOINGSOFTDEP(dvp))
896 softdep_change_directoryentry_offset(dp, dirbuf,
897 (caddr_t)nep, (caddr_t)ep, dsize);
898 else
899 memcpy((caddr_t)ep, (caddr_t)nep, dsize);
900 }
901 /*
902 * Update the pointer fields in the previous entry (if any),
903 * copy in the new entry, and write out the block.
904 */
905 if (ep->d_ino == 0 ||
906 (ufs_rw32(ep->d_ino, needswap) == WINO &&
907 memcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
908 if (spacefree + dsize < newentrysize)
909 panic("ufs_direnter: compact1");
910 dirp->d_reclen = spacefree + dsize;
911 } else {
912 if (spacefree < newentrysize)
913 panic("ufs_direnter: compact2");
914 dirp->d_reclen = spacefree;
915 ep->d_reclen = ufs_rw16(dsize, needswap);
916 ep = (struct direct *)((char *)ep + dsize);
917 }
918 dirp->d_reclen = ufs_rw16(dirp->d_reclen, needswap);
919 dirp->d_ino = ufs_rw32(dirp->d_ino, needswap);
920 if (dvp->v_mount->mnt_maxsymlinklen <= 0) {
921 #if (BYTE_ORDER == LITTLE_ENDIAN)
922 if (needswap == 0) {
923 #else
924 if (needswap != 0) {
925 #endif
926 u_char tmp = dirp->d_namlen;
927 dirp->d_namlen = dirp->d_type;
928 dirp->d_type = tmp;
929 }
930 }
931 memcpy((caddr_t)ep, (caddr_t)dirp, (u_int)newentrysize);
932 if (DOINGSOFTDEP(dvp)) {
933 softdep_setup_directory_add(bp, dp,
934 dp->i_offset + (caddr_t)ep - dirbuf,
935 ufs_rw32(dirp->d_ino, needswap), newdirbp, 0);
936 bdwrite(bp);
937 } else {
938 error = VOP_BWRITE(bp);
939 }
940 dp->i_flag |= IN_CHANGE | IN_UPDATE;
941 /*
942 * If all went well, and the directory can be shortened, proceed
943 * with the truncation. Note that we have to unlock the inode for
944 * the entry that we just entered, as the truncation may need to
945 * lock other inodes which can lead to deadlock if we also hold a
946 * lock on the newly entered node.
947 */
948 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
949 if (DOINGSOFTDEP(dvp) && (tvp != NULL))
950 VOP_UNLOCK(tvp, 0);
951 (void) VOP_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cr, p);
952 if (DOINGSOFTDEP(dvp) && (tvp != NULL))
953 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
954 }
955 return (error);
956 }
957
958 /*
959 * Remove a directory entry after a call to namei, using
960 * the parameters which it left in nameidata. The entry
961 * dp->i_offset contains the offset into the directory of the
962 * entry to be eliminated. The dp->i_count field contains the
963 * size of the previous record in the directory. If this
964 * is 0, the first entry is being deleted, so we need only
965 * zero the inode number to mark the entry as free. If the
966 * entry is not the first in the directory, we must reclaim
967 * the space of the now empty record by adding the record size
968 * to the size of the previous entry.
969 */
970 int
971 ufs_dirremove(dvp, ip, flags, isrmdir)
972 struct vnode *dvp;
973 struct inode *ip;
974 int flags;
975 int isrmdir;
976 {
977 struct inode *dp;
978 struct direct *ep;
979 struct buf *bp;
980 int error;
981 #ifdef FFS_EI
982 const int needswap = UFS_MPNEEDSWAP(dvp->v_mount);
983 #endif
984
985 dp = VTOI(dvp);
986
987 if (flags & DOWHITEOUT) {
988 /*
989 * Whiteout entry: set d_ino to WINO.
990 */
991 error = VOP_BLKATOFF(dvp, (off_t)dp->i_offset, (void *)&ep,
992 &bp);
993 if (error)
994 return (error);
995 ep->d_ino = ufs_rw32(WINO, needswap);
996 ep->d_type = DT_WHT;
997 goto out;
998 }
999
1000 if ((error = VOP_BLKATOFF(dvp,
1001 (off_t)(dp->i_offset - dp->i_count), (void *)&ep, &bp)) != 0)
1002 return (error);
1003
1004 if (dp->i_count == 0) {
1005 /*
1006 * First entry in block: set d_ino to zero.
1007 */
1008 ep->d_ino = 0;
1009 } else {
1010 /*
1011 * Collapse new free space into previous entry.
1012 */
1013 ep->d_reclen =
1014 ufs_rw16(ufs_rw16(ep->d_reclen, needswap) + dp->i_reclen,
1015 needswap);
1016 }
1017 out:
1018 if (DOINGSOFTDEP(dvp)) {
1019 if (ip) {
1020 ip->i_ffs_effnlink--;
1021 softdep_change_linkcnt(ip);
1022 softdep_setup_remove(bp, dp, ip, isrmdir);
1023 }
1024 bdwrite(bp);
1025 } else {
1026 if (ip) {
1027 ip->i_ffs_effnlink--;
1028 ip->i_nlink--;
1029 DIP_ASSIGN(ip, nlink, ip->i_nlink);
1030 ip->i_flag |= IN_CHANGE;
1031 }
1032 error = VOP_BWRITE(bp);
1033 }
1034 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1035 return (error);
1036 }
1037
1038 /*
1039 * Rewrite an existing directory entry to point at the inode
1040 * supplied. The parameters describing the directory entry are
1041 * set up by a call to namei.
1042 */
1043 int
1044 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir, iflags)
1045 struct inode *dp, *oip;
1046 ino_t newinum;
1047 int newtype;
1048 int isrmdir;
1049 int iflags;
1050 {
1051 struct buf *bp;
1052 struct direct *ep;
1053 struct vnode *vdp = ITOV(dp);
1054 int error;
1055
1056 error = VOP_BLKATOFF(vdp, (off_t)dp->i_offset, (void *)&ep, &bp);
1057 if (error)
1058 return (error);
1059 ep->d_ino = ufs_rw32(newinum, UFS_MPNEEDSWAP(vdp->v_mount));
1060 if (vdp->v_mount->mnt_maxsymlinklen > 0)
1061 ep->d_type = newtype;
1062 oip->i_ffs_effnlink--;
1063 if (DOINGSOFTDEP(vdp)) {
1064 softdep_change_linkcnt(oip);
1065 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1066 bdwrite(bp);
1067 } else {
1068 oip->i_nlink--;
1069 DIP_ASSIGN(oip, nlink, oip->i_nlink);
1070 oip->i_flag |= IN_CHANGE;
1071 error = VOP_BWRITE(bp);
1072 }
1073 dp->i_flag |= iflags;
1074 return (error);
1075 }
1076
1077 /*
1078 * Check if a directory is empty or not.
1079 * Inode supplied must be locked.
1080 *
1081 * Using a struct dirtemplate here is not precisely
1082 * what we want, but better than using a struct direct.
1083 *
1084 * NB: does not handle corrupted directories.
1085 */
1086 int
1087 ufs_dirempty(ip, parentino, cred)
1088 struct inode *ip;
1089 ino_t parentino;
1090 struct ucred *cred;
1091 {
1092 doff_t off;
1093 struct dirtemplate dbuf;
1094 struct direct *dp = (struct direct *)&dbuf;
1095 int error, namlen;
1096 size_t count;
1097 const int needswap = UFS_IPNEEDSWAP(ip);
1098 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1099
1100 for (off = 0; off < ip->i_size;
1101 off += ufs_rw16(dp->d_reclen, needswap)) {
1102 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1103 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
1104 /*
1105 * Since we read MINDIRSIZ, residual must
1106 * be 0 unless we're at end of file.
1107 */
1108 if (error || count != 0)
1109 return (0);
1110 /* avoid infinite loops */
1111 if (dp->d_reclen == 0)
1112 return (0);
1113 /* skip empty entries */
1114 if (dp->d_ino == 0 || ufs_rw32(dp->d_ino, needswap) == WINO)
1115 continue;
1116 /* accept only "." and ".." */
1117 #if (BYTE_ORDER == LITTLE_ENDIAN)
1118 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0 || needswap != 0)
1119 namlen = dp->d_namlen;
1120 else
1121 namlen = dp->d_type;
1122 #else
1123 if (ITOV(ip)->v_mount->mnt_maxsymlinklen <= 0 && needswap != 0)
1124 namlen = dp->d_type;
1125 else
1126 namlen = dp->d_namlen;
1127 #endif
1128 if (namlen > 2)
1129 return (0);
1130 if (dp->d_name[0] != '.')
1131 return (0);
1132 /*
1133 * At this point namlen must be 1 or 2.
1134 * 1 implies ".", 2 implies ".." if second
1135 * char is also "."
1136 */
1137 if (namlen == 1 &&
1138 ufs_rw32(dp->d_ino, needswap) == ip->i_number)
1139 continue;
1140 if (dp->d_name[1] == '.' &&
1141 ufs_rw32(dp->d_ino, needswap) == parentino)
1142 continue;
1143 return (0);
1144 }
1145 return (1);
1146 }
1147
1148 /*
1149 * Check if source directory is in the path of the target directory.
1150 * Target is supplied locked, source is unlocked.
1151 * The target is always vput before returning.
1152 */
1153 int
1154 ufs_checkpath(source, target, cred)
1155 struct inode *source, *target;
1156 struct ucred *cred;
1157 {
1158 struct vnode *vp = ITOV(target);
1159 int error, rootino, namlen;
1160 struct dirtemplate dirbuf;
1161 const int needswap = UFS_MPNEEDSWAP(vp->v_mount);
1162
1163 vp = ITOV(target);
1164 if (target->i_number == source->i_number) {
1165 error = EEXIST;
1166 goto out;
1167 }
1168 rootino = ROOTINO;
1169 error = 0;
1170 if (target->i_number == rootino)
1171 goto out;
1172
1173 for (;;) {
1174 if (vp->v_type != VDIR) {
1175 error = ENOTDIR;
1176 break;
1177 }
1178 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1179 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1180 IO_NODELOCKED, cred, NULL, (struct proc *)0);
1181 if (error != 0)
1182 break;
1183 #if (BYTE_ORDER == LITTLE_ENDIAN)
1184 if (vp->v_mount->mnt_maxsymlinklen > 0 ||
1185 needswap != 0)
1186 namlen = dirbuf.dotdot_namlen;
1187 else
1188 namlen = dirbuf.dotdot_type;
1189 #else
1190 if (vp->v_mount->mnt_maxsymlinklen == 0 &&
1191 needswap != 0)
1192 namlen = dirbuf.dotdot_type;
1193 else
1194 namlen = dirbuf.dotdot_namlen;
1195 #endif
1196 if (namlen != 2 ||
1197 dirbuf.dotdot_name[0] != '.' ||
1198 dirbuf.dotdot_name[1] != '.') {
1199 error = ENOTDIR;
1200 break;
1201 }
1202 if (ufs_rw32(dirbuf.dotdot_ino, needswap) == source->i_number) {
1203 error = EINVAL;
1204 break;
1205 }
1206 if (ufs_rw32(dirbuf.dotdot_ino, needswap) == rootino)
1207 break;
1208 vput(vp);
1209 error = VFS_VGET(vp->v_mount,
1210 ufs_rw32(dirbuf.dotdot_ino, needswap), &vp);
1211 if (error) {
1212 vp = NULL;
1213 break;
1214 }
1215 }
1216
1217 out:
1218 if (error == ENOTDIR)
1219 printf("checkpath: .. not a directory\n");
1220 if (vp != NULL)
1221 vput(vp);
1222 return (error);
1223 }
Cache object: 6b72ca3c5b4db0aaa4746ba5144aa911
|