1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/6.2/sys/ufs/ufs/ufs_lookup.c 162034 2006-09-05 13:20:41Z daichi $");
39
40 #include "opt_ffs_broken_fixme.h"
41 #include "opt_ufs.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/namei.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/proc.h>
50 #include <sys/stat.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/sysctl.h>
54
55 #include <vm/vm.h>
56 #include <vm/vm_extern.h>
57
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/inode.h>
61 #include <ufs/ufs/dir.h>
62 #ifdef UFS_DIRHASH
63 #include <ufs/ufs/dirhash.h>
64 #endif
65 #include <ufs/ufs/ufsmount.h>
66 #include <ufs/ufs/ufs_extern.h>
67
68 #ifdef DIAGNOSTIC
69 static int dirchk = 1;
70 #else
71 static int dirchk = 0;
72 #endif
73
74 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
75
76 /* true if old FS format...*/
77 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
78
79 /*
80 * Convert a component of a pathname into a pointer to a locked inode.
81 * This is a very central and rather complicated routine.
82 * If the filesystem is not maintained in a strict tree hierarchy,
83 * this can result in a deadlock situation (see comments in code below).
84 *
85 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
86 * on whether the name is to be looked up, created, renamed, or deleted.
87 * When CREATE, RENAME, or DELETE is specified, information usable in
88 * creating, renaming, or deleting a directory entry may be calculated.
89 * If flag has LOCKPARENT or'ed into it and the target of the pathname
90 * exists, lookup returns both the target and its parent directory locked.
91 * When creating or renaming and LOCKPARENT is specified, the target may
92 * not be ".". When deleting and LOCKPARENT is specified, the target may
93 * be "."., but the caller must check to ensure it does an vrele and vput
94 * instead of two vputs.
95 *
96 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
97 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
98 * method.
99 *
100 * vfs_cache_lookup() performs the following for us:
101 * check that it is a directory
102 * check accessibility of directory
103 * check for modification attempts on read-only mounts
104 * if name found in cache
105 * if at end of path and deleting or creating
106 * drop it
107 * else
108 * return name.
109 * return VOP_CACHEDLOOKUP()
110 *
111 * Overall outline of ufs_lookup:
112 *
113 * search for name in directory, to found or notfound
114 * notfound:
115 * if creating, return locked directory, leaving info on available slots
116 * else return error
117 * found:
118 * if at end of path and deleting, return information to allow delete
119 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
120 * inode and return info to allow rewrite
121 * if not at end, add name to cache; if at end and neither creating
122 * nor deleting, add name to cache
123 */
124 int
125 ufs_lookup(ap)
126 struct vop_cachedlookup_args /* {
127 struct vnode *a_dvp;
128 struct vnode **a_vpp;
129 struct componentname *a_cnp;
130 } */ *ap;
131 {
132 struct vnode *vdp; /* vnode for directory being searched */
133 struct inode *dp; /* inode for directory being searched */
134 struct buf *bp; /* a buffer of directory entries */
135 struct direct *ep; /* the current directory entry */
136 int entryoffsetinblock; /* offset of ep in bp's buffer */
137 enum {NONE, COMPACT, FOUND} slotstatus;
138 doff_t slotoffset; /* offset of area with free space */
139 int slotsize; /* size of area at slotoffset */
140 int slotfreespace; /* amount of space free in slot */
141 int slotneeded; /* size of the entry we're seeking */
142 int numdirpasses; /* strategy for directory search */
143 doff_t endsearch; /* offset to end directory search */
144 doff_t prevoff; /* prev entry dp->i_offset */
145 struct vnode *pdp; /* saved dp during symlink work */
146 struct vnode *tdp; /* returned by VFS_VGET */
147 doff_t enduseful; /* pointer past last used dir slot */
148 u_long bmask; /* block offset mask */
149 int namlen, error;
150 struct vnode **vpp = ap->a_vpp;
151 struct componentname *cnp = ap->a_cnp;
152 struct ucred *cred = cnp->cn_cred;
153 int flags = cnp->cn_flags;
154 int nameiop = cnp->cn_nameiop;
155 struct thread *td = cnp->cn_thread;
156 ino_t saved_ino;
157
158 bp = NULL;
159 slotoffset = -1;
160 /*
161 * XXX there was a soft-update diff about this I couldn't merge.
162 * I think this was the equiv.
163 */
164 *vpp = NULL;
165
166 vdp = ap->a_dvp;
167 dp = VTOI(vdp);
168 /*
169 * We now have a segment name to search for, and a directory to search.
170 *
171 * Suppress search for slots unless creating
172 * file and at end of pathname, in which case
173 * we watch for a place to put the new file in
174 * case it doesn't already exist.
175 */
176 slotstatus = FOUND;
177 slotfreespace = slotsize = slotneeded = 0;
178 if ((nameiop == CREATE || nameiop == RENAME) &&
179 (flags & ISLASTCN)) {
180 slotstatus = NONE;
181 slotneeded = DIRECTSIZ(cnp->cn_namelen);
182 }
183 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
184
185 #ifdef UFS_DIRHASH
186 /*
187 * Use dirhash for fast operations on large directories. The logic
188 * to determine whether to hash the directory is contained within
189 * ufsdirhash_build(); a zero return means that it decided to hash
190 * this directory and it successfully built up the hash table.
191 */
192 if (ufsdirhash_build(dp) == 0) {
193 /* Look for a free slot if needed. */
194 enduseful = dp->i_size;
195 if (slotstatus != FOUND) {
196 slotoffset = ufsdirhash_findfree(dp, slotneeded,
197 &slotsize);
198 if (slotoffset >= 0) {
199 slotstatus = COMPACT;
200 enduseful = ufsdirhash_enduseful(dp);
201 if (enduseful < 0)
202 enduseful = dp->i_size;
203 }
204 }
205 /* Look up the component. */
206 numdirpasses = 1;
207 entryoffsetinblock = 0; /* silence compiler warning */
208 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
209 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
210 case 0:
211 ep = (struct direct *)((char *)bp->b_data +
212 (dp->i_offset & bmask));
213 goto foundentry;
214 case ENOENT:
215 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
216 goto notfound;
217 default:
218 /* Something failed; just do a linear search. */
219 break;
220 }
221 }
222 #endif /* UFS_DIRHASH */
223 /*
224 * If there is cached information on a previous search of
225 * this directory, pick up where we last left off.
226 * We cache only lookups as these are the most common
227 * and have the greatest payoff. Caching CREATE has little
228 * benefit as it usually must search the entire directory
229 * to determine that the entry does not exist. Caching the
230 * location of the last DELETE or RENAME has not reduced
231 * profiling time and hence has been removed in the interest
232 * of simplicity.
233 */
234 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
235 dp->i_diroff >= dp->i_size) {
236 entryoffsetinblock = 0;
237 dp->i_offset = 0;
238 numdirpasses = 1;
239 } else {
240 dp->i_offset = dp->i_diroff;
241 if ((entryoffsetinblock = dp->i_offset & bmask) &&
242 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
243 return (error);
244 numdirpasses = 2;
245 nchstats.ncs_2passes++;
246 }
247 prevoff = dp->i_offset;
248 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
249 enduseful = 0;
250
251 searchloop:
252 while (dp->i_offset < endsearch) {
253 /*
254 * If necessary, get the next directory block.
255 */
256 if ((dp->i_offset & bmask) == 0) {
257 if (bp != NULL)
258 brelse(bp);
259 error =
260 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
261 if (error)
262 return (error);
263 entryoffsetinblock = 0;
264 }
265 /*
266 * If still looking for a slot, and at a DIRBLKSIZE
267 * boundary, have to start looking for free space again.
268 */
269 if (slotstatus == NONE &&
270 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
271 slotoffset = -1;
272 slotfreespace = 0;
273 }
274 /*
275 * Get pointer to next entry.
276 * Full validation checks are slow, so we only check
277 * enough to insure forward progress through the
278 * directory. Complete checks can be run by patching
279 * "dirchk" to be true.
280 */
281 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
282 if (ep->d_reclen == 0 || ep->d_reclen >
283 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
284 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
285 int i;
286
287 ufs_dirbad(dp, dp->i_offset, "mangled entry");
288 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
289 dp->i_offset += i;
290 entryoffsetinblock += i;
291 continue;
292 }
293
294 /*
295 * If an appropriate sized slot has not yet been found,
296 * check to see if one is available. Also accumulate space
297 * in the current block so that we can determine if
298 * compaction is viable.
299 */
300 if (slotstatus != FOUND) {
301 int size = ep->d_reclen;
302
303 if (ep->d_ino != 0)
304 size -= DIRSIZ(OFSFMT(vdp), ep);
305 if (size > 0) {
306 if (size >= slotneeded) {
307 slotstatus = FOUND;
308 slotoffset = dp->i_offset;
309 slotsize = ep->d_reclen;
310 } else if (slotstatus == NONE) {
311 slotfreespace += size;
312 if (slotoffset == -1)
313 slotoffset = dp->i_offset;
314 if (slotfreespace >= slotneeded) {
315 slotstatus = COMPACT;
316 slotsize = dp->i_offset +
317 ep->d_reclen - slotoffset;
318 }
319 }
320 }
321 }
322
323 /*
324 * Check for a name match.
325 */
326 if (ep->d_ino) {
327 # if (BYTE_ORDER == LITTLE_ENDIAN)
328 if (OFSFMT(vdp))
329 namlen = ep->d_type;
330 else
331 namlen = ep->d_namlen;
332 # else
333 namlen = ep->d_namlen;
334 # endif
335 if (namlen == cnp->cn_namelen &&
336 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
337 !bcmp(cnp->cn_nameptr, ep->d_name,
338 (unsigned)namlen)) {
339 #ifdef UFS_DIRHASH
340 foundentry:
341 #endif
342 /*
343 * Save directory entry's inode number and
344 * reclen in ndp->ni_ufs area, and release
345 * directory buffer.
346 */
347 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
348 ep->d_type == DT_WHT) {
349 slotstatus = FOUND;
350 slotoffset = dp->i_offset;
351 slotsize = ep->d_reclen;
352 dp->i_reclen = slotsize;
353 enduseful = dp->i_size;
354 ap->a_cnp->cn_flags |= ISWHITEOUT;
355 numdirpasses--;
356 goto notfound;
357 }
358 dp->i_ino = ep->d_ino;
359 dp->i_reclen = ep->d_reclen;
360 goto found;
361 }
362 }
363 prevoff = dp->i_offset;
364 dp->i_offset += ep->d_reclen;
365 entryoffsetinblock += ep->d_reclen;
366 if (ep->d_ino)
367 enduseful = dp->i_offset;
368 }
369 notfound:
370 /*
371 * If we started in the middle of the directory and failed
372 * to find our target, we must check the beginning as well.
373 */
374 if (numdirpasses == 2) {
375 numdirpasses--;
376 dp->i_offset = 0;
377 endsearch = dp->i_diroff;
378 goto searchloop;
379 }
380 if (bp != NULL)
381 brelse(bp);
382 /*
383 * If creating, and at end of pathname and current
384 * directory has not been removed, then can consider
385 * allowing file to be created.
386 */
387 if ((nameiop == CREATE || nameiop == RENAME ||
388 (nameiop == DELETE &&
389 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
390 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
391 (flags & ISLASTCN) && dp->i_effnlink != 0) {
392 /*
393 * Access for write is interpreted as allowing
394 * creation of files in the directory.
395 */
396 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
397 if (error)
398 return (error);
399 /*
400 * Return an indication of where the new directory
401 * entry should be put. If we didn't find a slot,
402 * then set dp->i_count to 0 indicating
403 * that the new slot belongs at the end of the
404 * directory. If we found a slot, then the new entry
405 * can be put in the range from dp->i_offset to
406 * dp->i_offset + dp->i_count.
407 */
408 if (slotstatus == NONE) {
409 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
410 dp->i_count = 0;
411 enduseful = dp->i_offset;
412 } else if (nameiop == DELETE) {
413 dp->i_offset = slotoffset;
414 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
415 dp->i_count = 0;
416 else
417 dp->i_count = dp->i_offset - prevoff;
418 } else {
419 dp->i_offset = slotoffset;
420 dp->i_count = slotsize;
421 if (enduseful < slotoffset + slotsize)
422 enduseful = slotoffset + slotsize;
423 }
424 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
425 /*
426 * We return with the directory locked, so that
427 * the parameters we set up above will still be
428 * valid if we actually decide to do a direnter().
429 * We return ni_vp == NULL to indicate that the entry
430 * does not currently exist; we leave a pointer to
431 * the (locked) directory inode in ndp->ni_dvp.
432 * The pathname buffer is saved so that the name
433 * can be obtained later.
434 *
435 * NB - if the directory is unlocked, then this
436 * information cannot be used.
437 */
438 cnp->cn_flags |= SAVENAME;
439 return (EJUSTRETURN);
440 }
441 /*
442 * Insert name into cache (as non-existent) if appropriate.
443 */
444 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
445 cache_enter(vdp, *vpp, cnp);
446 return (ENOENT);
447
448 found:
449 if (numdirpasses == 2)
450 nchstats.ncs_pass2++;
451 /*
452 * Check that directory length properly reflects presence
453 * of this entry.
454 */
455 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
456 ufs_dirbad(dp, dp->i_offset, "i_size too small");
457 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
458 DIP_SET(dp, i_size, dp->i_size);
459 dp->i_flag |= IN_CHANGE | IN_UPDATE;
460 }
461 brelse(bp);
462
463 /*
464 * Found component in pathname.
465 * If the final component of path name, save information
466 * in the cache as to where the entry was found.
467 */
468 if ((flags & ISLASTCN) && nameiop == LOOKUP)
469 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
470
471 /*
472 * If deleting, and at end of pathname, return
473 * parameters which can be used to remove file.
474 */
475 if (nameiop == DELETE && (flags & ISLASTCN)) {
476 /*
477 * Write access to directory required to delete files.
478 */
479 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
480 if (error)
481 return (error);
482 /*
483 * Return pointer to current entry in dp->i_offset,
484 * and distance past previous entry (if there
485 * is a previous entry in this block) in dp->i_count.
486 * Save directory inode pointer in ndp->ni_dvp for dirremove().
487 */
488 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
489 dp->i_count = 0;
490 else
491 dp->i_count = dp->i_offset - prevoff;
492 if (dp->i_number == dp->i_ino) {
493 VREF(vdp);
494 *vpp = vdp;
495 return (0);
496 }
497 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino,
498 LK_EXCLUSIVE, &tdp)) != 0)
499 return (error);
500 /*
501 * If directory is "sticky", then user must own
502 * the directory, or the file in it, else she
503 * may not delete it (unless she's root). This
504 * implements append-only directories.
505 */
506 if ((dp->i_mode & ISVTX) &&
507 VOP_ACCESS(vdp, VADMIN, cred, cnp->cn_thread) &&
508 VOP_ACCESS(tdp, VADMIN, cred, cnp->cn_thread)) {
509 vput(tdp);
510 return (EPERM);
511 }
512 *vpp = tdp;
513 return (0);
514 }
515
516 /*
517 * If rewriting (RENAME), return the inode and the
518 * information required to rewrite the present directory
519 * Must get inode of directory entry to verify it's a
520 * regular file, or empty directory.
521 */
522 if (nameiop == RENAME && (flags & ISLASTCN)) {
523 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)))
524 return (error);
525 /*
526 * Careful about locking second inode.
527 * This can only occur if the target is ".".
528 */
529 if (dp->i_number == dp->i_ino)
530 return (EISDIR);
531 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino,
532 LK_EXCLUSIVE, &tdp)) != 0)
533 return (error);
534 *vpp = tdp;
535 cnp->cn_flags |= SAVENAME;
536 return (0);
537 }
538
539 /*
540 * Step through the translation in the name. We do not `vput' the
541 * directory because we may need it again if a symbolic link
542 * is relative to the current directory. Instead we save it
543 * unlocked as "pdp". We must get the target inode before unlocking
544 * the directory to insure that the inode will not be removed
545 * before we get it. We prevent deadlock by always fetching
546 * inodes from the root, moving down the directory tree. Thus
547 * when following backward pointers ".." we must unlock the
548 * parent directory before getting the requested directory.
549 * There is a potential race condition here if both the current
550 * and parent directories are removed before the VFS_VGET for the
551 * inode associated with ".." returns. We hope that this occurs
552 * infrequently since we cannot avoid this race condition without
553 * implementing a sophisticated deadlock detection algorithm.
554 * Note also that this simple deadlock detection scheme will not
555 * work if the filesystem has any hard links other than ".."
556 * that point backwards in the directory structure.
557 */
558 pdp = vdp;
559 if (flags & ISDOTDOT) {
560 saved_ino = dp->i_ino;
561 VOP_UNLOCK(pdp, 0, td); /* race to get the inode */
562 error = VFS_VGET(pdp->v_mount, saved_ino,
563 cnp->cn_lkflags, &tdp);
564 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, td);
565 if (error)
566 return (error);
567 *vpp = tdp;
568 } else if (dp->i_number == dp->i_ino) {
569 VREF(vdp); /* we want ourself, ie "." */
570 *vpp = vdp;
571 } else {
572 error = VFS_VGET(pdp->v_mount, dp->i_ino,
573 cnp->cn_lkflags, &tdp);
574 if (error)
575 return (error);
576 *vpp = tdp;
577 }
578
579 /*
580 * Insert name into cache if appropriate.
581 */
582 if (cnp->cn_flags & MAKEENTRY)
583 cache_enter(vdp, *vpp, cnp);
584 return (0);
585 }
586
587 void
588 ufs_dirbad(ip, offset, how)
589 struct inode *ip;
590 doff_t offset;
591 char *how;
592 {
593 struct mount *mp;
594
595 mp = ITOV(ip)->v_mount;
596 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
597 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
598 if ((mp->mnt_flag & MNT_RDONLY) == 0)
599 panic("ufs_dirbad: bad dir");
600 }
601
602 /*
603 * Do consistency checking on a directory entry:
604 * record length must be multiple of 4
605 * entry must fit in rest of its DIRBLKSIZ block
606 * record must be large enough to contain entry
607 * name is not longer than MAXNAMLEN
608 * name must be as long as advertised, and null terminated
609 */
610 int
611 ufs_dirbadentry(dp, ep, entryoffsetinblock)
612 struct vnode *dp;
613 struct direct *ep;
614 int entryoffsetinblock;
615 {
616 int i, namlen;
617
618 # if (BYTE_ORDER == LITTLE_ENDIAN)
619 if (OFSFMT(dp))
620 namlen = ep->d_type;
621 else
622 namlen = ep->d_namlen;
623 # else
624 namlen = ep->d_namlen;
625 # endif
626 if ((ep->d_reclen & 0x3) != 0 ||
627 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
628 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
629 /*return (1); */
630 printf("First bad\n");
631 goto bad;
632 }
633 if (ep->d_ino == 0)
634 return (0);
635 for (i = 0; i < namlen; i++)
636 if (ep->d_name[i] == '\0') {
637 /*return (1); */
638 printf("Second bad\n");
639 goto bad;
640 }
641 if (ep->d_name[i])
642 goto bad;
643 return (0);
644 bad:
645 return (1);
646 }
647
648 /*
649 * Construct a new directory entry after a call to namei, using the
650 * parameters that it left in the componentname argument cnp. The
651 * argument ip is the inode to which the new directory entry will refer.
652 */
653 void
654 ufs_makedirentry(ip, cnp, newdirp)
655 struct inode *ip;
656 struct componentname *cnp;
657 struct direct *newdirp;
658 {
659
660 #ifdef DIAGNOSTIC
661 if ((cnp->cn_flags & SAVENAME) == 0)
662 panic("ufs_makedirentry: missing name");
663 #endif
664 newdirp->d_ino = ip->i_number;
665 newdirp->d_namlen = cnp->cn_namelen;
666 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
667 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
668 newdirp->d_type = IFTODT(ip->i_mode);
669 else {
670 newdirp->d_type = 0;
671 # if (BYTE_ORDER == LITTLE_ENDIAN)
672 { u_char tmp = newdirp->d_namlen;
673 newdirp->d_namlen = newdirp->d_type;
674 newdirp->d_type = tmp; }
675 # endif
676 }
677 }
678
679 /*
680 * Write a directory entry after a call to namei, using the parameters
681 * that it left in nameidata. The argument dirp is the new directory
682 * entry contents. Dvp is a pointer to the directory to be written,
683 * which was left locked by namei. Remaining parameters (dp->i_offset,
684 * dp->i_count) indicate how the space for the new entry is to be obtained.
685 * Non-null bp indicates that a directory is being created (for the
686 * soft dependency code).
687 */
688 int
689 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
690 struct vnode *dvp;
691 struct vnode *tvp;
692 struct direct *dirp;
693 struct componentname *cnp;
694 struct buf *newdirbp;
695 {
696 struct ucred *cr;
697 struct thread *td;
698 int newentrysize;
699 struct inode *dp;
700 struct buf *bp;
701 u_int dsize;
702 struct direct *ep, *nep;
703 int error, ret, blkoff, loc, spacefree, flags, namlen;
704 char *dirbuf;
705
706 td = curthread; /* XXX */
707 cr = td->td_ucred;
708
709 dp = VTOI(dvp);
710 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
711
712 if (dp->i_count == 0) {
713 /*
714 * If dp->i_count is 0, then namei could find no
715 * space in the directory. Here, dp->i_offset will
716 * be on a directory block boundary and we will write the
717 * new entry into a fresh block.
718 */
719 if (dp->i_offset & (DIRBLKSIZ - 1))
720 panic("ufs_direnter: newblk");
721 flags = BA_CLRBUF;
722 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
723 flags |= IO_SYNC;
724 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
725 cr, flags, &bp)) != 0) {
726 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
727 bdwrite(newdirbp);
728 return (error);
729 }
730 dp->i_size = dp->i_offset + DIRBLKSIZ;
731 DIP_SET(dp, i_size, dp->i_size);
732 dp->i_flag |= IN_CHANGE | IN_UPDATE;
733 vnode_pager_setsize(dvp, (u_long)dp->i_size);
734 dirp->d_reclen = DIRBLKSIZ;
735 blkoff = dp->i_offset &
736 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
737 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
738 #ifdef UFS_DIRHASH
739 if (dp->i_dirhash != NULL) {
740 ufsdirhash_newblk(dp, dp->i_offset);
741 ufsdirhash_add(dp, dirp, dp->i_offset);
742 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
743 dp->i_offset);
744 }
745 #endif
746 if (DOINGSOFTDEP(dvp)) {
747 /*
748 * Ensure that the entire newly allocated block is a
749 * valid directory so that future growth within the
750 * block does not have to ensure that the block is
751 * written before the inode.
752 */
753 blkoff += DIRBLKSIZ;
754 while (blkoff < bp->b_bcount) {
755 ((struct direct *)
756 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
757 blkoff += DIRBLKSIZ;
758 }
759 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
760 dirp->d_ino, newdirbp, 1) == 0) {
761 bdwrite(bp);
762 return (UFS_UPDATE(dvp, 0));
763 }
764 /* We have just allocated a directory block in an
765 * indirect block. Rather than tracking when it gets
766 * claimed by the inode, we simply do a VOP_FSYNC
767 * now to ensure that it is there (in case the user
768 * does a future fsync). Note that we have to unlock
769 * the inode for the entry that we just entered, as
770 * the VOP_FSYNC may need to lock other inodes which
771 * can lead to deadlock if we also hold a lock on
772 * the newly entered node.
773 */
774 if ((error = bwrite(bp)))
775 return (error);
776 if (tvp != NULL)
777 VOP_UNLOCK(tvp, 0, td);
778 error = VOP_FSYNC(dvp, MNT_WAIT, td);
779 if (tvp != NULL)
780 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
781 return (error);
782 }
783 if (DOINGASYNC(dvp)) {
784 bdwrite(bp);
785 return (UFS_UPDATE(dvp, 0));
786 }
787 error = bwrite(bp);
788 ret = UFS_UPDATE(dvp, 1);
789 if (error == 0)
790 return (ret);
791 return (error);
792 }
793
794 /*
795 * If dp->i_count is non-zero, then namei found space for the new
796 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
797 * in the directory. To use this space, we may have to compact
798 * the entries located there, by copying them together towards the
799 * beginning of the block, leaving the free space in one usable
800 * chunk at the end.
801 */
802
803 /*
804 * Increase size of directory if entry eats into new space.
805 * This should never push the size past a new multiple of
806 * DIRBLKSIZE.
807 *
808 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
809 */
810 if (dp->i_offset + dp->i_count > dp->i_size) {
811 dp->i_size = dp->i_offset + dp->i_count;
812 DIP_SET(dp, i_size, dp->i_size);
813 }
814 /*
815 * Get the block containing the space for the new directory entry.
816 */
817 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
818 if (error) {
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(dvp), 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 bcopy() 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 bcopy it.
849 */
850 spacefree += nep->d_reclen;
851 ep->d_ino = 0;
852 dsize = 0;
853 continue;
854 }
855 dsize = DIRSIZ(OFSFMT(dvp), 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 bcopy((caddr_t)nep, (caddr_t)ep, 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 (BYTE_ORDER == LITTLE_ENDIAN)
879 if (OFSFMT(dvp))
880 namlen = ep->d_type;
881 else
882 namlen = ep->d_namlen;
883 # else
884 namlen = ep->d_namlen;
885 # endif
886 if (ep->d_ino == 0 ||
887 (ep->d_ino == WINO && namlen == dirp->d_namlen &&
888 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
889 if (spacefree + dsize < newentrysize)
890 panic("ufs_direnter: compact1");
891 dirp->d_reclen = spacefree + dsize;
892 } else {
893 if (spacefree < newentrysize)
894 panic("ufs_direnter: compact2");
895 dirp->d_reclen = spacefree;
896 ep->d_reclen = dsize;
897 ep = (struct direct *)((char *)ep + dsize);
898 }
899 #ifdef UFS_DIRHASH
900 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
901 dirp->d_reclen == spacefree))
902 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
903 #endif
904 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
905 #ifdef UFS_DIRHASH
906 if (dp->i_dirhash != NULL)
907 ufsdirhash_checkblock(dp, dirbuf -
908 (dp->i_offset & (DIRBLKSIZ - 1)),
909 dp->i_offset & ~(DIRBLKSIZ - 1));
910 #endif
911
912 if (DOINGSOFTDEP(dvp)) {
913 (void) softdep_setup_directory_add(bp, dp,
914 dp->i_offset + (caddr_t)ep - dirbuf,
915 dirp->d_ino, newdirbp, 0);
916 bdwrite(bp);
917 } else {
918 if (DOINGASYNC(dvp)) {
919 bdwrite(bp);
920 error = 0;
921 } else {
922 error = bwrite(bp);
923 }
924 }
925 dp->i_flag |= IN_CHANGE | IN_UPDATE;
926 /*
927 * If all went well, and the directory can be shortened, proceed
928 * with the truncation. Note that we have to unlock the inode for
929 * the entry that we just entered, as the truncation may need to
930 * lock other inodes which can lead to deadlock if we also hold a
931 * lock on the newly entered node.
932 */
933 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
934 if (tvp != NULL)
935 VOP_UNLOCK(tvp, 0, td);
936 #ifdef UFS_DIRHASH
937 if (dp->i_dirhash != NULL)
938 ufsdirhash_dirtrunc(dp, dp->i_endoff);
939 #endif
940 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
941 IO_NORMAL | IO_SYNC, cr, td);
942 if (tvp != NULL)
943 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
944 }
945 return (error);
946 }
947
948 /*
949 * Remove a directory entry after a call to namei, using
950 * the parameters which it left in nameidata. The entry
951 * dp->i_offset contains the offset into the directory of the
952 * entry to be eliminated. The dp->i_count field contains the
953 * size of the previous record in the directory. If this
954 * is 0, the first entry is being deleted, so we need only
955 * zero the inode number to mark the entry as free. If the
956 * entry is not the first in the directory, we must reclaim
957 * the space of the now empty record by adding the record size
958 * to the size of the previous entry.
959 */
960 int
961 ufs_dirremove(dvp, ip, flags, isrmdir)
962 struct vnode *dvp;
963 struct inode *ip;
964 int flags;
965 int isrmdir;
966 {
967 struct inode *dp;
968 struct direct *ep;
969 struct buf *bp;
970 int error;
971
972 dp = VTOI(dvp);
973
974 if (flags & DOWHITEOUT) {
975 /*
976 * Whiteout entry: set d_ino to WINO.
977 */
978 if ((error =
979 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
980 return (error);
981 ep->d_ino = WINO;
982 ep->d_type = DT_WHT;
983 goto out;
984 }
985
986 if ((error = UFS_BLKATOFF(dvp,
987 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
988 return (error);
989 #ifdef UFS_DIRHASH
990 /*
991 * Remove the dirhash entry. This is complicated by the fact
992 * that `ep' is the previous entry when dp->i_count != 0.
993 */
994 if (dp->i_dirhash != NULL)
995 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
996 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
997 #endif
998 if (dp->i_count == 0) {
999 /*
1000 * First entry in block: set d_ino to zero.
1001 */
1002 ep->d_ino = 0;
1003 } else {
1004 /*
1005 * Collapse new free space into previous entry.
1006 */
1007 ep->d_reclen += dp->i_reclen;
1008 }
1009 #ifdef UFS_DIRHASH
1010 if (dp->i_dirhash != NULL)
1011 ufsdirhash_checkblock(dp, (char *)ep -
1012 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1013 dp->i_offset & ~(DIRBLKSIZ - 1));
1014 #endif
1015 out:
1016 if (DOINGSOFTDEP(dvp)) {
1017 if (ip) {
1018 ip->i_effnlink--;
1019 softdep_change_linkcnt(ip);
1020 softdep_setup_remove(bp, dp, ip, isrmdir);
1021 }
1022 if (softdep_slowdown(dvp)) {
1023 error = bwrite(bp);
1024 } else {
1025 bdwrite(bp);
1026 error = 0;
1027 }
1028 } else {
1029 if (ip) {
1030 ip->i_effnlink--;
1031 ip->i_nlink--;
1032 DIP_SET(ip, i_nlink, ip->i_nlink);
1033 ip->i_flag |= IN_CHANGE;
1034 }
1035 if (flags & DOWHITEOUT)
1036 error = bwrite(bp);
1037 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1038 bdwrite(bp);
1039 error = 0;
1040 } else
1041 error = bwrite(bp);
1042 }
1043 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1044 /*
1045 * If the last named reference to a snapshot goes away,
1046 * drop its snapshot reference so that it will be reclaimed
1047 * when last open reference goes away.
1048 */
1049 #if defined(FFS) || defined(IFS)
1050 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0)
1051 ffs_snapgone(ip);
1052 #endif
1053 return (error);
1054 }
1055
1056 /*
1057 * Rewrite an existing directory entry to point at the inode
1058 * supplied. The parameters describing the directory entry are
1059 * set up by a call to namei.
1060 */
1061 int
1062 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1063 struct inode *dp, *oip;
1064 ino_t newinum;
1065 int newtype;
1066 int isrmdir;
1067 {
1068 struct buf *bp;
1069 struct direct *ep;
1070 struct vnode *vdp = ITOV(dp);
1071 int error;
1072
1073 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1074 if (error)
1075 return (error);
1076 ep->d_ino = newinum;
1077 if (!OFSFMT(vdp))
1078 ep->d_type = newtype;
1079 oip->i_effnlink--;
1080 if (DOINGSOFTDEP(vdp)) {
1081 softdep_change_linkcnt(oip);
1082 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1083 bdwrite(bp);
1084 } else {
1085 oip->i_nlink--;
1086 DIP_SET(oip, i_nlink, oip->i_nlink);
1087 oip->i_flag |= IN_CHANGE;
1088 if (DOINGASYNC(vdp)) {
1089 bdwrite(bp);
1090 error = 0;
1091 } else {
1092 error = bwrite(bp);
1093 }
1094 }
1095 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1096 /*
1097 * If the last named reference to a snapshot goes away,
1098 * drop its snapshot reference so that it will be reclaimed
1099 * when last open reference goes away.
1100 */
1101 #if defined(FFS) || defined(IFS)
1102 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1103 ffs_snapgone(oip);
1104 #endif
1105 return (error);
1106 }
1107
1108 /*
1109 * Check if a directory is empty or not.
1110 * Inode supplied must be locked.
1111 *
1112 * Using a struct dirtemplate here is not precisely
1113 * what we want, but better than using a struct direct.
1114 *
1115 * NB: does not handle corrupted directories.
1116 */
1117 int
1118 ufs_dirempty(ip, parentino, cred)
1119 struct inode *ip;
1120 ino_t parentino;
1121 struct ucred *cred;
1122 {
1123 doff_t off;
1124 struct dirtemplate dbuf;
1125 struct direct *dp = (struct direct *)&dbuf;
1126 int error, count, namlen;
1127 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1128
1129 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1130 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1131 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1132 NOCRED, &count, (struct thread *)0);
1133 /*
1134 * Since we read MINDIRSIZ, residual must
1135 * be 0 unless we're at end of file.
1136 */
1137 if (error || count != 0)
1138 return (0);
1139 /* avoid infinite loops */
1140 if (dp->d_reclen == 0)
1141 return (0);
1142 /* skip empty entries */
1143 if (dp->d_ino == 0 || dp->d_ino == WINO)
1144 continue;
1145 /* accept only "." and ".." */
1146 # if (BYTE_ORDER == LITTLE_ENDIAN)
1147 if (OFSFMT(ITOV(ip)))
1148 namlen = dp->d_type;
1149 else
1150 namlen = dp->d_namlen;
1151 # else
1152 namlen = dp->d_namlen;
1153 # endif
1154 if (namlen > 2)
1155 return (0);
1156 if (dp->d_name[0] != '.')
1157 return (0);
1158 /*
1159 * At this point namlen must be 1 or 2.
1160 * 1 implies ".", 2 implies ".." if second
1161 * char is also "."
1162 */
1163 if (namlen == 1 && dp->d_ino == ip->i_number)
1164 continue;
1165 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1166 continue;
1167 return (0);
1168 }
1169 return (1);
1170 }
1171
1172 /*
1173 * Check if source directory is in the path of the target directory.
1174 * Target is supplied locked, source is unlocked.
1175 * The target is always vput before returning.
1176 */
1177 int
1178 ufs_checkpath(source, target, cred)
1179 struct inode *source, *target;
1180 struct ucred *cred;
1181 {
1182 struct vnode *vp;
1183 int error, namlen;
1184 ino_t rootino;
1185 struct dirtemplate dirbuf;
1186
1187 vp = ITOV(target);
1188 if (target->i_number == source->i_number) {
1189 error = EEXIST;
1190 goto out;
1191 }
1192 rootino = ROOTINO;
1193 error = 0;
1194 if (target->i_number == rootino)
1195 goto out;
1196
1197 for (;;) {
1198 if (vp->v_type != VDIR) {
1199 error = ENOTDIR;
1200 break;
1201 }
1202 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1203 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1204 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, (int *)0,
1205 (struct thread *)0);
1206 if (error != 0)
1207 break;
1208 # if (BYTE_ORDER == LITTLE_ENDIAN)
1209 if (OFSFMT(vp))
1210 namlen = dirbuf.dotdot_type;
1211 else
1212 namlen = dirbuf.dotdot_namlen;
1213 # else
1214 namlen = dirbuf.dotdot_namlen;
1215 # endif
1216 if (namlen != 2 ||
1217 dirbuf.dotdot_name[0] != '.' ||
1218 dirbuf.dotdot_name[1] != '.') {
1219 error = ENOTDIR;
1220 break;
1221 }
1222 if (dirbuf.dotdot_ino == source->i_number) {
1223 error = EINVAL;
1224 break;
1225 }
1226 if (dirbuf.dotdot_ino == rootino)
1227 break;
1228 vput(vp);
1229 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1230 LK_EXCLUSIVE, &vp);
1231 if (error) {
1232 vp = NULL;
1233 break;
1234 }
1235 }
1236
1237 out:
1238 if (error == ENOTDIR)
1239 printf("checkpath: .. not a directory\n");
1240 if (vp != NULL)
1241 vput(vp);
1242 return (error);
1243 }
Cache object: 395f6270cb660b0027fa8a201c4f39b9
|