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("$FreeBSD$");
39
40 #include "opt_ffs_broken_fixme.h"
41 #include "opt_ufs.h"
42 #include "opt_quota.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/namei.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/proc.h>
51 #include <sys/stat.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/sysctl.h>
55
56 #include <vm/vm.h>
57 #include <vm/vm_extern.h>
58
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/dir.h>
63 #ifdef UFS_DIRHASH
64 #include <ufs/ufs/dirhash.h>
65 #endif
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/ufs_extern.h>
68
69 #ifdef DIAGNOSTIC
70 static int dirchk = 1;
71 #else
72 static int dirchk = 0;
73 #endif
74
75 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76
77 /* true if old FS format...*/
78 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
79
80 /*
81 * Convert a component of a pathname into a pointer to a locked inode.
82 * This is a very central and rather complicated routine.
83 * If the filesystem is not maintained in a strict tree hierarchy,
84 * this can result in a deadlock situation (see comments in code below).
85 *
86 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
87 * on whether the name is to be looked up, created, renamed, or deleted.
88 * When CREATE, RENAME, or DELETE is specified, information usable in
89 * creating, renaming, or deleting a directory entry may be calculated.
90 * If flag has LOCKPARENT or'ed into it and the target of the pathname
91 * exists, lookup returns both the target and its parent directory locked.
92 * When creating or renaming and LOCKPARENT is specified, the target may
93 * not be ".". When deleting and LOCKPARENT is specified, the target may
94 * be "."., but the caller must check to ensure it does an vrele and vput
95 * instead of two vputs.
96 *
97 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
98 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
99 * method.
100 *
101 * vfs_cache_lookup() performs the following for us:
102 * check that it is a directory
103 * check accessibility of directory
104 * check for modification attempts on read-only mounts
105 * if name found in cache
106 * if at end of path and deleting or creating
107 * drop it
108 * else
109 * return name.
110 * return VOP_CACHEDLOOKUP()
111 *
112 * Overall outline of ufs_lookup:
113 *
114 * search for name in directory, to found or notfound
115 * notfound:
116 * if creating, return locked directory, leaving info on available slots
117 * else return error
118 * found:
119 * if at end of path and deleting, return information to allow delete
120 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
121 * inode and return info to allow rewrite
122 * if not at end, add name to cache; if at end and neither creating
123 * nor deleting, add name to cache
124 */
125 int
126 ufs_lookup(ap)
127 struct vop_cachedlookup_args /* {
128 struct vnode *a_dvp;
129 struct vnode **a_vpp;
130 struct componentname *a_cnp;
131 } */ *ap;
132 {
133 struct vnode *vdp; /* vnode for directory being searched */
134 struct inode *dp; /* inode for directory being searched */
135 struct buf *bp; /* a buffer of directory entries */
136 struct direct *ep; /* the current directory entry */
137 int entryoffsetinblock; /* offset of ep in bp's buffer */
138 enum {NONE, COMPACT, FOUND} slotstatus;
139 doff_t slotoffset; /* offset of area with free space */
140 int slotsize; /* size of area at slotoffset */
141 int slotfreespace; /* amount of space free in slot */
142 int slotneeded; /* size of the entry we're seeking */
143 int numdirpasses; /* strategy for directory search */
144 doff_t endsearch; /* offset to end directory search */
145 doff_t prevoff; /* prev entry dp->i_offset */
146 struct vnode *pdp; /* saved dp during symlink work */
147 struct vnode *tdp; /* returned by VFS_VGET */
148 doff_t enduseful; /* pointer past last used dir slot */
149 u_long bmask; /* block offset mask */
150 int namlen, error;
151 struct vnode **vpp = ap->a_vpp;
152 struct componentname *cnp = ap->a_cnp;
153 struct ucred *cred = cnp->cn_cred;
154 int flags = cnp->cn_flags;
155 int nameiop = cnp->cn_nameiop;
156 struct thread *td = cnp->cn_thread;
157 ino_t saved_ino;
158
159 bp = NULL;
160 slotoffset = -1;
161 /*
162 * XXX there was a soft-update diff about this I couldn't merge.
163 * I think this was the equiv.
164 */
165 *vpp = NULL;
166
167 vdp = ap->a_dvp;
168 dp = VTOI(vdp);
169 /*
170 * We now have a segment name to search for, and a directory to search.
171 *
172 * Suppress search for slots unless creating
173 * file and at end of pathname, in which case
174 * we watch for a place to put the new file in
175 * case it doesn't already exist.
176 */
177 slotstatus = FOUND;
178 slotfreespace = slotsize = slotneeded = 0;
179 if ((nameiop == CREATE || nameiop == RENAME) &&
180 (flags & ISLASTCN)) {
181 slotstatus = NONE;
182 slotneeded = DIRECTSIZ(cnp->cn_namelen);
183 }
184 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
185
186 #ifdef UFS_DIRHASH
187 /*
188 * Use dirhash for fast operations on large directories. The logic
189 * to determine whether to hash the directory is contained within
190 * ufsdirhash_build(); a zero return means that it decided to hash
191 * this directory and it successfully built up the hash table.
192 */
193 if (ufsdirhash_build(dp) == 0) {
194 /* Look for a free slot if needed. */
195 enduseful = dp->i_size;
196 if (slotstatus != FOUND) {
197 slotoffset = ufsdirhash_findfree(dp, slotneeded,
198 &slotsize);
199 if (slotoffset >= 0) {
200 slotstatus = COMPACT;
201 enduseful = ufsdirhash_enduseful(dp);
202 if (enduseful < 0)
203 enduseful = dp->i_size;
204 }
205 }
206 /* Look up the component. */
207 numdirpasses = 1;
208 entryoffsetinblock = 0; /* silence compiler warning */
209 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
210 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
211 case 0:
212 ep = (struct direct *)((char *)bp->b_data +
213 (dp->i_offset & bmask));
214 goto foundentry;
215 case ENOENT:
216 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
217 goto notfound;
218 default:
219 /* Something failed; just do a linear search. */
220 break;
221 }
222 }
223 #endif /* UFS_DIRHASH */
224 /*
225 * If there is cached information on a previous search of
226 * this directory, pick up where we last left off.
227 * We cache only lookups as these are the most common
228 * and have the greatest payoff. Caching CREATE has little
229 * benefit as it usually must search the entire directory
230 * to determine that the entry does not exist. Caching the
231 * location of the last DELETE or RENAME has not reduced
232 * profiling time and hence has been removed in the interest
233 * of simplicity.
234 */
235 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
236 dp->i_diroff >= dp->i_size) {
237 entryoffsetinblock = 0;
238 dp->i_offset = 0;
239 numdirpasses = 1;
240 } else {
241 dp->i_offset = dp->i_diroff;
242 if ((entryoffsetinblock = dp->i_offset & bmask) &&
243 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
244 return (error);
245 numdirpasses = 2;
246 nchstats.ncs_2passes++;
247 }
248 prevoff = dp->i_offset;
249 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
250 enduseful = 0;
251
252 searchloop:
253 while (dp->i_offset < endsearch) {
254 /*
255 * If necessary, get the next directory block.
256 */
257 if ((dp->i_offset & bmask) == 0) {
258 if (bp != NULL)
259 brelse(bp);
260 error =
261 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
262 if (error)
263 return (error);
264 entryoffsetinblock = 0;
265 }
266 /*
267 * If still looking for a slot, and at a DIRBLKSIZE
268 * boundary, have to start looking for free space again.
269 */
270 if (slotstatus == NONE &&
271 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
272 slotoffset = -1;
273 slotfreespace = 0;
274 }
275 /*
276 * Get pointer to next entry.
277 * Full validation checks are slow, so we only check
278 * enough to insure forward progress through the
279 * directory. Complete checks can be run by patching
280 * "dirchk" to be true.
281 */
282 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
283 if (ep->d_reclen == 0 || ep->d_reclen >
284 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
285 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
286 int i;
287
288 ufs_dirbad(dp, dp->i_offset, "mangled entry");
289 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
290 dp->i_offset += i;
291 entryoffsetinblock += i;
292 continue;
293 }
294
295 /*
296 * If an appropriate sized slot has not yet been found,
297 * check to see if one is available. Also accumulate space
298 * in the current block so that we can determine if
299 * compaction is viable.
300 */
301 if (slotstatus != FOUND) {
302 int size = ep->d_reclen;
303
304 if (ep->d_ino != 0)
305 size -= DIRSIZ(OFSFMT(vdp), ep);
306 if (size > 0) {
307 if (size >= slotneeded) {
308 slotstatus = FOUND;
309 slotoffset = dp->i_offset;
310 slotsize = ep->d_reclen;
311 } else if (slotstatus == NONE) {
312 slotfreespace += size;
313 if (slotoffset == -1)
314 slotoffset = dp->i_offset;
315 if (slotfreespace >= slotneeded) {
316 slotstatus = COMPACT;
317 slotsize = dp->i_offset +
318 ep->d_reclen - slotoffset;
319 }
320 }
321 }
322 }
323
324 /*
325 * Check for a name match.
326 */
327 if (ep->d_ino) {
328 # if (BYTE_ORDER == LITTLE_ENDIAN)
329 if (OFSFMT(vdp))
330 namlen = ep->d_type;
331 else
332 namlen = ep->d_namlen;
333 # else
334 namlen = ep->d_namlen;
335 # endif
336 if (namlen == cnp->cn_namelen &&
337 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
338 !bcmp(cnp->cn_nameptr, ep->d_name,
339 (unsigned)namlen)) {
340 #ifdef UFS_DIRHASH
341 foundentry:
342 #endif
343 /*
344 * Save directory entry's inode number and
345 * reclen in ndp->ni_ufs area, and release
346 * directory buffer.
347 */
348 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
349 ep->d_type == DT_WHT) {
350 slotstatus = FOUND;
351 slotoffset = dp->i_offset;
352 slotsize = ep->d_reclen;
353 dp->i_reclen = slotsize;
354 enduseful = dp->i_size;
355 ap->a_cnp->cn_flags |= ISWHITEOUT;
356 numdirpasses--;
357 goto notfound;
358 }
359 dp->i_ino = ep->d_ino;
360 dp->i_reclen = ep->d_reclen;
361 goto found;
362 }
363 }
364 prevoff = dp->i_offset;
365 dp->i_offset += ep->d_reclen;
366 entryoffsetinblock += ep->d_reclen;
367 if (ep->d_ino)
368 enduseful = dp->i_offset;
369 }
370 notfound:
371 /*
372 * If we started in the middle of the directory and failed
373 * to find our target, we must check the beginning as well.
374 */
375 if (numdirpasses == 2) {
376 numdirpasses--;
377 dp->i_offset = 0;
378 endsearch = dp->i_diroff;
379 goto searchloop;
380 }
381 if (bp != NULL)
382 brelse(bp);
383 /*
384 * If creating, and at end of pathname and current
385 * directory has not been removed, then can consider
386 * allowing file to be created.
387 */
388 if ((nameiop == CREATE || nameiop == RENAME ||
389 (nameiop == DELETE &&
390 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
391 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
392 (flags & ISLASTCN) && dp->i_effnlink != 0) {
393 /*
394 * Access for write is interpreted as allowing
395 * creation of files in the directory.
396 */
397 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
398 if (error)
399 return (error);
400 /*
401 * Return an indication of where the new directory
402 * entry should be put. If we didn't find a slot,
403 * then set dp->i_count to 0 indicating
404 * that the new slot belongs at the end of the
405 * directory. If we found a slot, then the new entry
406 * can be put in the range from dp->i_offset to
407 * dp->i_offset + dp->i_count.
408 */
409 if (slotstatus == NONE) {
410 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
411 dp->i_count = 0;
412 enduseful = dp->i_offset;
413 } else if (nameiop == DELETE) {
414 dp->i_offset = slotoffset;
415 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
416 dp->i_count = 0;
417 else
418 dp->i_count = dp->i_offset - prevoff;
419 } else {
420 dp->i_offset = slotoffset;
421 dp->i_count = slotsize;
422 if (enduseful < slotoffset + slotsize)
423 enduseful = slotoffset + slotsize;
424 }
425 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
426 /*
427 * We return with the directory locked, so that
428 * the parameters we set up above will still be
429 * valid if we actually decide to do a direnter().
430 * We return ni_vp == NULL to indicate that the entry
431 * does not currently exist; we leave a pointer to
432 * the (locked) directory inode in ndp->ni_dvp.
433 * The pathname buffer is saved so that the name
434 * can be obtained later.
435 *
436 * NB - if the directory is unlocked, then this
437 * information cannot be used.
438 */
439 cnp->cn_flags |= SAVENAME;
440 return (EJUSTRETURN);
441 }
442 /*
443 * Insert name into cache (as non-existent) if appropriate.
444 */
445 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
446 cache_enter(vdp, *vpp, cnp);
447 return (ENOENT);
448
449 found:
450 if (numdirpasses == 2)
451 nchstats.ncs_pass2++;
452 /*
453 * Check that directory length properly reflects presence
454 * of this entry.
455 */
456 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
457 ufs_dirbad(dp, dp->i_offset, "i_size too small");
458 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
459 DIP_SET(dp, i_size, dp->i_size);
460 dp->i_flag |= IN_CHANGE | IN_UPDATE;
461 }
462 brelse(bp);
463
464 /*
465 * Found component in pathname.
466 * If the final component of path name, save information
467 * in the cache as to where the entry was found.
468 */
469 if ((flags & ISLASTCN) && nameiop == LOOKUP)
470 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
471
472 /*
473 * If deleting, and at end of pathname, return
474 * parameters which can be used to remove file.
475 */
476 if (nameiop == DELETE && (flags & ISLASTCN)) {
477 /*
478 * Write access to directory required to delete files.
479 */
480 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
481 if (error)
482 return (error);
483 /*
484 * Return pointer to current entry in dp->i_offset,
485 * and distance past previous entry (if there
486 * is a previous entry in this block) in dp->i_count.
487 * Save directory inode pointer in ndp->ni_dvp for dirremove().
488 */
489 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
490 dp->i_count = 0;
491 else
492 dp->i_count = dp->i_offset - prevoff;
493 if (dp->i_number == dp->i_ino) {
494 VREF(vdp);
495 *vpp = vdp;
496 return (0);
497 }
498 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino,
499 LK_EXCLUSIVE, &tdp)) != 0)
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 VOP_ACCESS(vdp, VADMIN, cred, cnp->cn_thread) &&
509 VOP_ACCESS(tdp, VADMIN, cred, cnp->cn_thread)) {
510 vput(tdp);
511 return (EPERM);
512 }
513 *vpp = tdp;
514 return (0);
515 }
516
517 /*
518 * If rewriting (RENAME), return the inode and the
519 * information required to rewrite the present directory
520 * Must get inode of directory entry to verify it's a
521 * regular file, or empty directory.
522 */
523 if (nameiop == RENAME && (flags & ISLASTCN)) {
524 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)))
525 return (error);
526 /*
527 * Careful about locking second inode.
528 * This can only occur if the target is ".".
529 */
530 if (dp->i_number == dp->i_ino)
531 return (EISDIR);
532 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino,
533 LK_EXCLUSIVE, &tdp)) != 0)
534 return (error);
535 *vpp = tdp;
536 cnp->cn_flags |= SAVENAME;
537 return (0);
538 }
539
540 /*
541 * Step through the translation in the name. We do not `vput' the
542 * directory because we may need it again if a symbolic link
543 * is relative to the current directory. Instead we save it
544 * unlocked as "pdp". We must get the target inode before unlocking
545 * the directory to insure that the inode will not be removed
546 * before we get it. We prevent deadlock by always fetching
547 * inodes from the root, moving down the directory tree. Thus
548 * when following backward pointers ".." we must unlock the
549 * parent directory before getting the requested directory.
550 * There is a potential race condition here if both the current
551 * and parent directories are removed before the VFS_VGET for the
552 * inode associated with ".." returns. We hope that this occurs
553 * infrequently since we cannot avoid this race condition without
554 * implementing a sophisticated deadlock detection algorithm.
555 * Note also that this simple deadlock detection scheme will not
556 * work if the filesystem has any hard links other than ".."
557 * that point backwards in the directory structure.
558 */
559 pdp = vdp;
560 if (flags & ISDOTDOT) {
561 saved_ino = dp->i_ino;
562 VOP_UNLOCK(pdp, 0, td); /* race to get the inode */
563 error = VFS_VGET(pdp->v_mount, saved_ino,
564 cnp->cn_lkflags, &tdp);
565 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, td);
566 if (error)
567 return (error);
568 *vpp = tdp;
569 } else if (dp->i_number == dp->i_ino) {
570 VREF(vdp); /* we want ourself, ie "." */
571 *vpp = vdp;
572 } else {
573 error = VFS_VGET(pdp->v_mount, dp->i_ino,
574 cnp->cn_lkflags, &tdp);
575 if (error)
576 return (error);
577 *vpp = tdp;
578 }
579
580 /*
581 * Insert name into cache if appropriate.
582 */
583 if (cnp->cn_flags & MAKEENTRY)
584 cache_enter(vdp, *vpp, cnp);
585 return (0);
586 }
587
588 void
589 ufs_dirbad(ip, offset, how)
590 struct inode *ip;
591 doff_t offset;
592 char *how;
593 {
594 struct mount *mp;
595
596 mp = ITOV(ip)->v_mount;
597 if ((mp->mnt_flag & MNT_RDONLY) == 0)
598 panic("ufs_dirbad: %s: bad dir ino %lu at offset %ld: %s",
599 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
600 else
601 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
602 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
603 }
604
605 /*
606 * Do consistency checking on a directory entry:
607 * record length must be multiple of 4
608 * entry must fit in rest of its DIRBLKSIZ block
609 * record must be large enough to contain entry
610 * name is not longer than MAXNAMLEN
611 * name must be as long as advertised, and null terminated
612 */
613 int
614 ufs_dirbadentry(dp, ep, entryoffsetinblock)
615 struct vnode *dp;
616 struct direct *ep;
617 int entryoffsetinblock;
618 {
619 int i, namlen;
620
621 # if (BYTE_ORDER == LITTLE_ENDIAN)
622 if (OFSFMT(dp))
623 namlen = ep->d_type;
624 else
625 namlen = ep->d_namlen;
626 # else
627 namlen = ep->d_namlen;
628 # endif
629 if ((ep->d_reclen & 0x3) != 0 ||
630 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
631 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
632 /*return (1); */
633 printf("First bad\n");
634 goto bad;
635 }
636 if (ep->d_ino == 0)
637 return (0);
638 for (i = 0; i < namlen; i++)
639 if (ep->d_name[i] == '\0') {
640 /*return (1); */
641 printf("Second bad\n");
642 goto bad;
643 }
644 if (ep->d_name[i])
645 goto bad;
646 return (0);
647 bad:
648 return (1);
649 }
650
651 /*
652 * Construct a new directory entry after a call to namei, using the
653 * parameters that it left in the componentname argument cnp. The
654 * argument ip is the inode to which the new directory entry will refer.
655 */
656 void
657 ufs_makedirentry(ip, cnp, newdirp)
658 struct inode *ip;
659 struct componentname *cnp;
660 struct direct *newdirp;
661 {
662
663 #ifdef DIAGNOSTIC
664 if ((cnp->cn_flags & SAVENAME) == 0)
665 panic("ufs_makedirentry: missing name");
666 #endif
667 newdirp->d_ino = ip->i_number;
668 newdirp->d_namlen = cnp->cn_namelen;
669 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
670 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
671 newdirp->d_type = IFTODT(ip->i_mode);
672 else {
673 newdirp->d_type = 0;
674 # if (BYTE_ORDER == LITTLE_ENDIAN)
675 { u_char tmp = newdirp->d_namlen;
676 newdirp->d_namlen = newdirp->d_type;
677 newdirp->d_type = tmp; }
678 # endif
679 }
680 }
681
682 /*
683 * Write a directory entry after a call to namei, using the parameters
684 * that it left in nameidata. The argument dirp is the new directory
685 * entry contents. Dvp is a pointer to the directory to be written,
686 * which was left locked by namei. Remaining parameters (dp->i_offset,
687 * dp->i_count) indicate how the space for the new entry is to be obtained.
688 * Non-null bp indicates that a directory is being created (for the
689 * soft dependency code).
690 */
691 int
692 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
693 struct vnode *dvp;
694 struct vnode *tvp;
695 struct direct *dirp;
696 struct componentname *cnp;
697 struct buf *newdirbp;
698 {
699 struct ucred *cr;
700 struct thread *td;
701 int newentrysize;
702 struct inode *dp;
703 struct buf *bp;
704 u_int dsize;
705 struct direct *ep, *nep;
706 int error, ret, blkoff, loc, spacefree, flags, namlen;
707 char *dirbuf;
708
709 td = curthread; /* XXX */
710 cr = td->td_ucred;
711
712 dp = VTOI(dvp);
713 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
714
715 if (dp->i_count == 0) {
716 /*
717 * If dp->i_count is 0, then namei could find no
718 * space in the directory. Here, dp->i_offset will
719 * be on a directory block boundary and we will write the
720 * new entry into a fresh block.
721 */
722 if (dp->i_offset & (DIRBLKSIZ - 1))
723 panic("ufs_direnter: newblk");
724 flags = BA_CLRBUF;
725 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
726 flags |= IO_SYNC;
727 #ifdef QUOTA
728 if ((error = getinoquota(dp)) != 0) {
729 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
730 bdwrite(newdirbp);
731 return (error);
732 }
733 #endif
734 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
735 cr, flags, &bp)) != 0) {
736 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
737 bdwrite(newdirbp);
738 return (error);
739 }
740 dp->i_size = dp->i_offset + DIRBLKSIZ;
741 DIP_SET(dp, i_size, dp->i_size);
742 dp->i_flag |= IN_CHANGE | IN_UPDATE;
743 vnode_pager_setsize(dvp, (u_long)dp->i_size);
744 dirp->d_reclen = DIRBLKSIZ;
745 blkoff = dp->i_offset &
746 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
747 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
748 #ifdef UFS_DIRHASH
749 if (dp->i_dirhash != NULL) {
750 ufsdirhash_newblk(dp, dp->i_offset);
751 ufsdirhash_add(dp, dirp, dp->i_offset);
752 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
753 dp->i_offset);
754 }
755 #endif
756 if (DOINGSOFTDEP(dvp)) {
757 /*
758 * Ensure that the entire newly allocated block is a
759 * valid directory so that future growth within the
760 * block does not have to ensure that the block is
761 * written before the inode.
762 */
763 blkoff += DIRBLKSIZ;
764 while (blkoff < bp->b_bcount) {
765 ((struct direct *)
766 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
767 blkoff += DIRBLKSIZ;
768 }
769 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
770 dirp->d_ino, newdirbp, 1) == 0) {
771 bdwrite(bp);
772 return (UFS_UPDATE(dvp, 0));
773 }
774 /* We have just allocated a directory block in an
775 * indirect block. Rather than tracking when it gets
776 * claimed by the inode, we simply do a VOP_FSYNC
777 * now to ensure that it is there (in case the user
778 * does a future fsync). Note that we have to unlock
779 * the inode for the entry that we just entered, as
780 * the VOP_FSYNC may need to lock other inodes which
781 * can lead to deadlock if we also hold a lock on
782 * the newly entered node.
783 */
784 if ((error = bwrite(bp)))
785 return (error);
786 if (tvp != NULL)
787 VOP_UNLOCK(tvp, 0, td);
788 error = VOP_FSYNC(dvp, MNT_WAIT, td);
789 if (tvp != NULL)
790 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
791 return (error);
792 }
793 if (DOINGASYNC(dvp)) {
794 bdwrite(bp);
795 return (UFS_UPDATE(dvp, 0));
796 }
797 error = bwrite(bp);
798 ret = UFS_UPDATE(dvp, 1);
799 if (error == 0)
800 return (ret);
801 return (error);
802 }
803
804 /*
805 * If dp->i_count is non-zero, then namei found space for the new
806 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
807 * in the directory. To use this space, we may have to compact
808 * the entries located there, by copying them together towards the
809 * beginning of the block, leaving the free space in one usable
810 * chunk at the end.
811 */
812
813 /*
814 * Increase size of directory if entry eats into new space.
815 * This should never push the size past a new multiple of
816 * DIRBLKSIZE.
817 *
818 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
819 */
820 if (dp->i_offset + dp->i_count > dp->i_size) {
821 dp->i_size = dp->i_offset + dp->i_count;
822 DIP_SET(dp, i_size, dp->i_size);
823 }
824 /*
825 * Get the block containing the space for the new directory entry.
826 */
827 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
828 if (error) {
829 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
830 bdwrite(newdirbp);
831 return (error);
832 }
833 /*
834 * Find space for the new entry. In the simple case, the entry at
835 * offset base will have the space. If it does not, then namei
836 * arranged that compacting the region dp->i_offset to
837 * dp->i_offset + dp->i_count would yield the space.
838 */
839 ep = (struct direct *)dirbuf;
840 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
841 spacefree = ep->d_reclen - dsize;
842 for (loc = ep->d_reclen; loc < dp->i_count; ) {
843 nep = (struct direct *)(dirbuf + loc);
844
845 /* Trim the existing slot (NB: dsize may be zero). */
846 ep->d_reclen = dsize;
847 ep = (struct direct *)((char *)ep + dsize);
848
849 /* Read nep->d_reclen now as the bcopy() may clobber it. */
850 loc += nep->d_reclen;
851 if (nep->d_ino == 0) {
852 /*
853 * A mid-block unused entry. Such entries are
854 * never created by the kernel, but fsck_ffs
855 * can create them (and it doesn't fix them).
856 *
857 * Add up the free space, and initialise the
858 * relocated entry since we don't bcopy it.
859 */
860 spacefree += nep->d_reclen;
861 ep->d_ino = 0;
862 dsize = 0;
863 continue;
864 }
865 dsize = DIRSIZ(OFSFMT(dvp), nep);
866 spacefree += nep->d_reclen - dsize;
867 #ifdef UFS_DIRHASH
868 if (dp->i_dirhash != NULL)
869 ufsdirhash_move(dp, nep,
870 dp->i_offset + ((char *)nep - dirbuf),
871 dp->i_offset + ((char *)ep - dirbuf));
872 #endif
873 if (DOINGSOFTDEP(dvp))
874 softdep_change_directoryentry_offset(dp, dirbuf,
875 (caddr_t)nep, (caddr_t)ep, dsize);
876 else
877 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
878 }
879 /*
880 * Here, `ep' points to a directory entry containing `dsize' in-use
881 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
882 * then the entry is completely unused (dsize == 0). The value
883 * of ep->d_reclen is always indeterminate.
884 *
885 * Update the pointer fields in the previous entry (if any),
886 * copy in the new entry, and write out the block.
887 */
888 # if (BYTE_ORDER == LITTLE_ENDIAN)
889 if (OFSFMT(dvp))
890 namlen = ep->d_type;
891 else
892 namlen = ep->d_namlen;
893 # else
894 namlen = ep->d_namlen;
895 # endif
896 if (ep->d_ino == 0 ||
897 (ep->d_ino == WINO && namlen == dirp->d_namlen &&
898 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
899 if (spacefree + dsize < newentrysize)
900 panic("ufs_direnter: compact1");
901 dirp->d_reclen = spacefree + dsize;
902 } else {
903 if (spacefree < newentrysize)
904 panic("ufs_direnter: compact2");
905 dirp->d_reclen = spacefree;
906 ep->d_reclen = dsize;
907 ep = (struct direct *)((char *)ep + dsize);
908 }
909 #ifdef UFS_DIRHASH
910 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
911 dirp->d_reclen == spacefree))
912 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
913 #endif
914 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
915 #ifdef UFS_DIRHASH
916 if (dp->i_dirhash != NULL)
917 ufsdirhash_checkblock(dp, dirbuf -
918 (dp->i_offset & (DIRBLKSIZ - 1)),
919 dp->i_offset & ~(DIRBLKSIZ - 1));
920 #endif
921
922 if (DOINGSOFTDEP(dvp)) {
923 (void) softdep_setup_directory_add(bp, dp,
924 dp->i_offset + (caddr_t)ep - dirbuf,
925 dirp->d_ino, newdirbp, 0);
926 bdwrite(bp);
927 } else {
928 if (DOINGASYNC(dvp)) {
929 bdwrite(bp);
930 error = 0;
931 } else {
932 error = bwrite(bp);
933 }
934 }
935 dp->i_flag |= IN_CHANGE | IN_UPDATE;
936 /*
937 * If all went well, and the directory can be shortened, proceed
938 * with the truncation. Note that we have to unlock the inode for
939 * the entry that we just entered, as the truncation may need to
940 * lock other inodes which can lead to deadlock if we also hold a
941 * lock on the newly entered node.
942 */
943 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
944 if (tvp != NULL)
945 VOP_UNLOCK(tvp, 0, td);
946 #ifdef UFS_DIRHASH
947 if (dp->i_dirhash != NULL)
948 ufsdirhash_dirtrunc(dp, dp->i_endoff);
949 #endif
950 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
951 IO_NORMAL | IO_SYNC, cr, td);
952 if (tvp != NULL)
953 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
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
982 dp = VTOI(dvp);
983
984 if (flags & DOWHITEOUT) {
985 /*
986 * Whiteout entry: set d_ino to WINO.
987 */
988 if ((error =
989 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
990 return (error);
991 ep->d_ino = WINO;
992 ep->d_type = DT_WHT;
993 goto out;
994 }
995
996 if ((error = UFS_BLKATOFF(dvp,
997 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
998 return (error);
999 #ifdef UFS_DIRHASH
1000 /*
1001 * Remove the dirhash entry. This is complicated by the fact
1002 * that `ep' is the previous entry when dp->i_count != 0.
1003 */
1004 if (dp->i_dirhash != NULL)
1005 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
1006 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
1007 #endif
1008 if (dp->i_count == 0) {
1009 /*
1010 * First entry in block: set d_ino to zero.
1011 */
1012 ep->d_ino = 0;
1013 } else {
1014 /*
1015 * Collapse new free space into previous entry.
1016 */
1017 ep->d_reclen += dp->i_reclen;
1018 }
1019 #ifdef UFS_DIRHASH
1020 if (dp->i_dirhash != NULL)
1021 ufsdirhash_checkblock(dp, (char *)ep -
1022 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1023 dp->i_offset & ~(DIRBLKSIZ - 1));
1024 #endif
1025 out:
1026 if (DOINGSOFTDEP(dvp)) {
1027 if (ip) {
1028 ip->i_effnlink--;
1029 softdep_change_linkcnt(ip);
1030 softdep_setup_remove(bp, dp, ip, isrmdir);
1031 }
1032 if (softdep_slowdown(dvp)) {
1033 error = bwrite(bp);
1034 } else {
1035 bdwrite(bp);
1036 error = 0;
1037 }
1038 } else {
1039 if (ip) {
1040 ip->i_effnlink--;
1041 ip->i_nlink--;
1042 DIP_SET(ip, i_nlink, ip->i_nlink);
1043 ip->i_flag |= IN_CHANGE;
1044 }
1045 if (flags & DOWHITEOUT)
1046 error = bwrite(bp);
1047 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1048 bdwrite(bp);
1049 error = 0;
1050 } else
1051 error = bwrite(bp);
1052 }
1053 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1054 /*
1055 * If the last named reference to a snapshot goes away,
1056 * drop its snapshot reference so that it will be reclaimed
1057 * when last open reference goes away.
1058 */
1059 #if defined(FFS) || defined(IFS)
1060 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0)
1061 ffs_snapgone(ip);
1062 #endif
1063 return (error);
1064 }
1065
1066 /*
1067 * Rewrite an existing directory entry to point at the inode
1068 * supplied. The parameters describing the directory entry are
1069 * set up by a call to namei.
1070 */
1071 int
1072 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1073 struct inode *dp, *oip;
1074 ino_t newinum;
1075 int newtype;
1076 int isrmdir;
1077 {
1078 struct buf *bp;
1079 struct direct *ep;
1080 struct vnode *vdp = ITOV(dp);
1081 int error;
1082
1083 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1084 if (error)
1085 return (error);
1086 ep->d_ino = newinum;
1087 if (!OFSFMT(vdp))
1088 ep->d_type = newtype;
1089 oip->i_effnlink--;
1090 if (DOINGSOFTDEP(vdp)) {
1091 softdep_change_linkcnt(oip);
1092 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1093 bdwrite(bp);
1094 } else {
1095 oip->i_nlink--;
1096 DIP_SET(oip, i_nlink, oip->i_nlink);
1097 oip->i_flag |= IN_CHANGE;
1098 if (DOINGASYNC(vdp)) {
1099 bdwrite(bp);
1100 error = 0;
1101 } else {
1102 error = bwrite(bp);
1103 }
1104 }
1105 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1106 /*
1107 * If the last named reference to a snapshot goes away,
1108 * drop its snapshot reference so that it will be reclaimed
1109 * when last open reference goes away.
1110 */
1111 #if defined(FFS) || defined(IFS)
1112 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1113 ffs_snapgone(oip);
1114 #endif
1115 return (error);
1116 }
1117
1118 /*
1119 * Check if a directory is empty or not.
1120 * Inode supplied must be locked.
1121 *
1122 * Using a struct dirtemplate here is not precisely
1123 * what we want, but better than using a struct direct.
1124 *
1125 * NB: does not handle corrupted directories.
1126 */
1127 int
1128 ufs_dirempty(ip, parentino, cred)
1129 struct inode *ip;
1130 ino_t parentino;
1131 struct ucred *cred;
1132 {
1133 doff_t off;
1134 struct dirtemplate dbuf;
1135 struct direct *dp = (struct direct *)&dbuf;
1136 int error, count, namlen;
1137 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1138
1139 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1140 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1141 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1142 NOCRED, &count, (struct thread *)0);
1143 /*
1144 * Since we read MINDIRSIZ, residual must
1145 * be 0 unless we're at end of file.
1146 */
1147 if (error || count != 0)
1148 return (0);
1149 /* avoid infinite loops */
1150 if (dp->d_reclen == 0)
1151 return (0);
1152 /* skip empty entries */
1153 if (dp->d_ino == 0 || dp->d_ino == WINO)
1154 continue;
1155 /* accept only "." and ".." */
1156 # if (BYTE_ORDER == LITTLE_ENDIAN)
1157 if (OFSFMT(ITOV(ip)))
1158 namlen = dp->d_type;
1159 else
1160 namlen = dp->d_namlen;
1161 # else
1162 namlen = dp->d_namlen;
1163 # endif
1164 if (namlen > 2)
1165 return (0);
1166 if (dp->d_name[0] != '.')
1167 return (0);
1168 /*
1169 * At this point namlen must be 1 or 2.
1170 * 1 implies ".", 2 implies ".." if second
1171 * char is also "."
1172 */
1173 if (namlen == 1 && dp->d_ino == ip->i_number)
1174 continue;
1175 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1176 continue;
1177 return (0);
1178 }
1179 return (1);
1180 }
1181
1182 /*
1183 * Check if source directory is in the path of the target directory.
1184 * Target is supplied locked, source is unlocked.
1185 * The target is always vput before returning.
1186 */
1187 int
1188 ufs_checkpath(source, target, cred)
1189 struct inode *source, *target;
1190 struct ucred *cred;
1191 {
1192 struct vnode *vp;
1193 int error, namlen;
1194 ino_t rootino;
1195 struct dirtemplate dirbuf;
1196
1197 vp = ITOV(target);
1198 if (target->i_number == source->i_number) {
1199 error = EEXIST;
1200 goto out;
1201 }
1202 rootino = ROOTINO;
1203 error = 0;
1204 if (target->i_number == rootino)
1205 goto out;
1206
1207 for (;;) {
1208 if (vp->v_type != VDIR) {
1209 error = ENOTDIR;
1210 break;
1211 }
1212 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1213 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1214 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, (int *)0,
1215 (struct thread *)0);
1216 if (error != 0)
1217 break;
1218 # if (BYTE_ORDER == LITTLE_ENDIAN)
1219 if (OFSFMT(vp))
1220 namlen = dirbuf.dotdot_type;
1221 else
1222 namlen = dirbuf.dotdot_namlen;
1223 # else
1224 namlen = dirbuf.dotdot_namlen;
1225 # endif
1226 if (namlen != 2 ||
1227 dirbuf.dotdot_name[0] != '.' ||
1228 dirbuf.dotdot_name[1] != '.') {
1229 error = ENOTDIR;
1230 break;
1231 }
1232 if (dirbuf.dotdot_ino == source->i_number) {
1233 error = EINVAL;
1234 break;
1235 }
1236 if (dirbuf.dotdot_ino == rootino)
1237 break;
1238 vput(vp);
1239 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1240 LK_EXCLUSIVE, &vp);
1241 if (error) {
1242 vp = NULL;
1243 break;
1244 }
1245 }
1246
1247 out:
1248 if (error == ENOTDIR)
1249 printf("checkpath: .. not a directory\n");
1250 if (vp != NULL)
1251 vput(vp);
1252 return (error);
1253 }
Cache object: 92e697cd209072364cbe25932b0b77f5
|